Computers and Electronics in Agriculture 198 (2022) 107119

Contents lists available at ScienceDirect

Computers and Electronics in Agriculture

journal homepage: www.elsevier.com/locate/compag

Review

Towards leveraging the role of machine learning and artificial intelligence

in precision agriculture and smart farming
Tawseef Ayoub Shaikh a, *, Tabasum Rasool b, Faisal Rasheed Lone c
a
Department of Computer Science & Engineering, Pandit Deendayal Energy University, Gandhinagar, Gujarat, India
b
Interdisciplinary Centre for Water Research (ICWaR), Indian Institute of Science, Bangalore, India
c
Department of Computer Science & Engineering, National Institute of Technology, Jalandhar, Punjab, India

A R T I C L E I N F O A B S T R A C T

Keywords: The digitalization of data has resulted in a data tsunami in practically every industry of data-driven enterprise.
Smart Farming (SF) Furthermore, man-to-machine (M2M) digital data handling has dramatically amplified the information wave.
Precision Agriculture (PA) There has been a significant development in digital agriculture management applications, which has impacted
Automated Irrigation Control
information and communication technology (ICT) to deliver benefits for both farmers and consumers, as well as
Unarmed Aerial Vehicles (UAV)
pushed technological solutions into rural settings. This paper highlights the potential of ICT technologies in
Machine Learning
traditional agriculture, as well as the challenges that may arise when they are used in farming techniques. Ro­
botics, Internet of things (IoT) devices, and machine learning issues, as well as the functions of machine learning,
artificial intelligence, and sensors in agriculture, are all detailed. In addition, drones are being considered for
crop observation as well as crop yield optimization management. When applicable, worldwide and cutting-edge
IoT-based farming systems and platforms are also highlighted. We do a thorough review of the most recent
literature in each area of expertise. We conclude the present and future trends in artificial intelligence (AI) and
highlight existing and emerging research problems in AI in agriculture due to this comprehensive assessment.

1. Introduction to smart agriculture time period. Farmers must adopt more sustainable methods to boost
productivity and supply since food production demands more resources
Agriculture, which accounts for 6.4 percent of global GDP, is the than the present supply will allow. By 2030, meeting 40 percent of water
primary source of food and gross domestic product (GDP). Agriculture is needs may be challenging; at the same time, 20 percent of agriculture
a significant source of revenue in nine countries throughout the world. may be damaged. Farmers must adopt more sustainable strategies to
Millions of people rely on agriculture for fuel and employment (United improve productivity while conserving resources because food produc­
Nations, 2019). According to the UN’s Food and Agriculture Organiza­ tion takes more resources than the present supply will allow (Food and
tion (FAO), global food demand will climb by 70% by 2050 to fulfill Agriculture Organization of the United Nations - FAO, 2019).
rising population demands (Nation United, 2017). Even though enough To meet the needs of the global population, the world’s yearly cereal
food is produced to feed everyone on the earth, 500 million people are production must increase by 3 billion tonnes and meat production by
malnourished, and over 821 million people go hungry. In the future more than 200 percent by 2050 (Trendov et al., 2019). Crop size and
decades, cities will house around two-thirds of the world’s population. farm structure must be expanded, as well as a technologically advanced
Africa, Indonesia, Pakistan, the Democratic Republic of Congo, India, method. Although this may be conceivable, it is uncertain whether it can
Indonesia, Egypt, the United States, and Tanzania will account for more be done in a way that is both sustainable and inclusive. However,
than half of the total global population increase. By that year, 473 reengineering farm operations at a massive size and speed will neces­
million individuals in India and Nigeria are predicted to join the middle sitate a significant and rapid transition. Because of uncertain weather,
class (United Nations, 2019). A rapid rise in the population poses a rising food demand, and rapid population expansion, agricultural sec­
barrier to achieving the Sustainable Development Goals (SDGs) aim of tors are interested in using new technology to get more out of each
eradicating hunger; meeting 40% of water needs by 2030 may be harvest (Food, 2029). Artificial intelligence (AI) is finding more and
challenging, and 20% of agriculture may be degraded within the same more applications in agriculture as it advances. Simultaneously, the

* Corresponding author.
E-mail addresses: [email protected], [email protected] (T. Ayoub Shaikh).

https://doi.org/10.1016/j.compag.2022.107119
Received 30 June 2021; Received in revised form 30 May 2022; Accepted 31 May 2022
Available online 7 June 2022
0168-1699/© 2022 Elsevier B.V. All rights reserved.
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

Internet of Things (IoT) and the Fourth Industrial Revolution (Industry 1.1. Defining smart farming and precision agriculture
4.0) allow for the development of new technologies and ideas. Smart
agriculture, commonly known as “Agriculture 4.0,” is a set of current “Smart Farming (SF) is a technology that relies on its implementation
technologies and advances that can enhance crop production while on the use of AI and IoT in cyber-physical farm management” (Bacco
lowering water and energy use (Pathan et al., 2020). These will be et al., 2019). Smart agriculture addresses many issues related to crop
possible thanks to the integration of environmental sensors and pre­ production as it allows monitoring the changes in climate factors, soil
diction technologies. Agricultural operations may be adjusted to allow characteristics, soil moisture, etc. “Precision agriculture (PA) is the
for increased productivity while using fewer natural resources, thanks to science of improving crop yields and assisting management decisions
the new capabilities introduced by smart farming. Smart farming makes using high technology sensor and analysis tools. PA is a new concept
use of a variety of technologies and platforms to help farmers capitalise adopted worldwide to increase production, reduce labor time, and
on innovative ideas (Pathan et al., 2020). ensure the effective management of fertilizers and irrigation processes”
Agriculture will be transformed by the “digital agricultural revolu­ (Mulla, 2013). It uses a large amount of data and information to improve
tion,” which will improve efficiency, sustainability, inclusiveness, and the use of agricultural resources, yields, and the quality of crops. Pre­
transparency. Nonetheless, to benefit from their superior capabilities, cision agriculture employs data from multiple sources to improve crop
technologies must be integrated into the agricultural industry on a yields and increase the cost-effectiveness of crop management strate­
bigger scale. However, various security challenges in agriculture remain gies, including fertilizer inputs, irrigation management, and pesticide
to be solved due to these elements, such as compatibility, heterogeneity, application. Fig. 1 represents a theoretical prototype of a smart
the handling of enormous amounts of data, and the processing of farming scenario.
massive amounts of data. Agriculture 4.0 must correctly generate, PA was first used to target fertiliser application to specific soil con­
transfer, and process data while safeguarding against cyber-attacks ditions. PA has been utilised to build automated guiding systems for
(Varga et al.). Data-related technologies, such as analytics and smart agricultural vehicles and tools, autonomous machines and processes,
systems, cannot work well without diligent data integrity management. farm research, and agricultural production system management. The
When many resources are combined, security issues occur, such as loss adaptability, speed, precision, and cost-effectiveness of automated ma­
of privacy, preserving trust, and ensuring that the resource is available chine learning in agriculture are important principles. Artificial intelli­
in the first place. Because the internet of things, cellular, and wireless gence in agriculture not only aids farmers in their farming abilities but
technologies are all intrinsically linked, it’s possible that they’ll answer also shifts to directed farming to obtain higher yields and better quality
many of the new and present challenges to agriculture in our modern with fewer resources. Smart agriculture uses current technology like
civilization. It must deal with issues such as new-to-date integrity, date sensors and actuator networks, unmanned aircraft systems (UAS), sat­
correctness, and device security, as well as security elements such as ellite imaging, IoT, and drones to reduce farmers’ workloads while
encryption, data accuracy, and availability (Hassija et al., 2019). increasing farm productivity (Mulla, 2013). Smart agriculture is a crop
management concept that allows farmers to control geographical and
temporal diversity in agriculture, as well as intrusion threats and real-

Fig. 1. A Prototype of a Smart Farming.

2
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

time data monitoring. Chemical, pH, wind, rain, temperature, moisture, thus providing better results (Mekala and Viswanathan, 2017; Reddy
and auditory sensors are extensively used in modern agricultural oper­ et al., 2020). Fig. 3 presents an overall global framework of the IoT
ations. Each sensor has its own set of capabilities and hardware needs, applications in agriculture domains.
which are customised for each location’s conditions. Smart farm man­ Today, we witness innovative agricultural techniques and commer­
agement is aided by IoT peripherals, image processing systems, big data cial models utilising technological breakthroughs faster than ever before
capabilities, data analytics, and wireless technology. Fig. 2 gives an (Reddy et al., 2020). Farmers gain from rental programmes for farming
impression of the central components in precision agriculture. equipment, such as Trringo in India, because they save money and have
Smart and precision agriculture has an essential role in all aspects of access to support systems that automate procedures. A program like this
agriculture. The coming together of the internet of things (IoT) and in­ might be seen as increasing SF adoption in low-income areas. A French
formation technology (IT) to collect data from diverse sources to figure company is striving to make legacy infrastructure more flexible and
out, predetermine, and organize farm activity concerning their envi­ valuable by combining web-based and physical technology. Karnott
ronments. Sensors have become ubiquitous to collect disparate sets of Systems manufactures a control device that can be mounted on agri­
data (soil temperature, moisture, foliage, sunlight levels, and direction). cultural machinery and that combines collecting and exchanging real-
Smart agriculture (SA) is grounded on digital automation, data collec­ time data, geofencing capabilities, and geographic information to
tion, data transmission, decision-making, data processing, and data satisfy farmers’ demands. For receiving information, data can be shared
analysis (Matei et al., 2017). A sensor network is among the most via an internet service like the API-platform agro’s. Various farms have
common data collection and communication technologies in this sector. access to the data, which serves as an excellent research library (Reddy
In smart agriculture, environmental data plays a critical role (Bacco et al., 2020). Tarris provides aerial and satellite imagery to help with
et al., 2019). As the soil is a big factor in climatic disease forecasting, precision farming (PF) using artificial intelligence (AI) as a platform.
meteorological information can alert farmers of infection outbreaks. AgriOpenData offers a Decision Support System (DSS) that utilises
Increased yield and lower environmental impacts are possible with blockchain technology and enables other services. Robotic agriculture
increased use of the data collected for crop protection. Human has provides a complete solution for the entire plant life cycle, from seeding
limited capacity to process this data, so it must be made easier with the to harvesting, while consuming less water.
help of tools and techniques that facilitate analysis to benefit decision- Fig. 4 shows a graphic representation of the smart agriculture system
making (Agrawal et al., 2017). The use of data mining techniques is and the flowchart of the data analytics. It is made up of a perceptual
vital in analyzing data. Big data includes patterns, so using different device, network operations, and edge capabilities, all wrapped in cloud-
ways of looking at the data is critical to discovering these patterns. Data based services (Forum et al., 2009). Sensors, Global Positioning System
mining techniques have been applied to a number of other tasks in the (GPS), radio frequency identification (RFID) tags, and cameras may be
agricultural sector, including the identification of pests, disease detec­ put on the farm for these applications; data is captured and processed at
tion, the prediction of yield, and the planning of fertilizer and pesticide the perception layer. As a result, they are unable to process or store
use, to date, for example (Bacco et al., 2018). They are also used for crop information close to the client, perform in the cloud, or use these gad­
management, but other than that, they can explore other models that gets. This component can be utilised with the wireless sensor network
can assist in measuring the organization of farms. Therefore, collecting since it is linked to internal sensor networks using networking tech­
this data is a new input that can directly affect agricultural efficiency, nologies (WSN). On the creative edge, security features, data filters,

Fig. 2. A General Overview of the Chief Components in Smart and Precision Agriculture.

3
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

Fig. 3. A Global Framework of the IoT in Automated and Smart Agricultural Practices.

Fig. 4. Smart Agriculture System and Data Analytics: A Generic View.

intelligence, diversified processing, numerous in/outbound interfaces, helps store and delivers data to consumers. Although the big data world
and the gateway might all be found. Depending on the features of the will soon meet a vast volume of data created by perceptive devices,
appliance, the edge adds one or more resources. While some devices are cloud processing is expensive (Foley, 2019).
merely capable of data returns, others can perform far more complex
activities. The capacity to process data, make choices, send commands to
machines, and communicate data to the cloud is one of the advantages of 1.2. The drive for smart and automated farming
programmable gateways. The gateway connects to the cloud when you
connect to the Internet through an Internet Service Provider (ISP); it The global population is predicted to reach nine billion people by
2050. According to the United Nations’ International Food and

4
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

Agriculture Organization (FAO), the world’s food supply in 2050 will be protect natural resources.
70 percent bigger than it is now (Malchi et al., 2021). According to the As a result, agriculture is constantly in need of new strategies to
2018 National Climate Assessment of the United States Global Change enhance crop yields, as weather changes, population growth, and food
Research Program, climate change would seriously endanger the pro­ security make expansion challenging. In terms of artificial intelligence
ductivity of food, livestock, and the economic viability of rural com­ advancement, agriculture is increasingly being recognized as a techno­
munities in the United States. On most U.S. crops, yields will decline due logical industry leader. As a result, because this process is complex or
to hotter and dryer weather, soil erosion, disease, pest pressure, and impossible for people, we must automate it with methods and tools to
outbreak, especially in the region of these issues while they remain high. make decision-making easier. To address the aforementioned challenges
As the country’s agricultural land usage has led to the destruction of in today’s farming settings, advances in computer vision, machine
numerous ecosystems, including the prairies of North America and the learning, and deep learning technologies can be used to precisely,
Atlantic Forest of Brazil (Selim, 2018). As illustrated in Fig. 5, agricul­ quickly, and sooner diagnose crop diseases from a variety of existing
ture has accounted for 70% of world water demand during the last few crop diseases. One of the advantages of applying this technology is that
decades (Natarajan et al., 2020). Global water withdrawals have nearly it provides quick and accurate findings from computerised apprehen­
quadrupled from 500 to 2500 cubic kilometers as agricultural land has sions using image processing methods. Computer vision and precision
expanded. Increased food production has become a severe challenge due improvements can be used to reduce labor expenses and time waste
to population growth and climate change. It is clear that the agriculture while also improving crop quality and total output. Through proper
business must change its existing production and adoption techniques to delivery channels, early information on the state of the crop and the area
favor land and water conservation while minimising the risk of uncer­ of illness can aid in the spread of disease management.
tain markets. The agricultural industry can now enter a data-driven era
thanks to the application of numerous new information technologies.
1.3. Research contribution
India is an agricultural country, with agriculture accounting for 16
percent of India’s GDP and 10 percent of overall exports, respectively.
Our research’s significant contributions can be stated as follows:
Approximately 75% of India’s population is dependent on agriculture,
1: The essential techniques in agricultural pre-harvesting automation
either directly or indirectly. Around 3,50,000 ha of land in India are
chores such as soil and seed, crop disease detection, irrigation, weed
utilized for tomato farming, yielding 5.3 million tonnes of tomatoes and
control, pesticides, yield management, and so on are covered in this
ranking India as the world’s third-largest tomato producer. India is the
work.
world’s second-largest producer of potatoes, with 48.60 million tonnes
2: We have discussed how machine learning (ML) in general and
produced annually (AO, , 2016; Glaroudis et al., 2020). India is the
deep learning (DL), in particular, have helped identify fruit and leaf
world’s sixth-largest apple grower, with a production of roughly 2371.0
diseases in plants. The role of IoT and smart agriculture, drones, and
million tonnes. With a yield of approximately 116.42 million metric
improved crop yield tasks, which consist of both image-processing ap­
tonnes, India ranks second in rice production. Diseases squander more
proaches and other techniques involved in this process, are also
than 15% of India’s crops, which ultimately weakens its economy.
discussed.
Agriculture today accounts for more than 10% of greenhouse gas
3. Various diseases and infections seen in various crops such as ap­
emissions and 44% of water use in Europe, owing to the steady expan­
ples, grapes, bananas, and other fruits and vegetables, as well as their
sion of food production requirements (Glaroudis et al., 2020). Chemical
symptoms for classification purposes, are discussed. The steps involved
treatments (pesticides) are widely employed to increase fruit crop
in the automatic detection and classification of illnesses in plants are
market penetration, resulting in significant impacts on pollinators and
discussed, as well as the numerous strategies and algorithms that can be
the planet’s environment. Plants account for more than 80% of the
used in each phase.
human diet. As a result, there is a growing interest in new approaches for
4. Future promises and discussions, as outlined in this comprehen­
reducing water consumption and optimizing pesticide treatments to
sive study, are also offered, as are challenges and issues in the adoption

Fig. 5. Water Withdrawals by Sector (AO, , 2016).

5
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

of these machine learning models for smart agricultural chores. 2. Artificial intelligence and Machine learning techniques in
To elaborate on the innovation covered in this survey and focus on smart agriculture (SA) and precision farming (PF)
how our study is distinct and different from other existing surveys, a
comparison of this survey with various current available similar surveys Eventually, farmers will need to use fewer resources to produce the
is undertaken and given in Table 1. Most available surveys on autono­ same amount of food, reducing the amount of water used and decreasing
mous plant disease diagnosis using machine learning approaches only chemicals. With higher production and lower costs, farmers want to
offer a cursory review of the various elements. However, in our research, maximize the quantity and profitability while the public desires nutri­
we concentrated on many aspects of autonomous plant disease detection tious food. New products, practices, and new technologies are sought
utilizing machine learning approaches. after in the agriculture industry. To meet these different needs, farmers
can use precision agriculture. To remain innovative, agriculture must
1.4. Outline of the paper employ a combination of technologies that permits data collection and
analysis. Many of these new technologies, especially web technologies,
The rest of the paper is divided into five sections. Section 2 contains generate vast amounts of data for everyone to obtain and share.
information about the most commonly used machine learning ap­ Regarding smart agriculture, the critical attributes of data mining are
proaches in the agricultural field. The automation of the pre-harvesting these: (i) It is imperative that we must deal with large amounts of data so
tasks like soil and seed, disease diagnostics tools, crop phenotyping, we can explore that fast and effectively. (ii) For use in smart agriculture,
irrigation control, pesticides management, crop yield, etc., occupy the the data sources need to be diverse; for example, sensors will need to be
core part of Section 3. Various types of probable diseases present in in various formats, as will images, strings, numbers, and so on. We must
different plants such as tomato, rice, apple, and others are described, as understand and communicate with various device types, along with
well as the symptoms that may be utilized to diagnose and classify them. websites, and pull information from the web.
The generic steps involved in automatic plant disease detection and The term ’smart agriculture’ is used chiefly for soil and water use
classification using machine learning and deep learning, such as image planning, monitoring crop health, and reducing pollutants, such as
acquisition, data pre-processing, image segmentation, image extraction, pesticides and herbicides (Ahmed et al., 2016). Everything is becoming
and disease classification, are discussed in detail, as well as the various more brilliant in this fast-changing world of technology and agriculture.
algorithms and techniques commonly used in each step. The findings Performance relies both on historical data and on incorporating com­
and the research potentials in smart agriculture and smart farming are puter vision technology to meet the growing agricultural demand. While
discussed, debated, and presented in Section 4 as Discussions. Section 5 it does affect crop classification, agro chem production, disease detec­
presents various challenges and open issues in the deployment of these tion, and prevention, it does not guarantee it.
machine learning and computer vision models for plant disease detec­
tion and classification, as well as future research directions, based on 2.1. Common Machine learning techniques in precision agriculture
observations of various existing frameworks in the literature, which are
discussed in the paper. The size of the dataset and the classifiers The process of identifying and categorising input data into various
employed have a significant impact on how these frameworks are dis­ known classes is known as classification. The actual detection and seg­
played. As a result, Section 6 contains an attempt to summarise the mentation of incoming data into multiple classes takes place in this step,
existing research and issues identified in this study, as well as the paper’s and it is the most critical task in computer vision. Plant images are
final remarks. The complete layout of this manuscript is shown graph­ categorised and sorted into distinct categories depending on identified
ically in Fig. 6. diseases in this phase of plant leaf categorization. In this phase, leaves
are classed as healthy or diseased. The leaves that have been categorised
as contaminated are then further classified into the precise disease class
that the leaf may be suffering from. Illness detection is a little easier than

Table 1
Comparison with other Existing Surveys in the Literature.
Topics Covered (Andrei (Moshou (Li et al., (Lamba (Shivali Amit (Kibriya Our
et al., 2020) et al., 2005) 2021) et al., Wagle, 2021) et al., 2021) Paper
2021)

Study of various pre-harvesting, post-harvesting stages like a possible X x x # x x ✓

seed, soil, and crop diseases in different crops like beans, tomato,
wheat, and apple, along with their classification
In-depth study of different phases of agriculture like seed and soil, # # x x x x ✓
disease detection, crop phenotyping, pesticide control, weed
detection, irrigation, drones, crop yield, using automated machine
learning frameworks
In-depth study of various existing machine learning and deep learning- ✓ ✓ ✓ ✓ ✓ ✓ ✓
based classification techniques and frameworks in literature along
with their performance evaluations, the dataset used, and the feature
extraction method used
Study of various possible infections and diseases in different crops like ✓ # ✓ # ✓ # ✓
potato, tomato, rice, and apple, along with their causes & probable
symptoms
Study of the various datasets available in the literature for plant disease X # # ✓ # ✓ ✓
detection
Study of various commercial tools available in the literature for smart ✓ x ✓ # x ✓ ✓
agriculture and precise farming
Challenges and issues in automated plant disease detection and # ✓ x # ✓ x ✓
classification using machine learning and deep learning techniques
Future research directions in automated plant disease detection and ✓ # ✓ # ✓ ✓ ✓
classification using machine learning & deep learning techniques

✓ Covered, x Not covered, # Not covered in detail.

6
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

Fig. 6. Complete Layout of the Paper.

disease classification; hence in the case of disease classification, overfitting, and a slight change in the data can cause a large difference in
choosing appropriate classifiers based on the situation is critical. Ma­ the structure of the decision tree, causing instability. The Decision Tree
chine learning techniques are mainly divided into two categories. algorithm is inadequate for applying regression and predicting contin­
i: Supervised Machine Learning techniques: The final label of the uous values.
data is already known, and the models following under this category are k-NN: It is a sort of lazy learning called instance base learning (IBK).
trained so as to automatically find the class label of each data point. The stage of training model construction is frequently skipped, and the
Various techniques falling under this category are Support Vector Ma­ test instance directly relates to the training instances to construct a
chine (SVM), Decision tree, K-NN, Random Forest (RF), Genetic algo­ classification model. This method creates a locally optimised model
rithm (GA), Fuzzy logic, etc. Few Classification techniques in this specific to the test case. K-NN doesn’t require any training period as it is
category are discussed below. a lazy learning model, which makes it fit for adding seamlessly new data.
Support Vector Machine: To make the categorization, SVM uses It is straightforward to implement for multi-class problems and can be
linear criteria and is a suitable choice for both classification and used for classification and regression problems. Even though K-NN has
regression tasks. A single-level decision tree with a sensibly desired several advantages but there are certain constraints of K-NN like slow
multivariate split condition is equivalent to an SVM classifier (Shaikh algorithm, the curse of dimensionality, the need for homogeneous fea­
and Algorithms, 2018). SVM maximes the distances between nearest tures, the optimal number of neighbor and outlier sensitivity, etc.
data points and hyper-plane that ultimately becomes the deciding right Random Forest: The random forest classifier is a collection of de­
hyper-plane known as Margin. Also, the SVM algorithm has a technique cision tree classifiers that was created to increase classification accuracy.
called the kernel trick that takes low dimensional input space and It is an ensemble algorithm that quickly identifies important information
transforms it into a higher-dimensional space, i.e., it converts a not from vast datasets. The best class of any sample is determined by each
separable problem into a separable problem. Within Sequential Minimal tree classifier in this ensemble, and the ultimate class label is determined
Optimization (SMO), Weka uses a specific efficient optimization by a majority voting procedure. Random forest is robust to outliers,
approach for SVM. The purpose is to enhance the separating hyper­ works well with non-linear data, has a lower risk of overfitting, and runs
plane’s margin. As for similarity values, the pairwise dot product be­ efficiently on a large dataset with optimal performance margins. These
tween specific training examples and between different test instances is pros also come with the cons like slow training, biased while dealing
used, which opens the gates for data transformations into multidimen­ with categorical variables, and not being suitable for linear methods
sional space. SVM works well with a clear margin of separation and is with a lot of sparse features.
effective in high-dimensional spaces. It is effective in cases where the Naïve Bayes Algorithm: It is a unique technique that is based on the
number of dimensions is greater than the number of samples. While SVM Bayes theorem’s well-known foundation and belongs to the probabilistic
is a simple and straightforward algorithm having lots of advantages, it method of classifiers (Shaikh et al., 2020). The secret to determining
also has some cons. It doesn’t perform well on a large data set (not class conditional probability in terms of posterior and prior probabilities
scalable). It also doesn’t perform very well when the data set has more is the Bayes Classifier, also known as the generative model. The
noise, i.e., target classes are overlapping. assumption that all features are independent makes the Naive Bayes
Decision Tree: It splits the data into the appropriate labels using algorithm very fast, and it works well with high-dimensional data. It is
split criteria. A single characteristic, known as Univariate, or many also used to solve multi-class prediction problems as it’s pretty helpful.
features, known as Multivariate, can be used as the splitting condition For categorical values, the Naïve Bayes algorithm performs exception­
(Shaikh and Algorithms, 2018). The goal is to split training data in a ally well. The assumption that all features are independent is not usually
recursive manner in order to maximise the discrimination of distinct the case in real life so it makes the Naive Bayes algorithm less gener­
classes across different nodes. To quantify the same, the Gini-index, alized. Also, if the test data set has a categorical variable of a category
Information gain, and Entropy are utilised. Decision Tree is easy to that wasn’t present in the training data set, the Naive Bayes model will
understand, interpret, inference, and visual algorithm that works for assign it zero probability and won’t be able to make any predictions in
numerical and categorical features. It requires little data preprocessing this regard. This phenomenon is called ‘Zero Frequency’.
and no need for hot encoding. It is a non-parametric model which as­ ii: Un-Supervised Machine Learning techniques: The final label of
sumes no shapes about the data. It doesn’t require normalization as well the data is not known, and the models following under these techniques
as scaling of data. On the negative side, the Decision Tree suffers from are made to find the respective clusters themselves based upon the

7
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

various distance matric suitable for that kind of data. Mainly used Ayoub Shaikh, 2019). It finds the direction in which the data spread is
classifier techniques under this category are K-means, Fuzzy C-means, maximum. In other words, PCA finds those features in the data with a
and Artificial Neural Network (ANN). Few unsupervised ML techniques maximum in between variance content and thus reducing redundancy.
are discussed below. PCA removes correlated features for compression, improves algorithm
K-means: K-means clustering is a straightforward unsupervised performance, reduces overfitting, and improves visualization of multi­
learning approach for resolving clustering issues. It uses a simple process dimensional data. PCA does also ha ave few disadvantages associated
to divide a given data set into a number of clusters, each of which is with it. Principal Components are not as readable and interpretable as
designated by the letter “k” which is predetermined. The clusters are original features. Data standardization is a must before PCA. Although
then positioned as points. All observations or data points are associated Principal Components try to cover maximum variance among the fea­
with the closest collection, computed, and adjusted before repeating the tures in a dataset, not selecting the number of Principal Components
procedure with the new adjustments until the desired result is achieved with care may miss some information compared to the original list of
(Shaikh et al., 2020). Search engines, market segmentation, analytics, features. The trade-off between information loss and dimensionality
and even astronomy use K-means clustering. reduction. Balancing the trade-off between dimensionality reduction
The formula is as follows: and information loss is, unfortunately, a necessary compromise that we
1. The initial set of centroids is represented by K points in the object have to make when using PCA.
data space. Linear Discriminant Analysis (LDA): LDA is a traditional ML
2. The closest k is allocated to each item or data point. technique that uses within-class and inter-class variances to develop a
3. The positions of the k centroids are computed when all objects projection matrix that maximises the separation between classes in the
have been allocated. projected space. It uses a system similar to PCA’s, but it considers the
4. Repeat steps 2 and 3 until the centroids’ positions do not move supervised information provided by the labels in this situation. LDA has
anymore. the benefit that it is a simple, fast, and portable algorithm. Coming to the
K-means algorithm is relatively simple to implement, scales to large prons side, LDA requires normal distribution assumption on features/
data sets, guarantees convergence, warm-starts the positions of cen­ predictors.Sometimes not good for a few categories of variables.
troids, easily adapts to new examples, and generalizes to clusters of Average Neighborhood Margin Maximization (ANMM) ANMM is
different shapes and sizes. But, there are certain issues related to the K- a dimensionality reduction-focused ML algorithm that attempts to
means algorithm, like ‘k’ value selection. K-means has trouble clustering overcome some of the constraints of PCA and LDA (Tawseef Ayoub
data where clusters are of varying sizes and densities. Centroids can be Shaikh, 2019). The goal of ANMM is to find a linear transformation that
dragged by outliers, or outliers might get their own cluster instead of maximises an average neighborhood margin, which is defined as the
being ignored. difference between a sample’s average distance to its nearest neighbours
Fuzzy C-means: Based on the distance between the cluster center of different classes and its average distance to its nearest neighbours of
and the data point, this algorithm assigns membership to each data point the same class for each instance.
corresponding to each cluster center. The closer the data is to the cluster iv: Deep Learning.
center, the more likely it is to belong to that cluster center. Clearly, the Deep Learning (DL) is a machine learning subfield that employs
sum of each data point’s membership should equal one. Fuzzy C-means ANN. The task of creating a new feature extractor for each new problem
draw the soft boundaries and give the best results for overlapped data is simplified by DL. The neural network (ANN) is at the heart of deep
sets. Datapoint is assigned membership to each cluster center as a result learning techniques. CNN-based classifiers can be trained directly on
of which data point may belong to more than one cluster. The issues raw images without the need for feature extraction by humans. Re­
related to this mode are Apriori specification of the number of clusters, searchers have been able to improve the accuracy of disease detection
and Euclidean distance measures can unequally weight underlying and other classifications category models thanks to recent advances in
factors. computer vision, hardware technologies, and deep learning. CNN has
Artificial Neural Network: Supervised training is used by almost all excelled in agriculture, automation, and robots, as well as artificial in­
of today’s highly successful neural networks. Kohen’s Self-Organizing telligence (AI), audio, video, and picture identification. The present
Map (KSOM), which is utilised for grouping high-dimensional data, is sections explain a few of the deep learning approaches employed in this
the only neural network that is employed with unsupervised learning. paper.
KSOM is a clustering technique that replaces the standard K-Mean Convolutional Neural Network (CNN): CNNs are feedforward
approach. Both could be designed depending on the problem at hand. By neural networks in which data is sent from one layer to the next. CNN
definition, neural networks are function approximators, and the goal is designs consist of input, hidden, and fully connected (output) layers.
to use them to estimate an unknown function. When you know both the Convolutional, ReLU (rectified linear unit), and Pooling layers are hid­
input and output of an anonymous function, the problem becomes su­ den layers combined to form a single network. The image’s pixel values
pervised. However, in some situations, when the input is known, but the are sent into the input layer in the form of an array (Shaikh et al., 2021).
output is unknown, the neural network can be employed in an unsu­ The process of feature extraction is done automatically by the hidden
pervised manner. ANN learning methods are quite robust to noise in the layers. Convolutional layers provide the function of a feature extractor.
training data. ANN has the ability to perform parallel processing of the The image matrix and filters are used to execute the convolution oper­
data. ANN are gradually being broken down, they will not suddenly stop ation. With the use of ReLU activation, ReLU adds non-linearity into the
working, and these networks are gradually being broken down. ANN network by setting all negative pixels to zero. CNN provides translation
models do come with some cons also, and the most debated in the equivariance, meaning that shifting in the input data does not alter the
literature is the black box nature which means it provides us the solu­ representation of the input but rather linearly shifts the input in the
tion, but how come it to the solution is not known to us. This unex­ latent space. They yield themselves to be more readily parallelizable.
plained functioning of the ANN model reduces its reliability. Also, the While CNN is offering numerous benefits, they are not quite robust to
parallel processing power of the ANN makes them hardware dependent. adversarial attacks, rotations, and reflections, they don’t have a sense of
iii: Dimensionality Reduction Techniques. memory state, and they are rather limited for sequential modeling.
By finding a linear transformation from the dataset space to a lower Recurrent Neural Network (RNN): The way RNNs and feed-
dimensions Euclidean space, dimensionality reduction algorithms forward neural networks channel information gives them their names.
attempt to learn a distance. The information in a feed-forward neural network only flows in one
Principal Components Analysis (PCA): In unsupervised ML, PCA is direction: from the input layer to the output layer, passing through the
one of the most used dimensionality reduction approaches (Tawseef hidden layers (Shaikh et al., 2021). The data travels straight through the

8
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

network, never passing through the same node twice. Feed-forward increasing profitability, and improving soil health. pH values and soil
neural networks have no recollection of the information they receive fertility indices classification and prediction model were presented by
and are poor predictors of what will happen next. A feed-forward Suchithra and Pai (Suchithra and Pai, 2019). pH values and Soil Organic
network has no concept of time order because it only analyses the cur­ Matter (SOM) are critical markers of soil fertility; according to Yang
rent input. Except for its training, it has no recollection of what tran­ et al. (Yang et al., 2019), hence the authors predicted SOM and pH pa­
spired in the past. It examines the current input and what it already rameters in paddy soil. Organic carbon (OC), nitrogen (TN), and mois­
possesses when making a judgment. RNN offers a principal advantage ture content (MC) parameters of the soil have been predicted by
over ANN by modeling a collection of records (i.e., time collection) so Morellos et al. (Morellos et al., 2016). The goal of this research was to
that each pattern can be assumed to be dependent on previous ones. compare machine learning algorithms and linear multivariate algo­
Also, they are used with convolutional layers to extend the powerful rithms in terms of prediction performance. Seed germination is a critical
pixel neighborhood. element in seed quality, and a key determinant of yield and product
quality. Huang et al. (Huang et al., 2019) and Zhu et al. (Zhu et al.,
3. Computational Machine learning techniques in agricultural 2019) have shown applications of many computer vision, machine
automation learning, and convolution neural network (CNN) algorithms in auto­
mated seed sorting and soil quality estimations. Veeramani et al.
While performing agriculture tasks, the steps below are generally (Veeramani et al., 2018) employed a CNN-based deep neural network
followed by farmers. (DNN) model to estimate the number of seeds per pod in soybeans and
Step 1: Selection of Crop. categorize haploid seeds based on form, phenotypic expression, and
Step 2: Land Preparation. embryo position. To improve the accuracy of the classification method,
Step 3: Seed Sowing. Keling et al. (Keling et al., 2018) employed a multilayer perceptron
Step 4: Irrigation & fertilizing. neural network model to separate high-quality pepper seeds from low-
Step 5: Crop Maintenance [use of pesticides, crop pruning, etc.]. quality pepper seeds. A detailed summary of work done by different
Step 6: Harvesting. authors on soil and seed parameters is mentioned in Table 2.
Step 7: Post-Harvesting activities. The research displays that compared to the single predictor selection
As per the above algorithm, the agriculture-related tasks are cate­ method based on a single feature predictor model, the selection based on
gorized into three major sub-areas. Fig. 7 shows these three sub-domains multilayer perceptron network and binary logistic regression models
of agriculture tasks. In the following sections, the review of the most obtained higher germination percentages while maintaining a high se­
recent techniques of machine vision systems used for classification and lection rate. The multilayer perceptron neural network model, with 15
object detection in the pre-harvesting stage of farming is presented. features chosen as covariates, can be implemented within the agricul­
tural industry because of its promising outputs in the previous contri­
butions. The research visualizations indicate that the embryo’s features
3.1. Pre-harvesting are being extracted by the network, which may be used further to
classify the seeds, as are carried out manually by agricultural field
The real growth of the crops is influenced by pre-harvesting condi­ workers. Also, deeper architectures provide better classification accu­
tions. Machine learning is utilized in pre-harvesting to record soil, seed racies as compared to shallower architectures. Also, the use of a shal­
quality, fertilizer treatment, pruning, genetic and environmental vari­ lower network (VGG19) for transfer learning helps in the reduction of
ables, and irrigation factors. It is critical to concentrate on each training loss significantly. For the VGG19 to identify multiple defects of
component in order to reduce total production losses. A few vital pre- seeds and the difference between some seeds with and without defects, it
harvesting components are discussed, as well as how neural networks is necessary to extract deeper features like GoogLeNet, which had an
and machine learning are utilised to capture the parameters of each excellent ability to extract features based on the inception structure.
component in this section. ELM classifier with its different activation functions are used to predict
the fertility levels as low, medium and high using the soil parameters as
3.1.1. Seed and soil input values; and to predict the pH levels of the corresponding soil using
The classification and assessment of soil attributes aid farmers in different activation functions which include the hyperbolic tangent
reducing fertilizer costs, reducing the need for soil analysis experts,

Fig. 7. General Categorization and Important Parameters Considered at each Agriculture Task.

9
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

Table 2
Analysis of Pre-harvesting Parameters of Soil and Seed.
References Property Dataset used (Public / Own Models / Method / Algorithms compared Best model / Results
method/ algorithm

Suchithra and Pai Soil Public (reports available Extreme Learning Machine (ELM) with ELMs with Gaussian 80.00% accuracy
(2019) ( during the years 2014 to different activation functions like sine- radial basis function
Suchithra and 2017) squared, Gaussian radial basis, triangular
Pai, 2019) basis, hyperbolic tangent, and hard limit
Yang et al. (2019) ( Soil Own Four Machine Learning models Cubist ELM coefficient of determination (R2)
Yang et al., 2019) regression model (Cubist), extreme learning = 0.81
machines (ELM), least squares-support vector
machines (LS-SVM), and partial least squares
regression (PLSR)
Morellos et al. Soil Own Cubist, partial least squares regression (PLSR), LS-SVM is best for Root mean square error of
(2016) (Morellos least squares support vector machines (LS- MC and OC, and TN prediction (RMSEP):0.457%,
et al., 2016) SVM), and principal component regression is best by the Cubist residual prediction deviation
(PCR) (RPD):2.24
Huang et al. (2019) Seed Own Ensemble learning, K-nearest neighbor (KNN), GoogleNet 95.00% accuracy
(Huang et al., logistic regression, support vector machine
2019) (SVM), and Speeded Up Robust Features
(SURF) algorithm to classify the extracted
features, GoogLeNet, VGG19
Zhu et al. (2019) ( Cotton Own, dataset collected from SVM, PLS-DA, and LR models based on deep self-design CNN 80% 80.00% accuracy
Zhu et al., 2019) Seed Shihezi, Xinjiang Uyghur features extracted by self-design CNN and
Autonomous Region, China ResNet models
Veeramani et al. Haploid DeepSort, Support Vector DeepSort DeepSort 96.10% accuracy
(2018) ( maize Machine (SVM), Random
Veeramani et al., seeds Forest (RF), and Logistic
2018) Regression (LR)
Keling et al. (2018) Pepper Own Multilayer perceptron (MLP); BLR binary multilayer 90.00% accuracy
(Keling et al., seeds logistic regression, single feature models perceptron and
2018) binary logistic
regression

function (elm_- tanh), triangular basis transfer function (elm_tribas), for example, such as Gamaya, employs precisely this practice. Trace
hard limit transfer function (elm_hardlim), gaussian radial basis func­ Genomics’ (Khirade and Patil, 2015) approach to soil health differs in
tion (elm_grbf) and sine-squared function (elm_sinsq). that it looks at soil quality and how the particular plant species respond
to the quality of the soil rather than at the conditions surrounding them.
3.1.2. Crop disease Detection, Prediction, and monitoring ML is used to keep crops healthy so that the company is trying to prevent
Plants are vulnerable to infection because they are surrounded by the disease instead of finding it. Some farmers ship their soil directly to
outside environment. The spread of disease depends on the current crop Trace Genetics and receive an evidence-based answer about the soil
conditions and susceptibility to infection. It is crucial to identify plant condition in advance. Here, a detailed survey of the recent techniques in
diseases to avoid any losses in the yield and quantity of agricultural computer science for automated disease diagnosis and detection in the
products. Ranging from leaves, stems, seeds, and roots of the plant to agricultural domain is performed.
flowers, fruit, and seeds in the plant (Gulve et al., 2015). For many re­
gions of the world, therefore, early diagnosis is problematic. Smart­ 3.1.2.1. Detecting and classifying crop diseases using Conventional data
phone diagnosis is increasingly being enabled by deep learning advances mining methods. Computer vision can use various classification and
in computer vision. Risk reduction is critical because of the tedious task detection algorithms and methods for crop disease. Using deep convo­
of spotting and then counting diseases among large populations of crops lution neural networks (DNNs) helped (Guo et al., 2020) to achieve a
on large farms; thus, automation is helpful. Even in underdeveloped 99.53% correct result in diagnosing the plant disease. Neural networks
countries, most infections are too subtle for the naked eye to perceive, have also been applied to issues in rice, like identifying illness. Gener­
and for that reason, medical assistance is critical. Therefore, using image ally, principal component analysis (PCA) (Mehra et al., 2016), support
processing software to help classify diseases of plants became a high vector machines (SVM) (Schor et al., 2016), and multiple linear models
priority, given that many species could not be distinguished by looking are alternative models most often used in the smart farming context.
at photographs alone. While discovering and classifying their visual similarities, color and
The wide range of soil conditions and nitrogen levels are vital players texture have been found to be helpful in agriculture and horticulture
in the good or bad health of crops. Pesticides were traditionally applied disease detection techniques. The start-up Prosper developed a system
equally to each square of the field. If farmers are overly budgeting for the that consists of networked cameras, sensors, and an algorithm based on
water use on their farms, it significantly reduces the number of species machine learning to keep track of crops and notify farmers as soon as
that are pollinators for flora and fauna. Machine learning (ML) can be they’re sick (Castro and New, 2016). By applying neutral network pro­
used in conjunction with analysis software for crop health assessment. cessing procedures to hyperspectral data, Golhani et al. (Golhani et al.,
The data is used to pinpoint which infestation areas are most critical, 2018) concentrated on the study of plant disease.
thus allowing farmers to target pesticide usage in those specific loca­ Based on an experiment conducted by Rangan et al. (Rangarajan
tions. Environmental customization can be significant. An example of et al., 2018), where they generated a dataset of six types of tomato leaves
this is Plantix, a German start-up, which combines machine learning from PlantVillage for the classification of tomato crop disease. Both the
with image recognition in an app to identify plant diseases and nutrients AlexNet and the VGG16 net used a deep learning architecture, and the
(Khirade and Patil, 2015). Tools like these make a difference for smaller dataset from PlantVillage was used as input to it. The accuracy of clas­
farmers because they’re flexible. Bigger businesses use digital platforms sification of VGG16′ s 13,262 image signatures was 97.29%, and of
connected to the internet of things (IoT) devices to collect both visual AlexNet’s was noted at 97.49%. Machine vision techniques are more
and thermal data. A concept commonly used by Swiss-based companies, effective in identifying plant disease at the beginning stages of growth

10
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

rather than in the late stages (Backhaus et al., 2011). The start of process An improved artificial plant optimization (IAPO) calculation using ML
begins with the preparation and acquisition of a sample. The machine has been presented in (Sengar et al., 2018) that distinguishes the PD &
vision process has been applied to crops like rice (Chung et al., 2016), arranges the leaves into sound & tainted on a dataset of 236 pictures.
papaya, and chili pepper (Habib et al., 2018) with accuracies of 87.9%, (Janarthan et al., 2020) gave the acknowledgment and categorization of
87.5%, and 90.15%, respectively. Fig. 8 displays a demo of the diseases maize plant leaf illnesses by utilizing a deep forest method. In (Ali et al.,
in apple fruits and apple leaves. 2017), a global pooling dilated CNN (GPDCNN) is suggested for recog­
The SVM classifier was used by researchers to detect grape leaf dis­ nizing PD. The authors in (Bah and Hafiane, 2018) developed a vision-
ease (Ji et al., 2020). Powdery mildew & DownyMildew are two grape based program to detect symptoms of olive quick decline syndrome
plant diseases that were identified with an accuracy rate of 88.89 (OQDS) on leaves of Olea europaea. Again, a convolutional neural
percent (Gavhale et al., 2014). In (Sreedhar and Kumar, 2020), authors network for end-to-end detection of red grape vine (Sangiovese) using
detected diseases such as anthracnose and canker on oranges, limes, color images of leaf clippings is recommended (Zhang et al., 2019).
lemons, and grapefruit citrus trees. The research outcome received an In (Selvaraj et al., 2019), researchers developed a system to detect
actual accuracy rate of 95 percent. The authors showed an overall disease in glycine max with a k-means-based segmentation algorithm.
classification accuracy of around 90 percent (Kaur et al., 2018) while The design catches downy mildew, frog eye, and septoria leaf blight
detecting septoria leaf blight, frog eye, downy mildew, and diseases in from images collected by the PlantVillage project. Again, in (Singh and
soybean crops by using a large dataset. The authors of (Sujatha et al., Singh, 2019), the authors proposed an intensity thresholding method for
2021) designed a system for diagnosing diseases in tea crops and measuring the progression of powdery mildew in cherries. A hybrid
detecting 3 different diseases by using limited features and claim to have method for detecting and identifying diseases in citrus plants is pre­
gained a 90 percent accuracy rate. In (Reddy and Sashirekhak, 2020), sented in (Saraansh et al., 2019). An approach based on color difference
the authors developed a system to determine disease presence in wheat to segment diseased parts of known mandarin leaves is given (Gargade
crop photos with a fuzzy classifier. The accuracy of unhealthy and and Khandekar, 2021). A method to detect diseases in beans using CNN
healthy leaves classification was 56 percent and 88 percent, respec­ is offered in (Sharifa et al., 2018). The research is carried out on wheat
tively. A thorough survey on crop disease identification with the KNN disease detection in (Padol and Yadav, 2016) and banana field images in
classifier is performed in (Too et al., 2019). The authors (Hossain et al., (Warne and Ganorkar, 2015). The study in (Kaur and Singla, 2016)
2019) used the GLCM feature extraction approach with the KNN clas­ proposed a method to classify apples into rotten or good apples and
sifier for cotton crop disease (Grey Mildew) and achieved 82.5 percent found SVM as the best classifier with 98.9% performance. An approach
accuracy. An in-depth data on plant disease detection using an ANN for apple leaf disease detection that achieved 98.54% classification ac­
classifier is offered in (Gupta et al., 2020). The authors (Arora and curacy is given in (Patil and Zambre, 2014). A hybrid method for the
Agrawal, 2020) analyzed various supervised machine learning algo­ detection and classification of diseases in citrus plants is recommended
rithms like RF, SVM, DT, KNN, NB, and KNN with image processing in (Sannakki et al., 2013) and achieves 97% classification accuracy on
methods to uncover the best algorithm for plant disease classification. the citrus disease image gallery dataset. Table 3 gives an idea of the state
RF algorithm achieved 89 percent accuracy compared with other of the art works on automated disease detection in the agriculture
algorithms. domain.
Similarly, the authors (Zhang et al., 2019) evaluated the perfor­
mance of SVM, RF, Stochastic Gradient Descent (SGD)) & DL (Inception- 3.1.2.2. Detecting and classifying crop diseases using deep learning. This
v3, VGG-16, VGG-19) in citrus plant disease detection in which DL section provides a brief summary of different DL methods for the
methods overtook the ML methods. The work (Cruz et al., 2017) illu­ transformation of traditional agriculture into a creative and intelligent
minates various types of plant disease (PD), and different progressed ML practice. A plant disease detection model was constructed using the
& image processing techniques to identify plant PD. (Cruza et al., 2019) Berkley vision and learning center DL framework (Zhu et al., 2018). The
aimed at tweaking & assessing cutting-edge deep CNN for picture-based model is able to identify 13 different types of disease. An approach
PD characterization. (Kaur et al., 2018) proposed a method for plant leaf proposed in 2017 that combined CNN and K-means feature for plant
disease detection (PLDD) & characterization utilizing the KNN classifier. disease identification and suppression is put forward in (Sladojevic

Fig. 8. Diseases in Apple Fruits and Leaves (a) Apple Scab (b) Apple Blotch (c) Apple Rot (d) Rust disease on the leaves of an apple tree (e) Apple Mosaic and (e)
Powdery Mildew.

11
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

Table 3 Table 4
A Summary of the Papers Dealing with Detecting and Classifying Diseases using A Summary of the Papers Dealing with the Task of Detecting and Classifying
Conventional Data Mining Methods. Diseases using Deep Learning.
References Diseases Methods Precision Reference Title Deep Learning Algorithms used

Padov and Yadav Grape leaf disease; SVM 88.99% Yu et al. (2016) (Yu Dual-camera high CNN-RNN
(2016) (Padol and Powdery mildew, et al., 2016) throughput phenotyping
Yadav, 2016) downy mildew (HTP) platform on an
Warne and Ganorkar Cotton leaf diseases; ANN 89.56% unmanned aerial vehicle
(2015) (Warne and Red leaf spot, (UAV) for large scale
Ganorkar, 2015) Cercospora leaf spot, soybean yield
Alternaria Chu and Yu (2020) (Chu An end-to-end model for Propagation Neural
Kaur and Singla Potato late blight ANN 100.00% and Yu, 2020) rice yield prediction using Networks (BPNNs) and
(2016) (Kaur and deep learning fusion Independently Recurrent
Singla, 2016) Neural Network (IndRNN)
Patil and Zambre Cotton leaf disease SVM 89.00% Tedesco-Oliveira et al. Convolutional neural Convolutional Neural
(2014) (Patil and (2020) (Oliveira et al., networks in predicting Networks (CNN)
Zambre, 2014) 2020) cotton yield from images
Sannakki et al. (2013) Grape leaf disease; BPNN 100.00% of commercial fields
(Sannakki et al., Powdery mildew, Nevavuori et al. (2019) ( Crop yield prediction with Convolutional Neural
2013) downy mildew Nevavuori et al., deep convolutional neural Networks (CNN)
Barbedo (2019) ( Plant Diseases CNN 96.17% 2019) networks
Barbedo, 2019)
Liu et al. (2018) (Liu Apple leaf diseases AlexNet 91.66% Maimaitijiang et al. Soybean yield prediction Deep Neural Networks
et al., 2018) Precursor, VGG (2020) (Maimaitijiang from UAV using (DNN)
19, Insecption et al., 2020) multimodal data fusion
Kour and Arora Apple fruit diseases FRADD 91.66% and deep learning
(2018) (Kour and Yang et al. (2019) (Yang Deep convolutional neural Convolutional Neural
Arora, 2018) et al., 2019) networks for rice grain Networks (CNN)
Turkoglu and Hanbay Plant disease and pest EML, SVM 98.00% yield estimation at the
(2019) (Turkoglu detection ripening stage using
and Hanbay, 2019) UAV-based remotely
sensed images
Khaki and Wang (2019) Crop Yield Prediction Convolutional Neural
et al., 2016). Identifying and generalizing features may lead to the (Khaki and Wang, Using Deep Neural Networks (CNN)
2019) Networks Deep Neural
degradation of consistent results due to human errors in creative design.
Networks (DNN)
This indicates that the application of DL and K-means resulted in a Rahnemoonfar and Real-time yield estimation Convolutional Neural
92.89% accuracy of recognition (Tang et al., 2017). Sheppard based on deep learning Networks (CNN)
Istanbul Technical University said in 2017 that CNN algorithms (2017) (
outperform feature-based learning algorithms in the discrimination of Rahnemoonfar and
Sheppard, 2017)
phenological stages (Yalcin, 2017). A new deep learning model for plant
Chen et al. (2019) (Chen Strawberry Yield Faster Region-based
phenotype was proposed by Namin et al. (Namin et al., 2017). New et al., 2019) Prediction Based on a Convolutional Neural
LSTM designs were created by combining CNNs. While CNN is partic­ Deep Neural Network Networks (Faster R-CNN)
ularly beneficial for reducing the error from sensor readings due to its Using High-Resolution
time-series data (Minh et al., 2017), recurrent neural network (RNN) is Aerial
Orthoimages
additionally used for crop yield prediction, which uses a time series Sun et al. (2019) (Sun County-Level Soybean The combination of
approach to reduce sensor-derived bias. The work in (Bu and Wang, et al., 2019) Yield Prediction Using Convolutional Neural
2019) implements two RNN-based classifiers, one a long short term Deep CNN-LSTM Model Networks and Long-Short
memory unit (LSTM) model and another a gated recurrent unit (GRU), in Term Memory Networks
(CNN-LSTM)
winter plant phenology mapping to find out how much water to give the
Khaki et al. (2020) ( A CNN-RNN Framework Prediction The
crops and the crop environment. Table 4 presents state-of-the-art deep Khaki et al., 2020) for Crop Yield combination of
learning works on automated disease detection in the agriculture Convolutional Neural
domain. Networks and Recurrent
The authors in (Shin et al., 2021) used deep learning (DL) to detect Neural
Networks (CNN-RNN)
powdery mildew (PM) and persistent fungal disease in strawberries to Terliksiz and Altylar Use Of Deep Neural 3D Convolutional Neural
reduce the amount of unnecessary fungicide use and the need for field (2019) (Terliksiz and Networks For Crop Yield Networks (3D CNN)
scouts. ResNet-50 gave the highest CA of 98.11% in classifying the Altylar, 2019) Prediction: A Case Study
healthy and infected leaves; however, considering the computation Of Soybean Yield in
Lauderdale County,
time, AlexNet had the fastest processing time, at 40.73 s, to process 2320
Alabama, USA
images with a CA of 95.59%. The four selected CNNs, AlexNet, Squee­ Lee et al. (2019) (Lee A Self-Predictable Crop Based On Crop Diseases
zeNet, GoogLeNet, and ResNet-50, were found to be commonly used in et al., 2019) Yield Platform (SCYP) Using Deep Learning
agriculture applications (Jiang et al., 2019). In (Durmus et al., 2017), Convolutional Neural
authors evaluated AlexNet and SqueezeNet for detecting disease in to­ Networks (CNN)
Elavarasan and Vincent Crop Yield Prediction Deep Recurrent Q-
matoes and yielded similar accuracies. Similarly, GoogLeNet (Inception) (2020) (Elavarasan Using Deep Reinforcement Network
has been used for disease detection in cassava (Ramcharan et al., 2017; and Vincent, 2020) Learning Model for
Szegedy et al., 2016). Lastly, (Fuentes et al., 2017) compared different Sustainable Agrarian
versions of the ResNet algorithm for detecting tomato diseases, where it applications
Wang et al. (2020) ( Winter Wheat Yield The combination of
was determined that ResNet-50 outperformed ResNeXt-50. The authors
Wang et al., 2020) Prediction at County Level Convolutional Neural
(Mohanty et al., 2016) trained a deep convolutional neural network to and Uncertainty Analysis Networks and Long-Short
identify 14 crop species and 26 diseases. The trained model achieves an in Main Wheat-
accuracy of 99.35% on a held-out test set, demonstrating the feasibility (continued on next page)

12
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

Table 4 (continued ) recent years (Feng et al., 2021). The general overview of phenotyping
Reference Title Deep Learning Algorithms used across various scales is depicted in Fig. 9 below.

Producing Regions of Term Memory (CNN-

China with Deep Learning LSTM)
3.1.4. Optimised management of input fertilisers and pesticides
Approaches Filtration of soil involves enormous effort and money. That study
Ju et al. (2020) (Ju Machine learning Short Term Memory found that a conservative estimate says about 5.6 billion pounds of
et al., 2020) approaches for crop yield (LSTM) Convolutional pesticides are applied every year globally. Pattern recognition and de­
prediction with MODIS Neural Networks (CNN),
cades of crop conditions allow an effective reduction in pesticide usage.
and weather data Long- Stacked-Sparse
Autoencoder (SSAE) Identifying pests is made easier with image-based pest control apps, and
then one can use an app to guide in finding the proper one. Agents help
deliver only the appropriate amount of pesticides to the crop, as
of this approach. excessive quantities can be dangerous to humans. Diverse environ­
mental conditions, such as improper temperature and humidity during
3.1.3. Crop phenotyping plant production cycles, allow plant pathogens, such as fungal and
The total of all observable characteristics of an organism can be nematode microorganisms and rodents, to proliferate. Thirty years ago,
defined as phenotype. The essential features include personality, products for farmers that contained agrochemicals started appearing on
biochemical properties, size, and color. The plant’s ontogenic, physio­ the market and rapidly changed agriculture. With new technological
logical, anatomic, and biochemical characteristics vary with the stage of advances in chemical treatments, including pesticides, antibiotics, and
development, while it has constant morphological characteristics (Guo insecticides, farmers finally made unwanted insects and bacteria less of a
et al., 2017). Additionally, during growth and development, the problem. Of course, crop damage is lessened by using these chemicals,
phenotype incorporates many functions, structures, processes, and but on both the environment and human health, and it does so in
structures. For cultivar development, successful breeding efficient different ways like more environmentally friendly and less toxic chem­
phenotype evaluation is required. Research bears that the main chal­ icals. The researchers at the University of Delhi were able to produce
lenge in plant breeding is improving the yield. chemicals only for specific insect species using ML tool-based technol­
In order to increase the time and lower the cost of data acquisition ogy called NeuroPIpred. However, in insects and certain other animals
and analyses, Dee and French (Dee and French, 2015) aimed to propose such as mollusks, so-like neuropeptides are considered necessary for
an automated system that could identify and measure objects from an their biological and behavioral activities such as metabolism and
image without human assistance in 2015. Coppens et al. in 2017 (Cop­ reproduction. NeuroPIpred takes in an insect neuropeptide composition
pens et al., 2017) present advances in robotized imagery characteriza­ and turns this into a new kind of insect poison that kills only the target
tion techniques which allow increases in the number of phenotype species. Agrochemical pollution has been a worldwide concern for the
evaluations, mitigating the supposed measurement bottleneck for past few decades.
functional genotyping research. In 2016, Bai et al. (Bai et al., 2016)
utilized five different types of sensors, ultrasonic distance sensors, 3.1.4.1. Pest detection using Conventional data mining methods. Several
portable spectrometers that use red, green, RGB cameras, and near- works have been carried out by researchers for rice pest detection. Zhou
infrared spectroscopy for high throughput phenotyping in plant et al. (Zhou et al., 2019) proposed an enhanced area-based fully con­
breeding and developmental phenotyping. There are four different volutional networks panicle identification and counting system in order
hyperspectral sensor technologies: push broom, filter-based, non-imag­ to detect and count the rice in the field, and this system can be used to
ing, and two, which are specialized for various purposes. They may be automate the rice phenotype measurements. Liu et al. (Liu et al., 2019)
used for the resistance of a particular disease in agriculture, however. proposed a UAV-based computer vision system to identify the agricul­
Another prominent example of a set of an application of mathe­ tural airboat position information, which is used in autonomous fertil­
matical algorithms to classification is the simplex volume maximization izing as well as herbicide applications. The presented system can inspect
(SVM) approach (Mahlein et al., 2019). SVM methods are primarily the paddy field with the help of UAV. Barrero et al. (Barrero and Per­
applied in the literature for classification and stress phenotyping. domo, 2018) proposed a method to fuse high-resolution RGB and low-
However, an improved understanding of the process may make it resolution multispectral images for identifying the Gramineae weed in
possible to apply methods such as K-means clustering, ANN, Gaussian the case of rice fields dealing with plants by considering fifty days after
Mixture Models, etc., more efficiently. Autonomous ground vehicles can emergence. Li et al. (Li et al., 2019) focus on Capsnet to track rice im­
be used for things like a telepresence robot that can test plant stalks for ages. The Capsnet is used to train classification, and the output vector is
strength and gather plant genetic data with an array of sensors that do predicted by focusing on routing by agreement protocol. Kitpo et al.
not make contact, more importantly (Mueller-Sim et al., 2017). Another (Kitpo and Inoue, 2018) proposed an IoT-based drone implementation
phenotypic platform is a metaphorical platform (Naito et al., 2017). system for the detection and classification of rice diseases. Qin et al. (Qin
Autonomous ground vehicles (Vinobot) and mobile observation plat­ et al., 2016) focus on the impact of spraying parameters such as oper­
forms (Vinoculer) work together to form an architecture. This is ad­ ation velocity and height of UAV on the droplet deposition in the case of
vantageous because the ground vehicle could be outfitted with sensors rice canopy as well as the protection efficiency against the hoppers
to obtain specific information about individual plants, while the lookout (plant). Table 5 gives the state of the artworks on automated pesticide
tower could monitor an entire field and focus on particular specimens. tasks in the agriculture domain.
Remote sensing and on-the-the-the-ground technologies are yet another
way to access or collect data. Deep Phenomics (pre-trained neural net­ 3.1.4.2. Pest detection using deep learning methods. The authors in
works) provide pre-trained models for phenotyping, as well as personal (Albanese et al., 2021) present an embedded system enhanced with ML
application to plant scientists (Shafiekhani et al., 2017). Three experi­ functionalities, ensuring continuous detection of pest infestation inside
ments were performed to evaluate the image-based phenotype tasks: the fruit orchards. The embedded solution is based on a low-power
leaf counting test, the classification of mutants, and the age-regression embedded sensing system along with a neural accelerator able to cap­
work. Reynolds et al. (Ubbens and Stavness, 2017) studied cost- ture and process images inside common pheromone-based traps. Results
effective imaging devices and environmental sensors, focusing on their show how it is possible to automate the task of pest infestation for an
trade-off in investment. Lowering the cost of ecological sensors or unlimited time without the farmer’s intervention. Deep learning tech­
embedded mobile phone sensors has allowed new options such as niques have been used to achieve an entirely automated, real-time pest
“affordable phenotyping” or “cost-effective phenotyping” to emerge in

13
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

Fig. 9. Schematic Overview of Phenotyping Platforms and across Different Scales Context.

identification follow some basic approaches for pre-processing by

Table 5
removing the background and segmenting the lesion tissue of the
A Summary of the Papers Dealing with the Optimal use of Fertilizers & Pesticides
diseased plant. In (Jiang et al., 2021), the authors propose a method to
in Smart agriculture.
detect infections in apple fruit and timely prevention of further in­
Reference Goal Technique Used data
fections caused by environmental factors and claimed to have achieved
Boniecki et al. Prediction of 6 types Neural Digital images of an 88% recognition rate. Jiang et al. (Peng, 2019) proposed a CNN
(2015) ( of pests in an apple Networks pests model for apple leaves disease detection by improving CNN for real-time
Boniecki orchard
et al., 2015)
detection of the disease from an image dataset. A new mobile applica­
Rupnik et al. Prediction of Pests Random Forest, Data of pest tion to automatically classify pests using a deep-learning solution for
(2018) ( (population size and Time series supporting specialists and farmers is introduced in (Karar et al., 2021).
Rupnik et al., its dynamics) analysis: The study has been successfully validated on five groups of pests called
2018)
Aphids, Cicadellidae, Flax Budworm, Flea Beetles, and Red Spider, and
Lottes et al. Detection of beet Random Forest, Images of two
(2016) (Lottes sugar plants/weeds Random different fields the method displayed 99.0% accuracy. Various capabilities of state-of-
et al., Markov Field captured by robots the-art (SoA) object detection models based on CNN for the task of
(2016)”) detecting beetle-like pest insects on non-homogeneous images are
Yu et al. (2019) Detection of weeds DCNN Images of different studied in work (Butera et al., 2021). Table 6 gives some more state-of-
(Yu et al., places
the-art works on automated pesticide tasks in the agriculture domain.
2019)
Bosilj et al. Detection and SVM Vegetation images
(2018) (Bosilj classification of crops 3.1.5. Weed detection and management
et al., 2018) and weeds Weeds proliferate and interfere with yield and profits, followed by
Padalalu et al. Suggestion of the Naive Bayes temperature,
fungus. Currently, they are the most popular choice, but in terms of both
(2017) ( amount of fertilizer algorithm moisture
Padalalu and irrigation control effectiveness and cost, herbicides are problematic. Secondly, the longer
et al., 2017) herbicides are left in the soil, the more weeds develop resistance to them
and can survive. The system is about to change how farmers detect
infested regions fundamentally. Autonomous herbicide-spraying has
monitoring system by removing the human from the loop (Segalla et al., been implemented by California-based Blue River Technology’s crea­
2020). (Lima et al., 2020) is one of the recent works that exploit ML tion, called See & Spray, which can select and dispense herbicides only
techniques to classify insects. The researchers (Panchal et al., 2021) on unwanted plants. The technology reduces the number of chemicals
build a deep learning model having the ability to explore the meaningful required by 80%. Tailoring herbicide programs is also made easier
features in the task of classification automatically that actually prepare a because of the use of See & Spray’s weed profiling feature. Extreme over-
complete roadmap in order to detect the seriousness of the disease. The irrigation is a risk factor for pseudomonas, amoeba, and larvae of eels
VGG model reached its best, having an accuracy of 93.5% over existing and parasites and biocides (pharmaceuticals, chemicals, etc.) in the
validation data. Barbedo (Barbedo and Arnal, 2014) worked on the produce. Many data-mining techniques have been developed to study
image segmentation method to identify plant diseases in black and white the need for various crop and growth cycles for irrigation systems. Data
background and try to reduce human error while taking less time to mining is critical to measuring the consumption and distribution of
identify these diseases. water using climatic elements, crop factors, and economic factors. In a
Many of the papers (Hiary et al., 2011) on image-based plant disease farm, it is imperative to have control of the pests. Many researchers have

14
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

Table 6 water management is critical. Nevertheless, here again, this has more to
A Summary of the Papers on Optimal use of Fertilizers & Pesticides in Smart do with the environment as the global population and food consumption
agriculture Context Using Deep Learning. are increasing. This is a problem in the worldwide market, and ML-
Author Dataset used Best model Testing Parameters powered smart irrigation solutions could help to solve it. Advanced
accuracy meteorological and geographic sensor systems are connected to multiple
Chohan et al. PlantVillage VGG-19 98.3% 143 million in-field sensors or satellites that can measure the temperature, moisture,
(2020) ( rainfall, and growth data. Given enough data, the irrigation system be­
Chohan et al., comes smart and minimizes efforts to use water while increasing water
2020)
use. For example, among other things, California-based ConsertWater
Ferentinos PlantVillage VGGNet 99.5% 138 million
(2018) ( uses satellite data, weather, topography, and geographical location to
Ferentinos, determine the amount of water required for a particular field. The app’s
2018) beauty is that it doesn’t require ground sensors to be installed and works
Mohameth PlantVillage ResNet-50 98% 25 million all the time. Even though the app looks simple, it claims to save 30% of
et al. (2020) ( + SVM
Mohameth
the water that’s used by farmers.
et al., 2020) Several methods for water control in an agricultural setting, as well
Tiwari et al. Potato leaf VGG-19 + 97.8% 143 million as a water quality analysis, have been implemented (Machado et al.,
(2020) ( PlantVillage Logistic 2019). An automated irrigation system to conduct the study of water
Tiwari et al., dataset Regression
needs in the field is proposed in (Kamienski et al., 2019). These sensors
2020)
Khamparia Tomato, potato, CAE 86.78% 3.3 million are installed throughout the field to gather data regarding different
et al. (2020) ( and maize leaf factors, namely soil moisture, soil temperature, and soil composition
Khamparia extracted from (PH). The Naive Bayes technique is used to estimate the required
et al., 2020) PlantVillage amount of water. For an agricultural operation, this technique considers
Punam Bedi Bacterial Spot CAE + CNN 98.38% 9,914
the weather forecasts to regulate the amount of water supply to be
(2021) (Bedi disease in peach
and Gole, plants applied to the crops and recommends specific fertilizers. Perea et al.
2021) (Perea et al., 2019) propose an intelligent irrigation system that ensures
a just and well-considered use of water by integrating the Decision Tree
with the Non-Dominated Sorting Genetic Algorithm II (NSGA II) genetic
studied plant mapping by using ML, though (ST, Suit and R Kumar­
algorithm.
aswamy, , 2019). Lots of unmanned aerial mapping machines for opti­
Xie et al. (Xie et al., 2017), in 2017, suggest a smart irrigation system
mization of the field have been designed and used. IoT-controlled flying
consisting of the components: A solar demand estimation is used to
machines NB-IoT is suitable for handling and shaping a large amount of
determine the time and energy required to do the following task using
data. Table 7 gives some more state-of-the-art works on automated weed
the support vector regression method, and on-demand scheduling is
detection in the agriculture domain.
used to reduce the cost of irrigation. This feature takes advantage of
numerical weather prediction (NWP) and incorporates it into the time of
3.1.6. Water analysis and optimal irrigation
use model (TOU). Results showed that water and energy resources could
If the fields are to be productive, they must be irrigated frequently.
be saved by 7.97%, and this resulted in a total amortization of 25.34%.
ML uses historical data to ensure adequate irrigation of each field’s
Goldstein et al. (Chen et al., 2020) devised an irrigation recommenda­
unique moisture levels. Optimal irrigation had been a challenge for
tion system in 2018. Several regression and classification algorithms
farmers in the past, but now machine learning has taken care of it. The
were applied using several prediction models. Data from different
ML robots can also measure the moisture of the field in real-time and
sources, including weather sensors, meteorological, and sensor-based
ensure that irrigation water application on all points of the field is
systems, was utilized to make real-time control possible.
allocated in an isotonic pattern. Even with traditional agriculture, it is
The work (Abdullah et al., 2021) proposes a framework that enables
impossible to accomplish this on the cheap. The necessity of the water to
advanced fuzzy logic to control a pump’s switching time according to
the global food supply can be imagined, given that 70% of global
user-defined variables, whereby sensors are the central aspect of and
freshwater is for irrigation. Especially in regions where rainfall is scarce,
contributor to the system. An overview of IoT-based irrigation systems

Table 7
Machine Learning Techniques for Weed Detection.
Author/Year Problem Definition Targeted Crop Dataset Model Accuracy

Alam et al. (2020) (Alam et al., 2020) Crop/weed detection and Unspecified Images were collected from private farm RF 95.00%
classification
Y-H Tu et al. (2019) (Tu et al., 2019) Measuring Canopy structure Avocado tree Avocado field Bundaberg, Australia RF 96.00%
J Gao et al. (2018) (Gao et al., 2018) Weeds recognition Maize Images were taken from crop field of Belgium RF 81.00%
KNN 76.95%
Castro et al. (2018) (Castro et al., Early Weed mapping Sunflower, Images were taken from crop fields of Spain RF 87.90%
2018) cotton
Aaron et al. (2019) (Etienne and Weed detection by UAV Maize Images were collected from private farm NVDI digit, YOLOv3 98.00%
Saraswat, 2019) Detector
D Chabot et al. (2018) (Chabot et al., Monitoring water aquatic Stratiotes Trent-Severn Waterway in Ontario, Canada RF 92.19%
2018) vegetation aloides
Brinkhoff et al. (2020) (Brinkhoff Land cover mapping 9 perennial Images were taken from the Riverina region SVM 84.80%
et al., 2020) crops in NSW, Australia
Zhang et al. (2019) (Zhang et al., Weeds species recognition 8 weed plants 1600 weed images were taken from South SVM 92.35%
2019) China crop field
Adel et al. (2018) (Bakhshipour and Weed detection using shape Sugar beet Images were taken Shiraz University, Iran SVM 95.00%
Jafari, 2018) feature
Abouzahir et al. (2018) (Abouzahir Weeds species detection Soybean Images were collected from Sâa José farm, SVM 95.07%
et al., 2018) Brazil

15
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

for agriculture is presented in (García et al., 2020). A review of moni­ Table 8

toring and control mechanisms for precision irrigation systems is Features of the Studies Dealing with the Irrigation Task in Smart Agriculture.
considered in (Abioye et al., 2020). IoT technologies, practices, and Reference Technology Monitoring Improvement/
future studies are investigated in (Ray, 2017) for smart agriculture, and Limitation
essential agricultural applications are highlighted and discussed. In Padalalu et al. Naive Bayes Estimation of the Sensors data
these studies, a detailed survey is carried out on smart agricultural (2017) ( algorithm precise amount (humidity, soil
monitoring in both simulated and real environments. A LoRaWAN and Kamienski of water and temperature, and soil
fog computing-based architecture for deploying smart irrigation systems et al., 2019) suggestion of the type (PH)) and
necessary weather forecasting
is proposed in (Lamas and CM-Echarri, L Azpilicueta, LP Iturri, F Fal­ fertilisers sites
cone, and F Carames, , 2020). In (Han et al., 2020), a graphical user Perea et al. Decision Trees, Prediction of the Crop, Julian day,
interface application for multifractal analysis of soil and plant structures (2019) ( Genetic Algorithm irrigation events bank holiday,
is presented for Windows platforms. In (Zaragoza et al., 2020), a user- Perea et al., weekday, and
2019) climatic data
friendly mobile application called REUTIVAR-App for fertigation
(temperature,
scheduling using reclaimed water is presented in olive groves, and daily humidity,
real-time irrigation and fertilization schedule recommendations at the precipitation event)
farm scale are monitored. The authors in (Syifa et al., 2020) used drone Xie et al. Support Vector Minimization of Soil moisture
remote sensing techniques to detect trees with similar symptoms to trees (2017) (Xie Regression irrigation cost information and
et al., 2017) method, Irrigation cloudless irradiance,
infected with PWD. The SVM had an overall accuracy of 94.13%, which estimation numerical weather
is 6.7% higher than the overall accuracy of the ANN, which was 87.43%. algorithm and an information (cloud
The study proposes developing the low-cost unmanned aerial system optimisation cover, humidity,
(UAS) for precision agriculture tasks called AgriQ (Oca and Flores, model. precipitation) and
solar energy data.
2021). Table 8 gives some more state-of-the-art works on automated
Goldstein et al. Gradient Boosted The weekly Meteorological
irrigation control in the agriculture domain. (2018) ( Regression Trees irrigation plan station data, actual
Sengupta (GBRT), Boosted prediction irrigation records,
3.1.7. Precision farming systems and platforms et al., 2021) Trees Classifier historical sensor data
For large-scale farming, labor costs form a significant part of farming (BTC), and linear
regression models
expenses. Like many other steps in traditional plantation agriculture,
Lamas (2020) ( Zigbee, LoRaWAN, Solar panel, Agricultural
harvesting, too, uses an enormous workforce. Anything that causes the Lamas and and WiFi agricultural monitoring system
crop to remain unharvested will lower the return on investment. Robotic CM-Echarri, monitoring with energy
harvesters identify and extract the crops at the appropriate stage. This is L harvesting
Azpilicueta,
both good for saving on labor costs and helps assure that the quality of
LP Iturri, F
the harvest is preserved and delivered to the customer at the right time Falcone, and
and in the process. The ML techniques and tools employed in crop F Carames, ,
production also show up in livestock production. The machines do the 2020)
work and supervise or manage the farms. The datasets are combined to Han (2020) ( MATLAB Soil and plant GUI-based
Han et al., structures (soil application (for
aid in the diagnosis of disease and/or damage detection. This example,
2020) porosity, tree Windows OS)
as shown here in Fig. 10, gives an excellent view of the general health of branching,
the plants and fruit conditions. Drones use scanning systems that cover biochar porosity)
large areas on pre-set routes. Computer vision is used throughout the Zaragoza REUTIVAR-App Reclaimed It provides daily real-
(2020) ( water, irrigation, time irrigation and
production of these as well. Images are used for creating multi-spectral
Zaragoza and fertilization fertilization schedule
images for crop analysis, plant monitoring, weed detection, and to et al., 2020) management recommendations at
determine if the disease is present and if plant health and drought are the farm scale.
affecting crop productivity. When taking the total amount of potential Pereira (2016) Wemos Mini D1, temperature, They proposed a
yields into consideration, this is a reasonable estimate of the crop yield. (Pereira Dg-2000 Ammonia relative prototype (a low
et al., 2020) Detector, DHT22, humidity, cost, hardware, and
More widely used robot applications include weed picking and precision
LDR, and MQ-137 luminosity, and software for
chemical application employing machine vision. electrochemical concentration of monitoring).
Various different uses of drones and automated vehicles in smart sensor ammonia in the
agriculture are discussed in the following sub-section. It’s a good match air
Ramli (2020) ( LoRaWAN and Crop, soil, They proposed a
for greenhouse horticulture due to its low maintenance demands and
Ramli et al., IEEE 802.11ac temperature, and reliable smart farm
controllable ecosystem. Environmental variability among crop param­ 2020) humidity. system.
eters makes it difficult for traditional agriculture and environmental Deng (2020) ( RFID and LoRa Temperature, They proposed a soil
regulations to adjust to plant growth cycles. IoT sensors and actuators Deng et al., moisture environment
are proposed to control the environmental conditions for a specific kind 2020) content, and monitoring system
chloride ion based on RFID
of plant species. Artificial neural network (ANN) determines the con­
concentration of sensors and LoRa.
ditions in the IoT cloud (Aliac and Maravillas, 2018). Aqeel-ur-Rehman, soil
2014 et al. (Rehman et al., 2014) reviewed the applications of WSNs in Pastor (2018) ( DCTA algorithm Orchid High and successful
agriculture and cited the necessity for sensors in this area. The principal Pastor et al., EC-5 and SHT-11 greenhouses data delivery rate
purpose of the authors’ work was to implement sensors and networks for 2018) sensor can be obtained
through DCTA.
beneficial and productive solutions in agriculture. Keshtari and Deljoo Thakur (2020) Arduino, Python, Water, soil The proposed system
(Keshtgari and Deljoo, 2012) applied WSN to farming in 2012. Usually, (Thakur Cloud, Soil moisture provides smart
WSNs are used for data collection, storage, collection, and data sharing. et al., 2020) Moisture and irrigation and
The primary goal is to be in tune with the real-time environmental Passive Infrared detection of
sensor intrusion.
properties of climatology and sensing data. High-quality remote moni­
toring and control for the design, construction, installation and verifi­
cation of a distributed WSN was the primary goal of this project.
Bhatnagar and Chandra (Bhatnagar and Chandra, 2020) focused on

16
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

Fig. 10. Autonomous Mobile Robots used in Precision Agriculture (a) An Unmanned See-and-Spray Machine (b) Agrobot Strawberry Harvester (c) Autonomous
Mapping Robot (d) Agriculture Water Spraying Robot (e) Agriculture Apple Harvesting Robot (f) Tomato Harvesting Robot (g) Autonomous Weed Removing Robot
(h) Agricultural Robot for Disease Detection.

soil health monitoring that showing the instantaneous reading of various et al. (Goswami et al., 2020) developed an IoT-based soil health moni­
parameters such as temperature, soil moisture, pH, and humidity on the toring system that monitors macronutrients N (Nitrogen), P (Phos­
screen of a farmer’s smartphone. Dasig et al. (Dasig, 2020) have dis­ phorus), K (Potassium), pH, soil moisture, and soil humidity. Authors in
cussed the implementation of IoT and WSN for precision agriculture in (Sengupta et al., 2021) designed and developed an IoT enhanced devi­
detail, emulating the advancement toward Agricultural 4.0. Yu et al. ce—FarmFox, which can analyze the sensed information and transmit it
(Yu, 2020) have developed “Plant Spike,” a low-cost, energy-efficient to the user via the internet. In (Cicioglu and Çalhan, 2021), the re­
WSN system for soil health monitoring. Nurzaman et al. (Nurzaman searchers propose a holistic smart agriculture application that consists of
et al., 2018) used IoT for crop health monitoring and also to maximize various agricultural sensors, drones, and IoT hardware and software
agricultural output. AgriTalk, a cost-effective IoT platform, was utilities. Table 9 gives some more state-of-the-art works on IoT and WSN
designed and tested for turmeric cultivation in (Chen, 2019). Goswami in the smart agriculture domain.

17
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

Table 9 Table 9 (continued )

Example of IoT/AI Applications in Smart Agriculture. Category Tool/Company Description
Category Tool/Company Description
farmers for effective decision-
allMETEO https A portal to manage IoT micro making.
://www.allmeteo. weather stations, to gather real- Cowlar A smart neck collar for monitoring
com time data access, and create a https://www. dairy animals to gather
Climate conditions weather map. It also provides an cowlar.com information on temperature,
Monitoring API for easy real-time data transfer rumination, activity and other
into developed or existing behavior. The intelligence
infrastructure. algorithm in the system allows for
Smart Elements A collection of products that the detection of health disorders
improve efficiency by eliminating before the appearance of visual
https://www. manual checking. They work by symptoms. It can monitor body
smartelements.io deploying a wide range of sensors movement patterns and gait to
generating a report back to an provide accurate oestrus detection
online dashboard, allowing rapid alerts. It uses a solar power base
and informed decisions based on unit, along with a waterproof and
real-time conditions. non-invasive monitoring system,
both comfortable for the animal
Pycno A software and sensor allowing and requiring minimum
https://www.pycno. continuous data collection and maintenance.
co flow from the farm to smartphone. FarmLogs This system monitors field
It also contains a dashboard to End-to end farm https://www.far conditions, facilitating the
apply the latest phenological and management mlogs.com planning and managing of crop
disease models to monitor trends systems production. It also markets
and assess the risk to agricultural agricultural products.
products. Cropio A decision-making tool used to
Farmapp A process of monitoring pests and https://www. optimize fertilisation and
www. farmappweb. diseases, generating reports for cropio.com irrigation to control the amount of
com mobile applications. It records the fertilizer and reduce the use of
data quickly and more efficiently water. It combines weather
Greenhouse than traditional methods (i.e., information and satellite data to
automation paper), allowing a smooth monitor crops and field forecasts.
implementation. The stored data is Farmshots A system analysing satellite and
synchronized with the server, Predictive https://www.farm drone images of farms fields to
enabling the following metrics to Analytics shots.com map potential sign of diseases,
be immediately observed: (1) a pests, and poor nutrition. It turns
satellite map with recorded points; images into a prescription map to
(2) the current sanitary status of optimise farm production and view
the farm; (3) comparative analytics on-farm performance.
heatmaps to easily compare Generated data in the cloud can be
previous measures with the current exportable into nearly all
situation; and (4) charts and agricultural software for
reports concerning pests and prescription creation.
diseases. aWhere A platform employed for weather
Growlink A platform that tightly integrates https://www.awh prediction and information on crop
https://www.gro hardware and software products, ere.com sustainability. Its goal is to deliver
wlink.com enabling smarter working, complete information and insight
including providing wireless for real-time agricultural decisions
automation and control, data on a daily basis and at a global
collection, optimization, and level.
monitoring and visualization. Plantix A machine learning based tool to
GreenIQ https:// A system to control irrigation and Crop and Soil https://www.pl control and manage the agriculture
www.easternpeak. lighting from all locations and to Health Monitoring antix.net process, disease control, and the
com connect IoT devices to automation cultivation of high-quality crops.
platforms. Trace Genomics A soil monitoring system
Arable A device that combines weather https://www. performing complex tests (i.e.,
Crop management https://www.ara and plant measurements, sending tracegenomics.com DNA) on soil samples.
ble.com data to the cloud for instant SkySquirrel A drone system aimed at helping
retrieval from all locations. It https://www.skysq users to improve their crop yield
offers continuous indicators of uirrel.ca and reduce costs. Users pre-
stress, pests, and disease. Agriculture program a drone route, and, once
Semios A platform focused on yield machines /drones deployed, the device will leverage
https://www.semios improvement. It enables farmers to computer vision to record images
.com assess and respond to insects, to be used for analysis. Once the
disease, and the health of crops drone completes its route, users
using real-time data, forming on- can transfer the data to a computer
site sensing, big data, and and upload it to a cloud drive. It
predictive analytics solutions for uses algorithms to integrate and
sustained agricultural products. analyse the captured images and
SCR/Allflex An advanced animal monitoring data to provide a detailed report on
https://www.scr system, aimed at the collection and the health and condition of crops.
Livestock dairy.com analysis of critical data, including See & Spray A robot designed to control weeds
monitoring and for individual animals. It delivers, https://www.sma and protect crops. It leverages
management when needed, the heat, health, and rtmachine.com computer vision to monitor and
nutrition insights required by precisely spray weeds and infected
plants.
(continued on next page)

18
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

Table 9 (continued ) MLP algorithm has maximum accuracy of 99%, a precision of 95%,
Category Tool/Company Description recall of 96%, minimum mean absolute error (MAE) of 0.04%, and root
mean square error (RMSE) of 0.006% for detecting sugarcane yield
CROO A robot that assists in the picking
https://www.harves and packing of crops. The
production. Kaburlasos et al. (Kaburlasos et al., 2020) presented a fuzzy-
tcroo.com manufacturer claimed that this based approach to predicting inclusion measure function for hellenic
robot can harvest eight acres in a sugar industry (HIS). Elavarasan et al. (Elavarasan et al., 2018) surveyed
single day and replace the work of publications on machine learning models associated with crop yield
thirty human labourers.
prediction based on climatic parameters. Liakos et al. (Liakos et al.,
Weather risk Arable offers an in-ground monitor that
https://www.ar collects and analyses 14 data about 2018) discussed the application of machine learning in the agricultural
able.com/ weather patterns, crop health, and sector. The authors in (Li et al., 2018) performed a review study on
soil quality. determining the ripeness of fruits to decide the optimal harvest time and
Farmers Edge provides farm management yield prediction.
https://www.farme software that collects and analyses
rsedge.ca data from satellites, weather
stations, as well as farm 3.1.8.2. Crop yield prediction using deep learning methods. Crop yield is a
equipment. Using image complicated task that is prejudiced by several factors, including geno­
processing and analytics, farmers type, environment, weather, and connections. Accurate yield prediction
are able to make data-driven
decisions about crop planting,
necessitates a fundamental understanding of the functional link between
cultivation, and harvesting. yield and these interaction components, which requires both large
Prospera autonomous crop management datasets and sophisticated algorithms to uncover. The crop yield forecast
www. prospera.ag solution that leverages their exist- problem is modeled using deep learning. It employs a semiparametric
15 ing algorithms to not only
deep neural network architecture that can account for complex
provide recommendations to
growers but also directly control nonlinear interactions in large datasets, as well as the known parametric
center pivots. structure and unobserved cross-sectional variability.
Blue River produces a machine that attaches Chu et al. (Schwalbert et al., 2020) studied innovative prototypes to
Technology to an existing tractor and precisely achieve in-season soybean yield forecasts in southern Brazil using Long-
www. blueriverte detects and applies herbicide to
chnology.com remove unwanted weeds from
Short Term Memory (LSTM), Neural Networks, satellite imagery, and
fields while avoiding crops weather data. Satellite and weather data were masked with a non-crop-
FarmBot open-source robotics project that specific layer based on field borders received from Brazil’s Rural Envi­
https://farm.bot/ consists of a Cartesian coordinate ronment Registry, which is required of all. The authors found (i) soybean
machine that uses software to
yield forecasts at the municipality-scale with a mean absolute error
automatically plant seeds, detect
and control weeds, and water (MAE) of 0.24 Mg ha − 1 at DOY 64 (march 5) (ii) a superior perfor­
plants mance of the LSTM neural networks relative to the other algorithms for
Plantix has built up the world’s largest all the forecast dates except DOY 16 where multivariate OLS linear
https://www.fastco database of plant diseases and uses regression provided the best performance. For rice yield prediction,
mpany.com image recognition and deep neural
Reddy et al. (Reddy and Kumar, 2021) proposed the BBI-model, a unique
networks to identify the plant type
as well as possible disease, pest, or end-to-end prediction model that integrates two back-propagation
nutrient deficiency neural networks (BPNNs) with an independently recurrent neural
FFRobotics uses deep learning algorithms to network (IndRNN). When the number of layers in the network was
https://www.ffrob identify the fruit, determine
adjusted to six, the BBI-model achieved the lowest mean absolute error
otics.com ripeness, and send a linear robotic
arm to harvest the fruit. (MAE), and root means square error (RMSE) for the summer rice pre­
diction (0.0044 and 0.0057, respectively) and equivalent values of
0.0074 and 0.0192 for the winter rice prediction. The study (Jeong et al.,
3.1.8. Crop and yield management 2022) offered an approach for pixel-scale early rice production predic­
Crop yield prediction is an essential task for the decision-makers at tion using a crop model and a deep learning model for various agricul­
the national and regional levels (EU level) for rapid decision-making. An tural systems in South and North Korea. By forecasting the ideal model
accurate crop yield prediction model can help farmers to decide on what roughly two months before harvest time, models of five alternative deep
to grow and when to grow. There are different approaches to crop yield learning network architectures were used to help determine the hybrid
prediction. The primary reason for farmers’ concerns is the assumption structure of long-short term memory (LSTM) and one-dimensional
that crop yields will be low, with the environmental parameters ex­ convolutional neural network (1D-CNN) layers. With unique spatial
pected to have the most significant impact on the market. Like any other patterns of rice yields for South and North Korea, the proposed model
type of modeling, it can be characterized as a machine learning problem. performed well [R2 = 0.859, Nash-Sutcliffe model efficiency = 0.858,
Extensive research has been done on this subject. If yield mapping was root mean squared error = 0.605 Mg ha1].
done using an ML technique, it could be used in farms utilizing IoT in­ The work (Zhou et al., 2021) developed an autonomous strawberry
formation for yield surveillance through the old molecularly. Informa­ flower detecting system with minimal labor and time costs for yield
tion collected will be used to provide data on land-use yields. In prediction. For the detection and counting of blossoms, mature straw­
addition, ML systems, together with IoT, are utilized in agriculture to berries, and immature strawberries, a quicker region-based convolu­
estimate and improve yields. There is a direct correlation between the tional neural network (R-CNN) was employed. The suggested algorithm
Internet of Things and agriculture. Other works demonstrate that ML is employed to count strawberry blooms from 2 m aerial photos in
systems can be built on the IoT and can give real-time feedback (Araby, November and December and compare the results to a hand count. The
2019). average deep learning counting accuracy was 84.1 percent, with a 13.5
percent average occlusion. Tello et al. (Tello and Ko, 2021) used mul­
3.1.8.1. Crop yield prediction using Conventional data mining methods. tispectral pictures to offer a deep learning-based approach for enhancing
The authors in (Wang et al., 2021) proposed an improved multilayer the yield forecast process of spring wheat (triticum aestivum). The re­
perceptron (MLP) approach to predict the amount of sugar yield pro­ sults contrasted wheat photographs throughout the season and
duction in IoT agriculture. Experimental results show that the proposed confirmed that the most critical flight periods for capturing images were

19
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

during late-heading, late-flowering, dough development, and harvest­ Table 10

ing. Red and red-edge were the most beneficial colour bands for yield A Summary of the Works of Optimal Crop Yield in the Smart Agriculture
prediction. The study (Alwis et al., 2019) used a Long Short Term Context.
Memory (LSTM) and Attention score system to determine the most Reference Target Model Used
beneficial elements for tomato yield utilising data from tomato growing Schwalbert et al. Satellite-based soybean Long-Short Term Memory
under smart farm conditions. Using deep learning approaches, Apolo (2020) ( yield forecast: Integrating (LSTM)
et al. (Apolo et al., 2020) created an automated image processing Schwalbert et al., machine learning and
methodology to detect, count, and estimate the size of citrus fruits on 2020) weather data for
improving crop yield
individual plants. For yield estimation per tree and total yield estima­
prediction in southern
tion, a model based on Long Short-term Memory (LSTM) was trained. Brazil
When the actual and estimated yields per tree were compared, the Reddy and MR Crop Yield Prediction Convolutional Neural
approximation error was found to be SE = 4.53 percent, and the stan­ Kumar (2021), ( (CYP) using various deep Networks (CNN), RNN
dard deviation was found to be SD = 0.97 kg. The technician’s esti­ Reddy and Kumar, learning techniques
2021)
mating error was SE = 13.74 percent, whereas the model’s mistakes Elavarasan, and Foreseeing the crop yield Deep belief network (DBN)
were SE = 7.22 percent and SD = 4083.58 Kg. Similarly, in (Bi and Hu, Vincent (2021) ( depending on climate, soil, and fuzzy neural networks
2021), a genetic algorithm (GA)-assisted deep learning solution Elavarasan and and water parameters system (FNN)
approach for agricultural yield prediction was developed. To compare Vincent, 2021)
Forsythe et al. Crop yield prediction using CNN-LSTM Model
the proposed method to other gradient-based algorithms, a case study of
(2021) (Forsythe, deep neural networks to
agricultural yield prediction is done. The results reveal that in terms of 2021) increase food security in
convergence speed and prediction accuracy, the suggested method beats Senegal, Africa
gradient-based methods. Table 10 gives some more state of artworks of Jeong et al. (2022) ( Scalable rice yields from a LSTM and 1D-CNN
deep learning in crop yield prediction. Jeong et al., 2022) crop model using targetted
deep learning techniques
AGEYE [https://www.ageyetech.com] is an automation platform for
Haque et al. (2020) ( Deep Neural Networks in Deep Neural Networks (DNN)
vertical farms and greenhouses that monitors every plant 24 h a day, Haque et al., crop yield
seven days a week, and uses cutting-edge artificial intelligence to 2020)
transform visual inputs into growth development insights and autono­ Nosratabadi et al. ( Crop yield prediction Artificial neural networks-
Nosratabadi et al., models based on hybrid imperialist competitive
mous actions, increasing harvest predictability and profitability as well
2021) machine learning method algorithm (ANN-ICA) and
as profitability. aWhere is a mission for the transforming agricultural artificial neural networks-
innovation campaign [www.climateshot.earth]. For 1.9 million virtual gray wolf optimizer (ANN-
weather stations worldwide, aWhere delivers daily updated historical GWO)
observed weather data and a 15-day prediction. Over 200,000 farmers in Zhou et al. (2021) ( Strawberry Yield You Only Look Once
Zhou et al., 2021) Prediction Based UAV and (YOLOv3)
Africa have benefited from aWhere’s digital climate advising services.
Near-Ground Imaging on a
aWhere collaborates with partners worldwide to install weather-based Deep Neural Network
crop advisory models to help rural farmers embrace climate-wise agri­ Sharma et al. (2020) Method to predict wheat DCCN
culture. Blue River, a team of innovators whose mission is to empower (Sharma et al., crop yield in India from
2020) publicly available satellite
customers, and farmers, to implement more sustainable solutions,
imagery
optimize chemical usage, reimagine routine processes, and improve Khan et al. (2019) ( Use Of Deep Neural Scale conjugate gradient
farming yields year after year [bluerivertechnology.com]. Fasal is data- Khan et al., 2019) Networks For Fruit Yield backpropagation (SCG)
driven smart farming for irrigation alerts, weather forecasts, disease and Prediction
pest alerts, record activities and finances, and many more on the list [htt Qiao et al. (2021) ( Crop Yield Platform from SSTNN (Spatial-Spectral-
Qiao et al., 2021) multi-spectral images. Temporal Neural Network),
ps://www.fasal.com]. Table 11 below offers some commercially avail­
combining 3D convolutional
able AI-based smart tools for precision agriculture and farming. and recurrent neural
networks
4. Discussions Xu et al. (2021) Large-scale and small-scale Bayesian regularization BP
(Xu et al., 2021) cotton yield prediction (backpropagation)
Tello and Ko (2021) Spring wheat (triticum The combination of
Agriculture has encountered numerous difficulties, including a lack (Tello and Ko, aestivum) yield a Convolutional Neural
of irrigation infrastructure, climate change, groundwater density, food 2021) prediction process using Networks and Long-Short
scarcity and waste, and so on. The acceptance of various cognitive so­ multispectral images TermMemory (CNN-LSTM)
lutions has a substantial impact on the cultivation outcome. Despite the Wolanin et al. Estimating and Convolutional Neural
(2020) (Wolanin understanding crop yields Networks (CNN)
fact that large-scale research is currently ongoing and several applica­ et al., 2020) with explainable deep
tions are now available, the sector continues to be underserved. When it learning in the Indian
comes to dealing with real-world problems and fixing them with Wheat Belt
autonomous decision-making and predictive solutions, farming is still in Bhojani and Bhatt Wheat crop yield Deep Neural Networks (DNN)
(2020) (Bhojani prediction using new
its infancy.
and Bhatt, 2020) activation functions in
It should go without saying that to diagnose and identify disease neural network
from a leaf, two acts must be completed: Machine learning approaches Fathi et al. (2019) ( Crop Yield Prediction Deep Neural Networks (DNN)
are applied after image processing techniques such as pre-processing Fathi et al., 2019) Using Deep Learning in
segmentation, and others have been used to the image samples. Ac­ Mediterranean Region
Shidnal et al. (2019) Crop yield prediction: two- Convolutional Neural
cording to the most recent research, the Fuzzy K-NN has been shown to (Shidnal et al., tiered machine learning Networks (CNN)
be quite helpful and accurate in detecting nutritional deficiency and 2019) model approach
potassium deficiency in tomato leaves. In terms of disease detection and Nguyen et al. (2019) Spatial-Temporal Multi- Spatial-Temporal Multi-Task
categorization of tomato leaves, SVM and CNN beat K-NN and random (Nguyen et al., Task Learning for Within- Learning
2019) Field Cotton Yield
forest methods (Ranjan and Parida, 2019). The CNN accurately detected
Prediction
bacterial spots, early blight, late blight, leaf mold, septoria spots, spider Long-Short Term Memory
mite, healthy, target spot tomato mosaic virus, and tomato yellow leaf (continued on next page)
curl virus. Furthermore, the PlantVillage dataset was the most widely

20
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

Table 10 (continued ) Table 10 (continued )

Reference Target Model Used Reference Target Model Used

De Alwis et al. Duo Attention with Deep SAR data in Sahibganj

(2019) (Alwis Learning on Tomato Yield district, Jharkhand (India)
et al., 2019) Prediction and Factor
Interpretation Duo
Attention
Jiang et al. (2020) A deep learning approach Long-Short Term Memory Table 11
(Jiang et al., to conflating (LSTM) Commercially Available Artificial Intelligence (AI)-Based Tools for Smart
2020) heterogeneous geospatial Agriculture.
data for corn yield
estimation: A Company Website Products/Service
case study of the US Corn AGEYE ageyetech.com AI-powered platform for indoor
Belt at the county level Technologies farming
Saravi et al. (2019) ( Quantitative model of Deep Neural Networks (DNN) aWhere awhere.com Weather information with
Saravi et al., irrigation effect on maize machine learning algorithms in
2019) yield by deep neural connection with satellites to
network predict the weather, analyze
Kang et al. (2020) ( Comparative assessment of Long-Short Term Memory crop sustainability and
Kang et al., 2020) environmental variables (LSTM) and Convolutional evaluate farms for the presence
and machine learning Neural Networks (CNN) of diseases and pests
algorithms for maize Blue Reiver bluerivertechnology.com Smart farm machines to
yield prediction in the US Technology manage crops at a plant level
Midwest and protect crops from weeds
Zhang et al. (2020) ( Combining Optical, Long-Short Term Memory FarmShots farmshots.com Integrated scouting and
Zhang et al., Fluorescence, Thermal (LSTM) variable rate prescription
2020) Satellite, and platform for farmers based on
Environmental Data to images captured by satellites
Predict County- and drones
Level Maize Yield in China Fasal fasal.co AI-based solutions for the small
Using Machine Learning farmer to provide critical
Approaches parameters using affordable
Wang et al. (2020) ( Combining Multi-Source Deep Neural Networks (DNN) sensors
Wang et al., 2020) Data and Machine Harvest CROO harvestcroorobotics. com Robot system to pick and pack
Learning Approaches to Robotics vegetables
Predict Winter Wheat Yield HelioPas AI heliopas.com irrigation, fight AI-based soil moisture
in mildew, and deal with monitoring system to control
the Conterminous United drought
States Hortau Inc hortau.com Web-based irrigation
Ju et al. (2020) (Ju Machine learning Long-Short Term Memory management service Ibex
et al., 2020) approaches for crop yield (LSTM) Convolutional Neural Automation ibexautomation.co.uk Autonomous agricultural robot
prediction with MODIS Networks (CNN), systems for farmers, including
and weather data Stacked-Sparse AutoEncoder an autonomous precision weed
(SSAE) detection and spraying system
Yalcin (2019) ( An Approximation for A Deep Learning Convolutional PEAT plantix.net Deep Learning-empowered
Yalcin, 2019) Relative Crop Yield Neural Networks (CNN) image recognition application
Estimate from Field Images to identify potential defects
Using and nutrient deficiencies in soil
Shook et al. (2021) ( performance records from LSTM, RNN Root AI root-ai.com AI-based automated and
Shook et al., 2021) Uniform Soybean Tests robotic solutions for indoor
(UST) to dissect and farmers
predict genotype response Trace tracegenomics.com Soil analysis system to provide
in multiple-environments Genomics a sense of soil’s strengths and
Gomez et al. (2021) Wheat yield production PLSR, ANN, RF weaknesses using machine
(Gomez et al., learning
2021) VineVIew vineview.com Highly specialized aerial-based
Aplolo et al. (2020)( Maize DNN spectral sensors and a cloud-
Apolo et al., 2020) based image processing service
Kouadio et al. Artificial intelligence Extreme learning machine, to monitor crop health
(2018) (Kouadio approach for the prediction multiple linear regression,
et al., 2018) of Robusta coffee yield random forest
using soil fertility
used dataset for training in the majority of current research work in the
properties
Bi and Hu (2021) (Bi Applying machine learning Genetic algorithm (GA)- literature related to the categorization of disease and infected tomato
and Hu, 2021) for Crop Yield Production assisted deep learning leaves. Pretrained transfer learning approaches, such as the AlexNet
Taherei Ghazvinei Sugarcane growth Neural networks architecture, maybe a simple alternative for not only achieving high
et al. (2018) ( prediction based on detection accuracy but also saving both money and coding time when it
Ghazvinei et al., meteorological parameters
comes to identifying potato leaves. Among the available classifiers, CNN
2018) using extreme learning
machine and artificial has the best accuracy (98.33%), followed by BPNN, MSVM, ANN, SVM,
neural network and RF methods. Deep learning models like DCNN scored admirably,
Xu et al. (2019) (Xu Design of an integrated Random forest, support while machine learning approaches like MSVM, ANN, SVM, and RF also
et al., 2019) climatic assessment vector machine
did well. For their rice ailment diagnosis, many existing works had
indicator (ICAI) for wheat
production
captured real images. CNN-based models have achieved the best level of
Ranjan & Parida Paddy acreage mapping Linear regression accuracy in detecting and categorising diseases in paddy leaves.
(2019) (Ranjan and yield prediction using The research in smart agriculture and precision farming is seeing an
and Parida, 2019) sentinel-based optical and exponential trend shifting towards deep learning and its various variants

21
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

because of the various novel deep learning applications. In traditional pressing challenges: feeding an additional 2 billion people by 2052,
machine learning techniques, most of the applied features need to be despite climate change disrupting growing seasons, turning arable land
identified by a domain expert to reduce the complexity of the data and into deserts, and flooding once-fertile deltas with seawater. Farmers in
make patterns more visible to learning algorithms to work. The biggest many nations, including India, rely on the monsoon for their crops. They
advantage of deep learning algorithms is that they try to learn high-level rely primarily on weather forecasts from several departments, particu­
features from data in an incremental manner. This eliminates the need larly for rain-fed agriculture.
for domain expertise and hardcore feature extraction. Deep learning AI will help predict weather and other agricultural circumstances
techniques tend to solve the problem end to end, whereas machine such as land quality, groundwater, crop cycle, and pest assault, among
learning techniques need the problem statements to break down into others. These sensors have a lot of potential in agriculture. Data such as
different parts to be solved first, and then their results to be combined at soil quality, weather, and groundwater level, among other things, can be
the final stage. DL can be considered as one of the core technology of AI, derived by agriculture scientists and used to optimize the cultivation
that has pushed AI to a new level, termed “Smarter AI” (Kumar et al., process (Unal, 2020). In order to collect data, AI-enabled sensors can be
2022). Thus, DL modeling is instrumental and is applied to almost all integrated into robotic harvesting equipment. It’s been suggested that
fields and is termed a universal learning method. Because of its capacity AI-based advisories could help enhance productivity by 30%. The most
to process vast amounts of features to build an effective data-driven challenging aspect of farming is crop damage caused by natural di­
model, it is a perfect fit for a large number of data piles. In terms of sasters, such as pest attacks. The majority of the time, farmers lose their
developing and training DL models, it relies on parallelized matrix and crops owing to a lack of sufficient information. In this cyber age, tech­
tensor operations as well as computing gradients and optimization. GPU nology might be beneficial to farmers in protecting their crops from
has become an integral part now of executing any DL-based model. cyber-attacks. Many businesses have adopted this strategy. Such oper­
Deep learning has achieved a lot of notable advances in the field of ations have shown to be helpful in the past, providing motivation to
agricultural disease image recognition because of enlarged chip pro­ develop a system to monitor and safeguard crops. NatureFresh Farms, a
cessing capabilities such as GPU units, cost-effective computer hard­ 20-year-old company that grows veggies on 185 acres between Ontario
ware, and current advancements in machine learning projects (Meshram and Ohio, is developing and researching the technique. Knowing how
et al., 2021). With a larger dataset, speed increment is also required, many tomatoes will be available to sell in the future makes the sales
which shifts the GPU to TPU (Tensor processing unit) and then HPC, team’s job easier and benefits the bottom line directly. According to data
which is high-performance computing/supercomputing, where HPC from prominent universities, there is a significant amount of food waste
takes care of both computational improvements in terms of hardware around the world, which may be addressed with the correct algorithms,
and software that will scale up deep learning. Robust deep learning which will not only save time and money but also lead to long-term
methods do not need a precise design feature. Instead, its process of development.
automatically learning and representing the best features for any task
makes it robust. The deep learning approach is generalized, which 5. Challenges and future scope of data mining in smart
means the same DL method can be applied with different types of agriculture
datasets or in different applications. That method is also termed transfer
learning. The DL method is extremely scalable in terms of data and A wide range of challenges applies to gathering, processing, and
computation. These and other properties of DL make it the right choice using data for agricultural productivity. One of the significant issues that
to be used in the smart agriculture and automated, precise farming farmers have to face to thrive in the age of information is data security
domain (Dokic et al., 2020). and ensuring privacy. Data availability and quality problems are
Deep learning methods, on the other hand, are heavily reliant on frequently found in agricultural information systems. When there is
data, making them prone to the following flaws (Sarker, 2021): more data in real-time, this gets much more complicated. Data efficiency
• In the absence of large-scale labeled training sets, the training and spatial semantic integration are often a struggle for data mining.
process is prone to over-fitting, making it difficult to construct an
optimal model. 5.1. Challenges
• As the complexity of the models increases, the number of param­
eters in the model grows exponentially, limiting their generalizability. The applications of machine learning and deep learning in agricul­
• Each new dataset and task requires the models to be trained from ture are huge, with many challenges.
scratch, which increases the hardware requirements and computational
expense, as well as limits their practical usefulness. 5.1.1. Privacy
Transfer learning allows you to adapt models learned in one domain The agricultural information contains their personal data (identity,
to another, resulting in high-quality model learning and construction geographical location, financial data, entrepreneurial knowledge, etc.).
with fewer data. This approach overcomes the disadvantages of deep Most farmers who have their information made available through digital
learning techniques, such as the need for large amounts of labeled platforms will be unaware of what it says about them (Jin et al., 2020).
training data. Transfer learning is particularly well suited for agricul­ True, farmers are unaware that their personal information is being
tural disease picture recognition when data resources are limited. Ap­ gathered and used, and even more worrisome, what it is being used for.
plications must be more resilient to explore the vast potential of AI in Data mining provides organizations with the opportunity to focus on
agriculture (Sarker, 2021). Only then would it be capable of handling people to accumulate and collect enormous amounts of information
frequent changes in external conditions, facilitating real-time decision about farmers, which could be enough to create and analyze a person­
making, and utilizing a suitable framework/platform for efficiently ality or psychological portrait of the subjects. Even if this data could be
collecting contextual data. Another significant factor is the high expense misused, there is the danger that it will injure his reputation or just get
of many cognitive farming technologies available on the market. To discovered and fall into the wrong hands. This could also make it diffi­
ensure that technology reaches the masses, solutions must become more cult for him to carry out his normal activities. They need to be reassured
affordable. The keys would be more economical if built on an open- that their data will be used to develop creative ideas but not for
source platform, leading to faster acceptance and higher penetration competitive advantage. Data mining can make it difficult for farmers to
among farmers. Farmers will benefit from the technology since it will keep their personal information private. Privacy and confidentiality
help them achieve higher yields and a more consistent seasonal harvest. policies require substantial time and resources to implement. Fig. 11
Artificial intelligence, from detecting pests to predicting which crops below demonstrates the various open challenges in implementing smart
will yield the best returns, can help humanity meet one of its most techniques in the precision agriculture sector.

22
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

Fig. 11. Overview of Open Challenges in the Smart Agriculture.

5.1.2. Available databases and size issue lot of background features. As a result, the system must remove all
It is challenging to obtain leaf images for specific plant infections. extraneous elements from the image so that just the desired segment
Due to this fact, the sizes of the available plant data sets are tiny. Only remains.
minor works have reported thousands of images for research purposes.
Due to the small database size problem, a large portion of the data set is 5.1.5. Image capture conditions
used for the training phase in most of the deep learning methods (Giua The images in all available datasets were recorded in a controlled
et al., 2022). Furthermore, the available database images are collected environment in laboratories, and in other cases, images were generated
in very constrained environmental conditions. We believe that images utilising animation techniques, according to the literature. Consider
must be gathered in real-world situations to make the algorithms more what happens if a farmer or capturing equipment in the field tries to
practical. Efficient image acquisition of leaf images is the need of the catch the same thing at different times of the day. Because of many
hour. If these images are captured in real-time scenarios, such databases variable characteristics such as extra light intensity, dampness, and
would be warmly welcomed in the research community. From this other environmental factors, it is challenging to acquire a similar image
survey, it is observed that many researchers use data source sites like in that circumstance. As a result, photographs of the same leaf must be
Kaggle, Meandly, IEEE Dataport, etc., to get the data to build models. If captured from several perspectives, at various times, and under different
the required data is not available, researchers need to develop their own climatic circumstances.
dataset.
5.1.6. Issues with available feature extraction methods
5.1.3. Quality and precision of data Preprocessing, feature extraction, and segmentation are crucial steps
A lot of data needs to be gathered and transformed into information in constructing a machine learning-based system. The type of data
if we want to succeed in agriculture. Traditional analytical tools have collection plays a role in determining the best preprocessing and seg­
been ineffective against the sheer weight of this new data. Crop man­ mentation strategy. The technique that is best suited for a particular
agement will be vastly improved with the use of data mining analysis. acquisition is usually the one that is used. We’ve seen a lot of variation in
Excellent data quality is needed to get helpful information from the DM the algorithms that have been published so far across different modules.
process (Lia et al., 2020). Agricultural data usually comes from different
databases and models and is thus messy, as well as the presence of many 5.1.7. Difficulties in classification module
missing pieces. Data collected through these systems is missing has lots For a long time, plant disease automation and detection have been a
of problems and inconsistencies. There is significant work that needs to hot topic of research. Researchers claim highly acceptable results
be done before the data can be processed in the data mining process. despite using very few photos for training and testing. Researchers in
Providing data for models can face several obstacles and challenges, this field are looking into a variety of classifiers (Lytos et al., 2020).
such as uncertainty, indefinite persistence, and incapability, and there­ According to the findings, backpropagation neural networks, SVM, and
fore calls for data processing (geographical and temporal). Using more discriminant analysis (especially linear) outperform the competition.
data is inevitable. Accordingly, both syntax and semantics must be After that, Nave Bayes, random forest, nearest neighbor, and multilayer
harmonized to guarantee data portability in a (definitions). Better data perceptron are used. However, newly established optimized deep neural
handling allows all kinds of analysis and product development. Reli­ networks significantly improve state-of-the-art outcomes. Deep con­
ability is a primary concern for IoT devices in terms of data transmission. volutional neural networks can assist enhance the outcomes of enor­
Devices need to gather and transfer reliable data to make appropriate mous data sets if they are used more effectively.
decisions when necessary. False reading will significantly reduce system
reliability. 5.1.8. Concerns about devices
Device standardization is necessary to ensure that technology may be
5.1.4. Unnecessary noise and background in crop‘s used in a wide range of applications. However, there are no data-
The technique of image segmentation is utilized to retrieve the processing standards formats (Klerkxa et al., 2019). And the
infected section from an image (Fraser, 2022). When the image back­ misreading of the mismatched code can result in different outcomes.
ground comprises colors other than white and black or other features Machine standardization can help to resolve system, application,
such as plants, leaves, dirt, grass, and so on, extracting the contaminated equipment, and product interoperability difficulties. Furthermore, the
leaf segment from an image becomes difficult. If farmers want to di­ development of the 5G network has made communication between de­
agnose crop illnesses in real-time from the fields, the photo may have a vices and servers 100 times faster than it was with 4G. 5G is a suitable

23
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

technology for sending data from remote sensors because it can trans­ 4) Use “Transfer Learning” approaches to shorten the time it takes to
port substantially more data. As a result, the adoption of 5G as a new train a model.
communication network is required to meet the needs of more secure 5) It is advised that the model be deployed in a real-time application
users and faster data transfer rates. One of the most serious issues is a to assist the intended users.
lack of interoperability.
5.2. Future scope
5.1.9. Spatial data importance
The overall goal of smart agriculture is to manage environmental The following paragraphs describe some elements that may help to
impacts and maximize profit. Conventional data mining approaches are improve and enhance the current state-of-the-art and provide re­
typically designed for relational databases but are not entirely appli­ searchers with some prospective ideas for future research:
cable to geographically dispersed data. For smart farming, new data
mining methods are required to consider spatial and temporal correla­ 5.2.1. Disease stage identification
tions in the data (Pivoto et al., 2018). One of the most critical aspects of plant disease identification is
disease stage identification. There are various stages to each disease. The
5.1.10. Incorporation of agricultural domain knowledge in data mining majority of the researchers concentrated their efforts on disease type
Agriculture is an interdisciplinary field that encompasses a variety of identification, but none focused on disease stage identification.
subjects, including environmental sciences, agronomy, soils, etc. The Furthermore, such systems must be able to recommend specific mea­
data we use in a data mining project can come from various sources. Due sures based on the stage of the disease. Disease forecasting will aid ag­
to the sensor and high data throughput, new challenges arise, particu­ riculturists in taking the appropriate activities and precautions to limit
larly operable semantics: figuring out how to keep the meaning of data the percentage of damage.
and correctly represent it over time? The problem of integrating domain
knowledge is one of the most challenging problems in data mining 5.2.2. Quantification of a disease
(Jerhamre et al., 2022). It can be thought of as a form of fusion of The quantification of a specific disease is another intriguing subject
agriculture, making agricultural domain knowledge compatible with to investigate. Even though significant research has been done in this
data mining research. area, only a few researchers have determined the extent of the disease’s
impact. They can be quite beneficial since corrective steps can be
5.1.11. Scalability of data mining algorithms implemented based on the severity of the sickness. This type of quan­
Smart agriculture is responsible for generating enormous amounts of tification will discover the fraction of a culture that is infected with a
data due to all the various gadgets in use. Heterogeneous data, in disease. Because the number of pesticides can be managed, this study
particular, is being produced by remote sensors. They will probably have perspective is critical. Typically, farmers use chemicals to treat diseases
a large data set, which will allow them to demonstrate the real re­ without first analyzing or quantifying them. Such behavior is exceed­
lationships. Data mining algorithms, however, face the constant ingly hazardous to human health. Developing a practical image pro­
requirement of needing to make countless decisions in the agricultural cessing application will aid in determining whether or not specific
sector: Can they handle the volume of data quickly. Data mining algo­ chemicals are required.
rithms need to be scalable with exponential search problems searching
through these large data sets. The development of parallel and distrib­ 5.2.3. Mobile and online applications
uted algorithms is critical in agricultural data mining. Several solutions for disease identification applications have been
presented in the literature. However, just a few of the portals and mobile
5.1.12. Drones apps are publicly published and accessible via the internet (Jianga et al.,
Because the drone can only fly for an hour or less, the flight line path 2020). Assess Software and Leaf Doctor are two of these applications,
must be set with the overlap between the flight lines in mind. Drones are both of which are free to use. These programs, on the other hand,
expensive, especially those with high-resolution cameras and thermal function with photographs that have a flat, black background. As a
cameras, as well as good software, hardware tools, and devices. Drone result, such online systems and apps are critical for identifying plant
operations require licenses, which can be challenging in many nations, diseases. The availability of technologies like this will aid farmers in
as well as a height not exceeding 400 feet. Climate conditions have an identifying a specific illness. Such software can be used to generate
impact on drone operation (Terence and Purushothaman, 2020). Wind analysis reports, which can then be sent to a disease expert for advice.
speed and wetness affect drone operation; thus, the weather should be
considered before beginning work. 5.2.4. Exploring transfer learning to increase data size
There are a few extra challenges to be aware of: Similarly, current tendencies in CV development, which are rapidly
1) Identifying the problem and understanding the business need. going toward DL approaches, are not very promising for plant disease
2) Gaining a better understanding of the consumer and how they diagnosis. Given the difficulty of the data, particularly at the training
engage with technology. stage, transfer learning is the best choice to consider. A heterogeneous
3) An application that is easy to use. domain strategy can be used to investigate knowledge transmission.
4) Model performance in real-world circumstances. LSTMs, optical flow frames, temporal pooling, and 3D convolution are
5) Model power consumption and battery restrictions for running the some of the keywords that might be explored in terms of automatic plant
model on devices. disease identification. Last but not least, better and more thoroughly
6) Camera configurations at the user end for computer vision models. constructed methodologies are required for further research in this area.
Following are a few recommendations based on the results of this in- The case of data augmentation, for example, might be examined further.
depth study to make the implementation process more efficient, accu­
rate, seamless, and deployable. 5.2.5. Incorporating new technologies in agriculture
1) Concentrate on developing a machine learning model to address a Agri-food supply chain traceability requires more attention and has
specific problem, such as classification or recommendation. significant IoT potential. Meanwhile, the most commonly utilised IoT
2) Create your dataset for training the model and make it available to communication technologies in these publications are LoRa, ZigBee, and
other researchers via open platforms such as Kaggle, Meandly, IEEE WiFi, while emerging high-speed communication technologies such as
Dataport, and so on. 5G and NB-IoT are expected to be widely used to enhance agricultural
3) Use publicly available datasets for model testing and validation. production modernization and intelligence (Cicioglu and Çalhan, 2021).

24
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

With the advancement of current technology, there is a lot of room for Declaration of Competing Interest
IoT-based agriculture to develop new and effective solutions. Low-cost
systems with advantages such as autonomous operation, low mainte­ The authors declare that they have no known competing financial
nance, energy efficiency, ease of use, and sturdy architecture are interests or personal relationships that could have appeared to influence
particularly in demand. the work reported in this paper.

6. Conclusions References

The value of artificial intelligence is widely recognized across a wide Abdullah, N., Durani, N., Shari, M.F.B., Siong, K.S., Hau, V.K.W., Siong, W.N., Ahmad, I.
R.K.A., 2021. Towards Smart Agriculture Monitoring Using Fuzzy Systems 9,
range of industries, including agriculture. A segment of technologies in 4097–4111.
the agricultural sector is becoming crucial to business survival (Liu et al., Abioye, E.A., Abidin, M.S.Z., Mahmud, M.S.A., Buyamin, S., Ishak, M.H.I., Rahman, M.K.
2022). Data lies at the heart of farming decisions, and the potential is I.A., 2020. A review on monitoring and advanced control strategies for precision
irrigation. Comput. Electron. Agric. 173 (105441), 1–37.
enormous. More importantly, ML is expected to be a source of greater Abouzahir, S., Sadik, M., Sabir, E., 2018. In: Enhanced Approach for Weeds Species
resource efficiency in sustainable resource utilization and an enabler of Detection Using Machine Vision. Optimization and Computer Science (ICECOCS),
massive environmental benefits. This technology can make a significant Kenitra, Morocco, pp. 1–6.
Agrawal, N., Singhai, J., Agarwal, D.K., 2017. “Grape leaf disease detection and
difference in agriculture if stakeholders in the process see it, have a classification using multi-class support vector machine. In: In International
different attitude towards it, and funding can be found. The agricultural Conference on Recent Innovations in Signal Processing and Embedded Systems
industry has a number of concerns, including a lack of appropriate (RISE), pp. 14–23.
Ahmed, H., Juraimi, A.S., Hamdani, S.M., 2016. Introduction to robotics agriculture in
irrigation systems, weeds, crop yields, plant monitoring issues owing to
pest control: a review. Pertanika Journal of Scholarly Research Reviews 2 (2), 80–93.
crop height, and extreme weather conditions. However, with the help of Akhter, R., Sofi, S.A., 2021. Precision agriculture using IoT data analytics and machine
technology, performance may be improved, and thus these issues can be learning. Journal of King Saud University – Computer and Information Sciences xxx,
resolved. It can be enhanced with AI-driven techniques such as remote 1–27.
Alam, M., Alam, M.S., Roman, M., Tufail, M., Khan, M.U., Khan, M.T., 2020. Real-Time
sensors for detecting soil moisture content and GPS-assisted automated Machine-Learning Based Crop/Weed Detection and Classification for Variable-Rate
irrigation (Akhter and Sofi, 2021). Farmers’ difficulty was that precision Spraying in Precision Agriculture. In Proceedings of the 7th International Conference
weeding techniques were able to offset the high amount of crops lost on Electrical and Electronics Engineering (ICEEE).
Albanese, A., Nardello, M., Brunelli, D., 2021. Automated Pest Detection With DNN on
during the weeding procedure. The self-driving robots not only increase the Edge for Precision Agriculture”. IEEE J. Emerging Sel. Top. Circuits Syst. 11 (3),
productivity but also cut the use of unneeded pesticides and herbicides. 458–467.
Aside from that, farmers may use drones to successfully spray pesticides Ali, H., Lali, M.I., Nawaz, M.Z., Sharif, M., Saleem, B.A., 2017. Symptom-based
automated detection of citrus diseases using color histogram and textural
and herbicides on their farms, and plant monitoring is no longer a hassle descriptors. Comput. Electron. Agric. 138, 92–104.
(Debauche et al., 2021). For starters, in agriculture difficulties, man- Aliac, C.J.G., Maravillas, E., 2018. IOT Hydroponics Management System. In: IEEE 10th
made brainpower can be used to understand resource and job short­ International Conference on Humanoid. Nanotechnology, Information Technology,
Communication and Control, Environment and Management (HNICEM), Baguio
ages. In traditional tactics, a significant amount of labor was necessary City, Philippines, pp. 1–5.
to get agricultural parameters such as plant height, soil texture, and Alwis, S.D., Zhang, Y., Na, M., Li, G., 2019. Duo attention with deep learning on tomato
content, which necessitated manual testing, which was time-consuming. yield prediction and factor interpretation. In: Pacific Rim International Conference
on Artificial Intelligence. Springer, Cham, pp. 704–715.
Quick and non-damaging high throughput phenotyping would be
Andrei, B.B., Torres, B., Atslands, R., Rocha, R., Ticiana, L., Silva, C.D., Souza, J.N.,
possible with the help of the various systems investigated, with the Gondim, R.S., 2020. Multilevel data fusion for the internet of things in smart
added benefit of flexible and favorable activity, on-demand access to agriculture. Comput. Electron. Agric. 171 (105309), 1–16.
information, and spatial goals. AO (2016). AQUASTAT database. Retrieved from http://www.fao.org/nr/water/
aquastat/data/query/index.html?lang=en. [Last visited 18 July 2021].
In this paper, we have discussed how ML in general and DL, in OEA Apolo, MJ Guanter, G Egea, P Raja, and PM Ruiz (2020) Deep learning techniques
particular, have helped identify the diseases in plants, optimal irriga­ for estimation of the yield and size of citrus fruits using a UAV. Eur J Agron 115
tion, pesticide control, crop phenotyping, and improved crop yield. If (126030): 1-34.data—a machine learning approach. IEEE J. Sel. Top. Appl. Earth
Obs. Remote Sens. 10 (7): 3254–3264.
the conditions are not correctly identified, they affect the crop yield and Araby, A.A., 2019. Smart IoT Monitoring System for Agriculture with Predictive
ultimately result in long-term issues, such as global warming and even Analysis. In: 8th International Conference on Modern Circuits and Systems
famine. It can be deduced from the studies that with precision farming, Technologies (MOCAST). Thessaloniki, Greece, pp. 1–4.
Arora, J., Agrawal, U., 2020. Classification of Maize leaf diseases from healthy leaves
more pragmatic farming may be carried out utilizing scientific meth­ using Deep Forest. Journal of Artificial Intelligence and Systems 2 (1), 14–26.
odologies such as remote sensing, GPS, data analytics, and so on, Bacco, M., Barsocchi, P., Ferro, E., Gotta, A., Ruggeri, M., 2019. The Digitisation of
resulting in increased agricultural productivity and reduced potential Agriculture: a Survey of Research Activities on Smart Farming. Array 3 (100009),
1–11.
environmental damage. In addition, the abnormality can be recognized M Bacco, A Berton, E Ferro, C Gennaro, A Gotta, S Matteoli, F Paonessa, M Ruggeri, G
in the plant at an early stage using image recognition software, artificial Virone, and A Zanella (2018), “Smart Farming: Opportunities, Challenges and
neural networks, and a variety of other technologies. Because there are Technology Enablers”, In: IoT Vertical and Topical Summit on Agriculture-Tuscany
(IOT Tuscany), IEEE 685: 1-6.
so many different types of crop diseases and so few datasets, fine-tuning
Backhaus, A., Bollenbeck, F., Seiffert, U., 2011. Robust classification of the nutrition state
models that have been pre-trained on large-scale public datasets like in crop plants by hyperspectral imaging and artificial neural networks. In 3rd
PlantVillage and ImageNet to be able to do image recognition based on Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote
few samples is desirable (Idoje et al., 2021). It can be demonstrated that Sensing (WHISPERS).
Bah, M.D., Hafiane, A., 2018. Canals R: Deep learning with unsupervised data labeling
these methods can reach extraordinary levels of accuracy when per­ for weed detection in line crops in UAV images. Remote Sens 10, 1690.
forming plant disease image detection when they are based on large- Bai, G., Ge, Y., Hussain, W., Baenziger, P.S., Graef, G., 2016. A multi-sensor system for
scale open-source datasets and pre-trained models. However, further high throughput field phenotyping in soybean and wheat breeding. Comput.
Electron. Agric. 128, 181–192.
research revealed that the scale of the plant image collection used for Bakhshipour, A., Jafari, A., 2018. Evaluation of support vector machine and artificial
modeling, as well as the context in which the photos were obtained, neural networks in weed detection using shape features. Comput. Electron. Agric.
might have a significant impact on recognition accuracy. The recogni­ 145, 153–160.
Barbedo, J.G.A., 2019. Plant disease identification from individual lesions and spots
tion accuracy of systems built using the PlantVillage dataset is more using deep learning. Biosyst. Eng. 180, 96–107.
remarkable since most of the photos in the dataset were captured with a Barbedo, J.G., Arnal, A., 2014. An automatic method to detect and measure leaf disease
simple background or in a laboratory environment with less symptoms using digital image processing. Plant Dis. 98 (12), 1709–1716.
Barrero, O., Perdomo, S.A., 2018. RGB and multispectral UAV image fusion for
interference. Gramineae weed detection in rice fields. Precis. Agric. 19 (5), 809–822.

25
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

Bedi, P., Gole, P., 2021. Plant disease detection using hybrid model based on Fathi, M.T., Ezziyyani, M., Ezziyyani, M., Mamoune, S., 2019. Crop yield prediction
convolutional autoencoder and convolution neural network. Artificial Intelligence in using deep learning in Mediterranean Region 20, 106–114.
Agriculture 5, 90–101. Feng, L., Chen, S., Zhang, C., Zhang, Y., He, Y., 2021. A comprehensive review on recent
Bhatnagar, V., Chandra, R., 2020. IoT-based soil health monitoring and recommendation applications of unmanned aerial vehicle remote sensing with various sensors for
system. Internet of Things and Analytics for Agriculture. 2, 1–21. high-throughput plant phenotyping. Comput. Electron. Agric. 182, 1–32.
Bhojani, S.H., Bhatt, N., 2020. Wheat crop yield prediction using new activation Ferentinos, K.P., 2018. Deep learning models for plant disease detection and diagnosis.
functions in neural network. Neural Comput. Appl, 1–11. Comput. Electron. Agric. 145, 311–318.
Bi, L., Hu, G., 2021. A genetic algorithm-assisted deep learning approach for crop yield J Foley (2019) A Five-Step Plan to Feed the World. Retrieved from https://www.
prediction. Soft. Comput. 25, 10617–10628. nationalgeographic.com/foodfeatures/feeding-9-billion/. [Last visited 10 November
Boniecki, P., Koszela, K., Piekarska-Boniecka, H., Weres, J., Zaborowicz, M., Kujawa, S., 2021].
Majewski, A., Raba, B., 2015. Neural identification of selected apple pests. Comput. Food and Agriculture Organization of the United Nations - FAO. Strengthened global
Electron. Agric. 110, 9–16. partnerships needed to end hunger and malnutrition. 2019. http://www.fa o.org/
Bosilj, P., Duckett, T., Cielniak, G., 2018. Connected attribute morphology for unified news/story/en/item/1194310/icode/. [Last visited 14 August 2021].
vegetation segmentation and classification in precision agriculture. Comput. Ind. 98, OECD. Food, A. O. Of the United Nations, OECD-FAO agricultural outlook 2020- 2029.
226–240. 2020. 10.1787/1112c23b-en. https://www.oecd-ilibrary.or g/content/publication/
Brinkhoff, J., Vardanega, J., Robson, A.J., 2020. Land Cover Classification of Nine 1112c23b-en. [Last visited 10 August 2021].
Perennial Crops Using Sentinel-1 and-2 Data. Remote Sens. 12, 96–129. Forsythe, M.M., 2021. Crop Yield Prediction Using Deep Neural Networks and LSTM.
Bu, F., Wang, X., 2019. A smart agriculture IoT system based on deep reinforcement Agriculture, Case Studies & Projects, Machine Learning, Remote Sensing 1, 1–18.
learning. Future Generation Computer Systems 99, 500–507. High-Level Expert Forum - Global Agricultu re Towards 2050. (2009, October 12).
Butera, L., Ferrante, A., Jermini, M., Prevostini, M., Alippi, C., 2021. Precise Agriculture: Retrieved from http://www.fao.org/fileadmin/templates/wsfs/docs/Issues_papers/
Effective Deep Learning Strategies to Detect Pest Insects. IEEE/CAA Journal of HLEF2050_Global_Agriculture.pdf. [Last visited 08 November 2021].
Automatics Sinica 9 (2), 246–258. Fraser, A., 2022. ‘You can’t eat data’?: Moving beyond the misconfigured innovations of
Castro, D., New, J., 2016. The promise of artificial intelligence. Center for Data smart farming. Journal of Rural Studies 91, 200–207.
Innovation 1–48. Fuentes, A., Yoon, S., Kim, S., Park, D., 2017. A robust deep-learning-based detector for
Castro, D., Torres-Sanchez, A.I., Peña, J., Jiménez-Brenes, J.M., Csillik, F.M., real-time tomato plant diseases and pests recognition. Sensors 17 (9), 2022–2051.
Granados, O.L., 2018. An automatic random forest-OBIA algorithm for early weed Gao, J., Nuyttens, D., Lootens, P., He, Y., Pieters, J.G., 2018. Recognising weeds in a
mapping between and within crop rows using UAV imagery. Remote Sens. 10, maize crop using a random forest machine-learning algorithm and near-infrared
285–3015. snapshot mosaic hyperspectral imagery. Biosyst. Eng. 170, 39–50.
Chabot, D., Dillon, C., Shemrock, A., Weissflog, N., Sager, E.P., 2018. An object-based García, L., Parra, L., Jimenez, J.M., Lloret, J., Lorenz, P., 2020. IoT-Based smart irrigation
image analysis workflow for monitoring shallow-water aquatic vegetation in systems: an overview on the recent trends on sensors and iot systems for irrigation in
multispectral drone imagery. ISPRS Int. J. Geo-Inf. 7, 294–318. precision agriculture. Sensors 20, 1042–1068.
Chen, W., 2019. AgriTalk: IoT for precision soil farming of turmeric cultivation. IEEE Gargade, A., Khandekar, S., 2021. Custard Apple Leaf Parameter Analysis, Leaf Diseases,
Internet Things J. 6 (3), 5209–5223. and Nutritional Deficiencies Detection Using Machine Learning. In: In Advances in
Chen, H., Chen, A., Xu, L., Xie, H., Qiao, H., Lin, Q., Cai, K., 2020. A deep learning CNN Signal and Data Processing. Lecture Notes in Electrical Engineering, p. 703.
architecture applied in smart near-infrared analysis of water pollution for Gavhale, R.K., Gawande, U., Hajari, K.O., 2014. Unhealthy region of citrus leaf detection
agricultural irrigation resources. Agric. Water Manag. 240 (106303), 1–8. using image processing techniques. In: In International Conference for Convergence
Chen, Y., Lee, W.S., Gan, H., Peres, N., Fraisse, C., Zhang, Y., He, Y., 2019. Strawberry for Technology IEEE, pp. 76–86.
yield prediction based on a deep neural network using high-resolution aerial Ghazvinei, P.T., Darvishi, H.H., Mosavi, A., Yusof, K.W., Alizamir, M., Shamshirband, S.,
orthoimages. Remote Sens. 11 (13), 1584–1617. Chau, K., 2018. Sugarcane growth prediction based on meteorological parameters
Chohan, M., Khan, A., Katper, S., Mahar, M., 2020. Plant disease detection using deep using extreme learning machine and artificial neural network. Eng. Appl. Comput.
learning. Int. J. Recent Technol. Eng. 9 (1), 909–914. Fluid Mech. 12 (1), 738–749.
Chu, Z., Yu, J., 2020. An end-to-end model for rice yield prediction using deep learning Giua, C., Materia, V.C., Camanzi, L., 2022. Smart farming technologies adoption: Which
fusion. Comput. Electron. Agric, 174–181. factors play a role in the digital transition. Technol. Soc. 68 (101869), 1–11.
Chung, C.L., Huang, K.J., Chen, S.Y., Lai, M.H., Chen, Y.C., Kuo, Y.F., 2016. Detecting Glaroudis, D., Iossifides, A., Chatzimisios, P., 2020. Survey, comparison and research
Bakanae disease in rice seedlings by machine vision. Comput. Electron. Agric. 121, challenges of IoT application protocols for smart farming. Comput. Netw. 107037
404–411. (168), 183.
Cicioglu, M., Çalhan, A., 2021. Smart agriculture with internet of things in cornfields. Golhani, K., Balasundram, S.K., Vadamalai, G., Pradhan, B., 2018. A review of neural
Comput. Electr. Eng. 90 (106982), 1–11. networks in plant disease detection using hyperspectral data. Information Processing
Cicioglu, M., Çalhan, A., 2021. Smart agriculture with the internet of things in cornfields. in Agriculture 5, 354–371.
Comput. Electr. Eng. 90 (106982), 1–11. Gomez, D., Salvador, P., Sanz, J., Casanova, J.L., 2021. Modelling wheat yield with
Coppens, F., Wuyts, N., Inze, D., Dhondt, S., 2017. Unlocking the potential of plant antecedent information, satellite and climate data using machine learning methods
phenotyping data through integration and data-driven approaches. Current Opinion in Mexico. Agric. For. Meteorol. 300 (108317), 1–21.
in Systems Biology 4, 58–63. Goswami, V., Singh, P., Dwivedi, P., Chauhan, S., 2020. Soil health monitoring system.
Cruz, A.C., Luvisi, A., De Bellis, L., Ampatzidis, Y., 2017. X-FIDO: an effective application Int. J. Res. Appl. Sci. Eng. Technol. 8 (5), 1536–1540.
for detecting olive quick decline syndrome with deep learning and data fusion. Gulve, P.P., Tambe, S.S., Pandey, M.A., Kanse, S.S., 2015. Leaf disease detection of the
Front. Plant Sci. 8, 1–12. cotton plant using image processing techniques. IOSR Journal of Electronics and
Cruza, A., Ampatzidisb, Y., Pierro, R., Materazzi, A., Panattoni, A., Bellis, L.D., Luvisi, A., Communication Engineering 41, 50–54.
2019. Detection of grapevine yellows symptoms in Vitis vinifera L. with artificial Guo, D., Juan, J., Chang, L., Zhang, J., Huang, D., 2017. Discrimination of plant root
intelligence. Comput. Electron. Agric. 157, 63–76. zone water status in greenhouse production based on phenotyping and machine
Dasig, D.D.J., 2020. Implementing IoT and wireless sensor networks for precision learning techniques. Sci. Rep. 7 (1), 1–23.
agriculture. Internet of Things and Analytics for Agriculture. 2, 23–44. Guo, Y., Zhang, J., Yin, C., Hu, X., Zou, Y., Xue, Z., Wang, W., 2020. Plant Disease
Debauche, O., Trani, J.P., Mahmoudi, S., Manneback, P., Bindelle, J., Mahmoudi, S.A., Identification Based on Deep Learning Algorithm in Smart Farming. Discrete
Guttadauria, A., Lebeau, F., 2021. Data management and internet of things : A Dynamics in Nature and Society 2479172, 1–11.
methodological review in smart farming. Internet of Things 14 (100378), 1–23. Gupta, D., Sharma, P., Choudhary, K., Gupta, K., Chawla, R., Khanna, A.,
Dee, H., French, A., 2015. From image processing to computer vision: plant imaging Albuquerque, V.H.C.D., 2020. Artificial plant optimization algorithm to detect
grows up. Funct. Plant Biol. 42 (5), 1–19. infected leaves using machine learning. Expert Syst. e12501.
Deng, F., Zuo, P., Wen, K., Wu, X., 2020. Novel soil environment monitoring system Habib, M.T., Majumder, A., Jakaria, A.Z.M., Akter, M., Uddin, M.S., Ahmed, F., 2018.
based on RFID sensor and LoRa. Comput. Electron. Agric. 169 (105169), 1–25. Machine vision-based papaya disease recognition. Journal of King Saud University -
K Dokic, L Blaskovic, and D Mandusic (2020) From machine learning to deep learning in Computer and Information Sciences 32 (3), 300–309.
agriculture – the quantitative review of trends, IOP Conf. Series: Earth and Han, L., Srocke, F., Masek, O., Smith, D.L., Lafond, J.A., Allaire, S., 2020. A graphical-
Environmental Science 614 (2020) 012138, 1-15. user-interface application for multifractal analysis of soil and plant structures.
Durmus, H., Günes, E.O., Kırcı, M., 2017. Disease detection on the leaves of the tomato Comput. Electron. Agric. 174 (105454), 1–29.
plants by using deep learning. In: In 6th International Conference on Agro- Haque, F.F., Abdelgawad, A., Yanambaka, V.P., Yelamarthi, K., 2020. Crop Yield
Geoinformatics, pp. 1–5. Prediction Using Deep Neural Network. In: In IEEE 6th World Forum on Internet of
Elavarasan, D., Vincent, D.R., Sharma, V., Zomaya, A.Y., Srinivasan, K., 2018. Things (WF-IoT), p. 1:12..
Forecasting yield by integrating agrarian factors and machine learning models: a Hassija, V., Chamola, V., Saxena, V., Jain, D., Goyal, P., Sikdar, B., 2019. A survey on IoT
survey. Comput. Electron. Agric. 155, 257–282. security: application areas, security threats, and solution architectures. IEEE Access
Elavarasan, D., Vincent, P.D., 2020. Crop yield prediction using deep reinforcement 1, 2169–2193.
learning model for sustainable agrarian applications. IEEE Access 8, 86886–88690. Hiary, H.A., Ahmad, S.B., Reyalat, M., Braik, M., Rahamneh, Z.A.L., 2011. Fast and
Elavarasan, D., Vincent, D.R.P.M., 2021. Fuzzy deep learning-based crop yield prediction accurate detection and classification of plant diseases. Inter. J. Com. Appl. 17 (1),
model for sustainable agronomical frameworks. Neural Comput & Applic 33, 31–38.
13205–13224. Hossain, E., Hossain, M.F., Rahaman, M.A., 2019. A color and texture-based approach for
Etienne, A., Saraswat, D., 2019. “Machine learning approaches to automate weed the detection and classification of plant leaf disease using KNN classifier. In:
detection by UAV based sensors”, In Autonomous Air and Ground Sensing Systems International Conference on Electrical. Computer and Communication Engineering
for Agricultural Optimization and Phenotyping IV. International Society for Optics (ECCE), IEEE, pp. 1–6.
and Photonics: Bellingham, WA, USA 11008, 110080–110087.

26
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

Huang, S., Fan, X., Sun, L., Shen, Y., Suo, X., 2019. Research on classification method of Kouadio, L., Deo, R.C., Byrareddy, V., Adamowski, J.F., Mushtaq, S., Nguyen, V.P., 2018.
maize seed defect based on machine vision. J Sens 2716975, 1–31. Artificial intelligence approach for the prediction of Robusta coffee yield using soil
Idoje, G., Dagiuklas, T., Iqbal, M., 2021. Survey for smart farming technologies: fertility properties. Comput. Electron. Agric. 155, 324–338.
Challenges and issues. Comput. Electr. Eng. 92 (107104), 1–14. V Kour, and S Arora (2018), “Fruit Disease Detection Using Rule-Based Classification”, In
Janarthan, S., Thuseethan, S., Rajasegarar, S., Lyu, Q., Zheng, Y., Yearwood, J., 2020. Proceedings of Smart Innovations in Communication and Computational Sciences,
Deep Metric Learning Based Citrus Disease Classification With Sparse Data. IEEE Advances in Intelligent Systems and Computing (ICSICCS-2018): 295–312.
Access 8, 162588–162600. Kumar, S., Durai, S., Shamili, M.D., 2022. Smart farming using Machine Learning and
Jeong, S., Ko, J., Yeom, J.M., 2022. Predicting rice yield at pixel scale through synthetic Deep Learning techniques. Decision Analytics Journal 3 (100041), 1–19.
use of crop and deep learning models with satellite data in South and North Korea. Lamas, P.F., Echarri, C.M., Azpilicueta, L., Iturri, L.P., Falcone, F., Carames, F., 2020.
Sci. Total Environ. 802 (149726). Design and empirical validation of a LoRaWAN IoT smart irrigation system. In Proc
Jerhamre, E., Carlberg, C.J.C., Zoest, V., 2022. Exploring the susceptibility of smart AMIA Annu Fall Symp 42, 62.
farming: Identified opportunities and challenges. Smart Agricultural Technology 2 Lamba, Monika, Gigras, Yogita, Dhull, Anuradha, 2021. Classification of plant diseases
(100026), 1–8. using machine and deep learning. Open Computer Science 11, 491–508.
Ji, M., Zhang, L., Wu, Q., 2020. Automatic grape leaf diseases identification via Lee, S., Jeong, Y., Son, S., Lee, B., 2019. A self-predictable crop yield platform (SCYP)
UnitedModel based on multiple convolutional neural networks. Information based on crop diseases using deep learning. Sustainability 11 (13), 3637–3659.
Processing in Agriculture 7 (3), 418–426. Li, B., Lecourt, J., Bishop, G., 2018. Advance in non-destructive early assessment of fruit
Jiang, B., He, J., Yang, S., Fu, H., Li, T., Song, H., He, D., 2019. Fusion of machine vision ripeness towards defining optimal time of harvest and yield prediction—a review.
technology and AlexNet-CNNs deep learning network for the detection of Plants 7 (1), 1–15.
postharvest apple pesticide residues. Artif. Intell. Agric. 1, 1–8. Li, Y., Qian, M., Liu, P., Cai, Q., Li, X., Guo, J., Yan, H., Yu, F., Yuan, K., Yu, J., 2019. The
Jiang, H., Hu, H., Zhong, R., Xu, J., Xu, J., Huang, J., Lin, T., 2020. A deep learning recognition of rice images by UAV based on capsule network. Cluster Comput. 22
approach to conflating heterogeneous geospatial data for corn yield estimation: a (4), 9515–9524.
case study of the US Corn Belt at the county level. Glob. Change Biol. 26 (3), Li, L., Zhang, S., Wang, B., 2021. Plant Disease Detection and Classification by Deep
1754–1766. Learning—A Review. IEEE Access 9, 56683–56698. https://doi.org/10.1109/
Jiang, H., Xiaoru, L., Safara, F., 2021. IoT based Agriculture: Deep Learning in Detecting ACCESS.2021.3069646.
Apple Fruit Diseases. Microprocess. Microsyst. 1–23. Lia, W., Clarka, B., Taylora, J.A., Kendalla, H., Jonesa, G., Lid, Z., Jina, S., Zhaoe, C.,
Jianga, H., Zhanga, C., Qiaob, Y., Zhangc, Z., Zhanga, W., 2020. Changqing Song CNN Yange, G., Shuaif, C., Chengf, X., Cheng, J., Yange, H., Frewer, L.J., 2020. A hybrid
feature-based graph convolutional network for weed and crop recognition in smart modeling approach to understanding adoption of precision agriculture technologies
farming. Comput. Electron. Agric. 174 (105450), 1–11. in Chinese cropping systems. Comput. Electron. Agric. 172 (105305), 1–12.
Jin, X.B., Yang, N.X., Wang, X.Y., Bai, Y.T., Su, T.L., Kong, J.L., 2020. Hybrid Deep Liakos, K.G., Busato, P., Moshou, D., Pearson, S., Bochtis, D., 2018. Machine learning in
Learning Predictor for Smart Agriculture Sensing Based on Empirical Mode agriculture: a review. Sensors 8, 1–21.
Decomposition and Gated Recurrent Unit Group Model. Sensors 20 (5), 1334–1359. Lima, M.C.F., Leandro, M.E.D.A., Valero, C., Coronel, L.C.P., Bazzo, C.O.G., 2020.
S Ju, H Lim, and J Heo (2020), “Machine learning approaches for crop yield prediction Automatic detection and monitoring of insect pests—A review. Agricult. 10 (5),
with MODIS and weather data”, In 40th Asian Conference on Remote Sensing: 161–179.
Progress of Remote Sensing Technology for Smart Future, ACRS 2019. Liu, Y., Noguchi, N., Liang, L., 2019. Development of a positioning system using UAV-
Ju, S., Lim, H., Heo, J., 2020. Machine learning approaches for crop yield prediction with based computer vision for airboat navigation in paddy field. Comput. Electron.
MODIS and weather data. In 40th Asian Conference on Remote Sensing: Progress of Agric. 162, 126–133.
Remote Sensing Technology for Smart Future. Liu, Y., Wang, J., Shi, Y., He, Z., Liu, F., Kong, W., He, Yong, 2022. Unmanned airboat
Kaburlasos, V.G., Spais, V., Petridis, V., Petrou, L., Kazarlis, S., Maslaris, N., 2020. technology and applications in environment and agriculture. Comput. Electron.
Intelligent clustering techniques for prediction of sugar production. Math. Comput. Agric. 197, 106920.
Simul 60, 159–168. Liu, B., Zhang, Y., He, D., Li, Y., 2018. Identification of apple leaf diseases based on deep
Kamienski, C., Soininen, J., Taumberger, M., Dantas, R., Toscano, A., Cinotti, T.S., convolutional neural networks. Symmetry (Basel) 10, 11–32.
Maia, R.F., Neto, A.T., 2019. Smart Water Management Platform: IoT-Based P Lottes, M Hoeferlin, S Sander, M Müter, P Schulze, LC Stachniss (2016)”, An effective
Precision Irrigation for Agriculture. Sensors 19 (2), 1–20. classification system for separating sugar beets and weeds for precision farming
Kang, Y., Ozdogan, M., Zhu, X., Ye, Z., Hain, C.R., Anderson, M.C., 2020. Comparative applications”, In 2016 IEEE International Conference on Robotics and Automation
assessment of environmental variables and machine learning algorithms for maize (ICRA), 5157–5163.
yield prediction in the US Midwest. Environ. Res. Lett, 1–23. Lytos, A., Lagkas, T., Sarigiannidis, P., Zervakis, M., Livanos, G., 2020. Towards smart
Karar, M.E., Alsunaydi, F., Albusaymi, S., Alotaibi, S., 2021. A new mobile application of farming: Systems, frameworks, and exploitation of multiple sources. Comput. Netw.
agricultural pests recognition using deep learning in cloud computing system. 172 (107147), 1–14.
Alexandria Engineering Journal 60, 4423–4432. MR Machado, TR Júnior, MR Silva and JB Martins (2019), “Smart Water Management
Kaur, S., Pandey, S., Goel, S., 2018. Semi-automatic leaf disease detection and System using the Microcontroller ZR16S08 as IoT Solution,” In IEEE 10th Latin
classification system for soybean culture. IET Image Proc. 12 (6), 1038–1048. American Symposium on Circuits & Systems (LASCAS), Armenia, Colombia., 169-
Kaur, S., Pandey, S., Goel, S., 2018. A semi-automatic leaf disease detection and 172.
classification system for soybean culture. IET Image Proces. 12 (6), 45–67. Mahlein, A.K., Kuska, M.T., Thomas, S., Wahabzada, M., Behmann, J., Rascher, U.,
Kaur, R., Singla, S., 2016. Classification of plant leaf diseases using gradient and texture Kersting, K., 2019. Quantitative and qualitative phenotyping of disease resistance of
feature. In: In International Conference on Advances in Information Communication crops by hyperspectral sensors: seamless interlocking of phytopathology, sensors,
Technology & Computing, pp. 96–107. and machine learning is needed! Curr. Opin. Plant Biol. 50, 156–162.
Keling, T.U., Linjuan, L.I., Liming, Y., Jianhua, W., Qun, S., 2018. Selection for high- Maimaitijiang, M., Sagan, V., Sidike, P., Hartling, S., Esposito, F., Fritschi, F.B., 2020.
quality pepper seeds by machine vision and classifiers. J Integr Agric 17 (9), Soybean yield prediction from UAV using multimodal data fusion and deep learning.
1999–2006. Remote Sens. Environ, 237–257.
Keshtgari, M., Deljoo, A., 2012. A wireless sensor network solution for precision Malchi, S.K., Kallam, S., Al-Turjman, F., Patan, R., 2021. A trust-based fuzzy neural
agriculture based on ZigBee technology. Wirel. Sens. Netw. 4, 25–30. network for smart data fusion in the internet of things. Comput. Electr. Eng. 89
Khaki, S., Wang, L., 2019. Crop yield prediction using deep neural networks. Front. Plant (106901), 1–35.
Sci. 10, 621–637. Matei, O., Rusu, T., Petrovan, A., Mihut, G., 2017. A data mining system for real-time soil
Khaki, S., Wang, L., Archontoulis, S.V., 2020. A cnn-rnn framework for crop yield moisture prediction. Procedia Eng. 181, 837–844.
prediction. Front. Plant Sci. 10, 1750–1782. Mehra, T., Kumar, V., Gupta, P., 2016. Maturity and disease detection in tomato using
Khamparia, A., Saini, G., Gupta, D., Khanna, A., Tiwari, S., Albuquerque, V.H.C., 2020. computer vision. In: In Fourth International Conference on Parallel, Distributed and
Seasonal crops disease prediction and classification using deep convolutional Grid Computing (PDGC). https://doi.org/10.1109/pdgc.2016.7913228.
encoder network. Circ. Syst. Sign. Proc. 39, 818–836. Mekala, M.S., Viswanathan, P., 2017. A Survey: smart agriculture IoT with cloud
T Khan, J Qiua, MAA Qureshi, MS Iqbal, R Mehmood, and W Hussain (2019), computing. In: International conference on microelectronic devices, circuits and
“Agricultural Fruit Prediction Using Deep Neural Networks”, International systems, ICMDCS Vellore. IEEE, India, pp. 1–7.
Conference on Identification, Information and Knowledge in the Internet of Things Meshram, V., Patil, K., Meshram, V., Hanchate, D., Ramkteke, S.D., 2021. Machine
(IIKI2019), Procedia Computer Science 174: 72–78. learning in agriculture domain: A state-of-art survey. Artificial Intelligence in the
Khirade, S.D., Patil, A.B., 2015. Plant disease detection using image processing. In: In Life Sciences 1 (100010), 1–23.
International Conference on Computing Communication Control and Automation. Minh, D.H.T., Ienco, D., Gaetano, R., Lalande, N., Ndikumana, E., 2017. Deep recurrent
IEEE Computer Society, pp. 768–771. neural networks for winter vegetation quality mapping via multitemporal SAR
Kibriya, H., Abdullah, I., Nasrullah, A., 2021. Plant Disease Identification and Sentinel. IEEE Geosci. Remote Sens. Lett. 99, 1–5.
Classification Using Convolutional Neural Network and SVM. In: 2021 International Mohameth, F., Bingcai, C., Sada, K.A., 2020. Plant disease detection with deep learning
Conference on Frontiers of Information Technology (FIT), pp. 264–268. https://doi. and feature extraction using Plant Village. J. Comp. Commun. 8 (6), 10–22.
org/10.1109/FIT53504.2021.00056. Mohanty, P.S., Hughes, D.P., Salathe, M., 2016. Using Deep Learning for Image-Based
Kitpo, N., Inoue, M., 2018. “Early rice disease detection and position mapping system Plant Disease Detection”. Front. Plant Sci. 7, 1–10.
using drone and IoT architecture”, In 12th South East Asian Technical University Morellos, A., Pantazi, X., Moshou, D., Alexandridis, T., Whetton, R., Tziotzios, G.,
Consortium (SEATUC). IEEE 1, 1–5. Wiebensohn, J., Bill, R., Mouazen, A.M., 2016. Machine learning-based prediction of
Klerkxa, L., Jakkub, E., Labarthe, P., 2019. A review of social science on digital soil total nitrogen, organic carbon and moisture content by using VIS-NIR
agriculture, smart farming and agriculture 4.0: New contributions and a future spectroscopy. Biosyst. Eng. 21 (52), 104–116.
research agenda. NJAS - Wageningen Journal of Life Sciences 90 (100315), 1–16.

27
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

Moshou, D., Bravo, C., Oberti, R., West, J., Bodria, L., McCartney, A., Ramon, H., 2005. Ranjan, A.K., Parida, B.R., 2019. Paddy acreage mapping and yield prediction using
Plant disease detection based on data fusion of hyperspectral and multispectral sentinel-based optical and SAR data in Sahibganj district, Jharkhand (India). Spatial
fluorescence imaging using Kohonen maps. Real-Time Imaging 11 (2), 75–83. Inf. Res 1–19.
Mueller-Sim, T., Jenkins, M., Abel, J., Kantor, G., 2017. The Robotanist: a ground-based Ray, P.P., 2017. Internet of things for smart agriculture: Technologies, practices and
agricultural robot for high-throughput crop phenotyping. In IEEE International future direction. J. Ambient Intell. Smart Environ. 9, 395–420.
Conference on Robotics and Automation (ICRA). https://doi.org/10.1109/ Reddy, D.J., Kumar, M.R., 2021. Crop Yield Prediction using Machine Learning
icra.2017.7989418. Algorithm. In: In 5th International Conference on Intelligent Computing and Control
Mulla, D.J., 2013. Twenty five years of remote sensing in precision agriculture: key Systems (ICICCS), pp. 1466–1470.
advances and remaining knowledge gaps. Biosyst. Eng. 114 (4), 358–371. Reddy, K.S.P., Roopa, Y.M., Rajeev, K.L.N., Nandan, N.S., 2020. IoT based Smart
Naito, H., Ogawa, S., Valencia, M.O., Mohri, H., Urano, Y., Hosoi, F., Omasa, K., 2017. Agriculture using Machine Learning. In: In Second International Conference on
Estimating rice yield-related traits and quantitative trait loci analysis under different Inventive Research in Computing Applications (ICIRCA), pp. 130–134.
nitrogen treatments using a simple tower-based field phenotyping system with Reddy, T.V., Sashirekhak, K., 2020. Examination on advanced machine learning
modified single-lens reflex cameras. ISPRS J. Photogramm. Remote Sens. 125, techniques for plant leaf disease detection from leaf imagery. J Crit Rev 7 (5),
50–62. 1208–1221.
Namin, S.T., Esmaeilzadeh, M., Najafi, M., Brown, T.B., Borevitz, J.O., 2017. Deep Rehman, A., Abbasi, A.Z., Islam, N., Shaikh, Z.A., 2014. A review of wireless sensors and
phenotyping: Deep learning for temporal phenotype/genotype classification. 134 networks’ applications in agriculture. Computer Standards & Interfaces 36 (2),
(205), 1–37. 263–270.
VA Natarajan, MS Kumar, R Patan, S Kallam, and MYN Mohamed (2020), “Segmentation Rupnik, R., Kukar, M., Vracar, P., Kosir, D., Pevec, D., Bosnic, Z., 2018. AgroDSS: a
of Nuclei in Histopathology images using Fully Convolutional Deep Neural decision support system for agriculture and farming. Comput. Electron. Agric. 161,
Architecture”, In International Conference on Computing and Information 260–271.
Technology (ICCIT-1441), IEEE:1-7). Sannakki, S.S., Rajpurohit, V.S., Nargund, V.B., Kulkarni, P., 2013. Diagnosis and
Nation United (2017). Sustainable development goals. https://sdgs.un.org/goals. [Last classification of grape leaf diseases using neural networks. In: In Fourth International
visited 18 November 2021]. Conference on Computing, Communications and Networking Technologies
Nevavuori, P., Narra, N., Lipping, T., 2019. Crop yield prediction with deep (ICCCNT), pp. 1–5.
convolutional neural networks. Comput. Electron. Agric, 163–189. Saraansh, B., Siddhant, K., Anuja, A., 2019. Deep Learning Convolutional Neural
Nguyen, L.H., Zhu, J., Lin, Z., Du, H., Yang, Z., Guo, W., Jin, F., 2019. Spatial-temporal Network for Apple Leaves Disease Detection. Technology and Management
multi-task learning for within-field cotton yield prediction. In: Pacific-Asia (SUSCOM). Amity University Rajasthan, Jaipur - India.
Conference on Knowledge Discovery and Data Mining. Springer, Cham, pp. 343–354. Saravi, B., Nejadhashemi, A.P., Tang, B., 2019. Quantitative model of irrigation effect on
S Nosratabadi, F Imre, K Kando, K Szell, A, Regia, S Ardabili, and A Mosavi (2021) maize yield by deep neural network. Neural Comput. Appl, 1–14.
Hybrid Machine Learning Models for Crop Yield Prediction, https://arxiv.org/ftp/ Sarker, I.H., 2021. Deep Learning: A Comprehensive Overview on Techniques,
arxiv/papers/2005/2005.04155.pdf, 1-5. Taxonomy, Applications and Research Directions. SN COMPUT. SCI. 2, 420. https://
Nurzaman, A., De, D., Hussain, I., 2018. Internet of things (IoT) for smart precision doi.org/10.1007/s42979-021-00815-1.
agriculture and farming in rural areas. IEEE Internet Things J. 5 (6), 4890–4899. Schor, N., Bechar, A., Ignat, T., Dombrovsky, A., Elad, Y., Berman, S., 2016. Robotic
Oca, A.M., Flores, G., 2021. The AgriQ: A low-cost unmanned aerial system for precision disease detection in greenhouses: combined detection of powdery mildew and
agriculture. Expert Syst. Appl. 182 (115163), 1–19. tomato spotted wilt virus. IEEE Rob. Autom. Lett. 1 (1), 354–360.
Oliveira, D.T., Silva, R.P., Maldonado, J.W., Zerbato, C., 2020. Convolutional neural Schwalbert, R.A., Amado, T., Corassa, G., Pott, L.P., Prasad, P.V., Ciampitti, I.A., 2020.
networks in predicting cotton yield from images of commercial fields. Comput. Satellite-based soybean yield forecast: Integrating machine learning and weather
Electron. Agric, 171–197. data for improving crop yield prediction in southern Brazil. Agric. For. Meteorol.
Padalalu, P., Mahajan, S., Dabir, K., Mitkar, S., Javale, D., 2017. Smart water dripping 284, 1–29.
system for agriculture/farming. In: In 2nd International Conference for Convergence Segalla, A., Fiacco, G., Tramarin, L., Nardello, M., Brunelli, D., 2020. Neural networks for
in Technology (I2CT), pp. 659–662. pest detection in precision agriculture. In Proc. IEEE Int. Workshop Metrol. Agricult.
Padol, P.B., Yadav, A.A., 2016. SVM classifier based grape leaf disease detection. In: In Forestry (MetroAgriFor) 7–12.
Conference on Advances in Signal Processing (CASP), pp. 175–179. Selim, H., 2018. Recognition and detection of tea leaf’s diseases using support vector
AV Panchal, SC Patel, K Bagyalakshmi, P Kumar, IR Khan, and M Son (2021) Image- machine. In IEEE 14th International Colloquium on Signal Processing & Its
based Plant Diseases Detection using Deep Learning”, Materials Today: Proceedings, Applications (CSPA).
xxx (xxxx) xxx: 1-7. Selvaraj, M.G., Vergara, A., Ruiz, H., Safari, N., Elayabalan, S., Ocimati, W., Blomme, G.,
Pastor, F.F., Chamizo, G.J., Hidalgo, N.M., Martínez, M.J., 2018. Precision agriculture 2019. AI-powered banana diseases and pest detection. Plant Methods 15, 92.
design method using a distributed computing architecture on the internet of things Sengar, N., Dutta, M.K., Travieso, C.M., 2018. Computer vision-based technique for
context. Sensors 18, 1731–1749. identification and quantification of powdery mildew disease in cherry leaves.
Pathan, M., Patel, N., Yagnik, H., Shah, M., 2020. Artificial cognition for applications in Computing 1–13.
smart agriculture: A comprehensive review. Artificial Intelligence in Agriculture 4, Sengupta, A., Debnath, B., Das, A., De, D., 2021. armFox: A QuadSensor-Based IoT Box
81–95. for Precision Agriculture. IEEE Consumer Technology Society 21 (62), 63–68.
Patil, S.P., Zambre, S.R., 2014. Classification of cotton leaf spot disease using support Shafiekhani, A., Kadam, S., Fritschi, F., DeSouza, G., 2017. Vinobot and vinoculer: two
vector machine. Int. J. Eng. Res. Appl. 4, 92–97. robotic platforms for high-throughput field phenotyping. Sensors 17 (12), 214–241.
Peng, J., 2019. Real-time detection of apple leaf diseases using deep learning approach Shaikh, T.A., Ali, R., Beg, M.M.S., 2020. Transfer learning privileged information fuels
based on improved convolutional neural networks. IEEE Access 7, 59069–59080. CAD diagnosis of breast cancer. Mach. Vis. Appl. 31, 9. https://doi.org/10.1007/
Perea, R.G., Poyato, E.C., Montesinos, P., Díaz, J.A.R., 2019. Prediction of irrigation s00138-020-01058-5.
event occurrence at farm level using optimal decision trees. Comput. Electron. Agric. Tawseef Ayoub Shaikh and Rashid Ali Applying Machine Learning Algorithms for Early
157, 173–180. Diagnosis and Prediction of Breast Cancer Risk, 2018, Proceedings of 2nd
Pereira, W.F., Fonseca, L.S., Putti, F.F., Goes, B.C., Naves, L.P., 2020. Environmental International Conference on Communication, Computing and Networking, Lecture
monitoring in a poultry farm using an instrument developed with the internet of Notes in Networks and Systems: 589-598, Springer.
things concept. Comput. Electron. Agric. 170 (105257), 1–20. Tawseef Ayoub Shaikh, Rashid Ali, Waseem Ahmad Mir, “An Intelligent Healthcare
Pivoto, D., Waquil, P.D., Talamini, E., Finocchio, C.P.S., Corte, V.F.D., Mores, G.V., 2018. System for Automated Alzheimer’s Disease Prediction and Personalized Care.”
Scientific development of smart farming technologies and their application in Brazil. Volume 12, Issue 2, pp: p240-253, 2021.
Information processing in Agriculture 5, 21–32. Sharifa, M., Khana, M.A., Iqbal, Z., Azam, M.F., Lalib, M.I.U., Younus, M., 2018. “Javed
Qiao, M., He, X., Cheng, X., Li, P., Luo, H., Zhang, L., Tian, Z., 2021. Crop yield Detection and classification of citrus diseases in agriculture based on optimized
prediction from multi-spectral, multi-temporal remotely sensed imagery using weighted segmentation and feature selection. Comput. Electron. Agric. 150,
recurrent 3D convolutional neural networks. Int. J. Appl. Earth Obs. Geoinf. 102 220–234.
(102436), 1–12. S Sharma, S Rai, and NC Krishnan (2020) Wheat Crop Yield Prediction Using Deep LSTM
Qin, W.C., Qiu, B.J., Xue, X.Y., Chen, C., Xu, Z.F., Zhou, Q.Q., 2016. Droplet deposition Model CoRR abs/2011.01498, https://arxiv.org/abs/2011.01498: 1-8.
and control effect of insecticides sprayed with an unmanned aerial vehicle against Shidnal, S., Latte, M.V., Kapoor, A., 2019. Crop yield prediction: two-tiered machine
planthoppers. Crop Prot. 85, 79–88. learning model approach. Int. J. Inf. Technoly 10, 1–9.
Rahnemoonfar, M., Sheppard, C., 2017. Real-time yield estimation based on deep Shin, J., Chang, K.Y., Heung, B., Quang, T.N., Price, G.W., Mallahi, A.A., 2021. A deep
learning. In Autonomous Air and Ground Sensing Systems for Agricultural learning approach for RGB image-based powdery mildew disease detection on
Optimization and Phenotyping 10218, 1021809–1021821. strawberry leaves. Comput. Electron. Agric. 183 (106042), 1–8.
Ramcharan, A., Baranowski, K., McCloskey, P., Ahmed, B., Legg, J., Hughes, D.P., 2017. Shivali Amit Wagle, 2021. Harikrishnan R, Jahariah Sampe, Faseehuddin Mohammad,
Deep learning for image-based cassava disease detection. Front. Plant Sci. 8, Sawal Hamid Md Ali, “Effect of Data Augmentation in the Classification and
1852–1875. Validation of Tomato Plant Disease with Deep Learning Methods”. Traitement du
Ramli, M.R., Daely, P.T., Kim, D.S., Lee, J.M., 2020. IoT-based adaptive network Signal 38 (6), 1657–1670.
mechanism for reliable smart farm system. Comput. Electron. Agric. 170 (105287), Shook, J., Gangopadhyay, T., Wu, L., Ganapathysubramanian, B., Sarkar, S., Singh, A.K.,
1–17. 2021. Crop yield prediction integrating genotype and weather variables using deep
Rangarajan, A.K., Purushothaman, R., Ramesh, A., 2018. Tomato crop disease learning. PlosOne 1371, 1–19.
classification using a pre-trained deep learning algorithm. Procedia Comput. Sci. Singh, S., Singh, N.P., 2019. “Machine Learning-Based Classification of Good and Rotten
133, 1040–1047. Apple”, In Recent Trends in Communication, Computing, and Electronics. Lecture
Notes in Electrical Engineering, Springer, Singapore, p. 524.

28
T. Ayoub Shaikh et al. Computers and Electronics in Agriculture 198 (2022) 107119

Sladojevic, S., Arsenovic, M., Anderla, A., Culibrk, D., Stefanovic, D., 2016. Deep neural Wang, X., Huang, J., Feng, Q., Yin, D., 2020. Winter wheat yield prediction at county
networks-based recognition of plant diseases by leaf image classification. level and uncertainty analysis in main wheat-producing regions of china with deep
Computational Intelligence and Neuroscience 3289801, 11–41. learning approaches. Remote Sens. 12 (11), 1744–1777.
Sreedhar, B., Kumar, M.S., 2020. In: A Comparative Study of Melanoma Skin Cancer Wang, Y., Zhang, Z., Feng, L., Du, Q., Runge, T., 2020. Combining multi-source data and
Detection in Traditional and Current Image Processing Techniques. and Cloud)(I- machine learning approaches to predict winter wheat yield in the conterminous
SMAC) IEEE, pp. 654–658. United States. Remote Sens. 12 (8), 1232–1259.
Suchithra, M.S., Pai, M.L., 2019. Improving the prediction accuracy of soil nutrient Warne, P.P., Ganorkar, S.R., 2015. Detection of diseases on cotton leaves using K-mean
classification by optimizing extreme learning machine parameters. Inf Process clustering method. Int. Res. J. Eng. Technol. 2 (4), 1–29.
Agricul 7 (1), 72–82. Wolanin, A., García, G.M., Valls, G.C., Chova, L.G., Meroni, M., Duveiller, G., Guanter, L.,
Suit, S.T., Kumaraswamy, R., 2019. Performance Comparison of Weed Detection 2020. Estimating and understanding crop yields with explainable deep learning in
Algorithms. In: International Conference on Communication and Signal Processing the Indian Wheat Belt. Environ. Res. Lett. 15 (2), 1–23.
(ICCSP). Chennai, India, pp. 0843–0847. Xie, T., Huang, Z., Chi, Z., Zhu, T., 2017. Minimizing amortized cost of the on-demand
Sujatha, R., Chatterjee, J.M., Jhanjhi, N.Z., Brohi, S.N., 2021. Performance of deep irrigation system in smart farms. In: In: 3rd International Workshop on Cyber-
learning vs machine learning in plant leaf disease detection. Microprocess. Physical Systems for Smart Water Networks, pp. 43–46.
Microsyst. 80 (103615), 1–11. Xu, W., Chen, P., Zhan, Y., Chen, S., Zhang, L., Lan, Y., 2021. Cotton yield estimation
Sun, J., Di, L., Sun, Z., Shen, Y., Lai, Z., 2019. County-level soybean yield prediction model based on machine learning using time series UAV remote sensing data. Int. J.
using deep CNN-LSTM model. Sensors 19 (20), 4363–4391. Appl. Earth Obs. Geoinf. 104 (102511), 1–13.
Syifa, M., Park, S.J., Wook, C., 2020. Lee Detection of the Pine Wilt Disease Tree Xu, X., Gao, P., Zhu, X., Guo, W., Ding, J., Li, C., Wu, X., 2019. Design of an integrated
Candidates for Drone Remote Sensing Using Artificial Intelligence Techniques. climatic assessment indicator (ICAI) for wheat production: a case study in Jiangsu
Engineering 6, 919–926. Province. China. Ecol. Ind. 101, 943–953.
Szegedy, C., Vanhoucke, V., Ioffe, S., Shlens, J., Wojna, Z., 2016. Rethinking the Yalcin, H., 2017. Plant phenology recognition using deep learning. In: 6th International
inception architecture for computer vision. In: In Proceedings of the IEEE conference Conference on Deep-Pheno. Agro-Geoinformatics, IEEE, pp. 31–44.
on computer vision and pattern recognition, pp. 2818–2826. Yalcin, H., 2019. An approximation for a relative crop yield estimate from field images
Tang, J.L., Wang, D., Zhang, Z.G., He, L., Xin, J., Xu, Y., 2017. Weed identification based using deep learning. In: In 8th International Conference on Agro-Geoinformatics
on K-means feature learning combined with the convolutional neural network. (Agro-Geoinformatics), pp. 1–6.
Comput. Electron. Agric. 135, 63–70. Yang, Q., Shi, L., Han, J., Zha, Y., Zhu, P., 2019. Deep convolutional neural networks for
Tawseef Ayoub Shaikh, 2019. Rashid Ali, Automated atrophy assessment for Alzheimer’s rice grain yield estimation at the ripening stage using UAV-based remotely sensed
disease diagnosis from brain MRI images. Magn. Reson. Imaging 62, 167–173. images. Field Crops Res. 235, 142–153.
Tello, J.T., Ko, S.B., 2021. Identifying Useful Features in Multispectral Images with Deep Yang, M., Xu, D., Chen, S., Li, H., Shi, Z., 2019. Evaluation of machine learning
Learning for Optimizing Wheat Yield Prediction. In: In IEEE International approaches to predict soil organic matter and pH using vis-NIR spectra. Sensors 19
Symposium on Circuits and Systems (ISCAS), pp. 1–5. (2), 263–277.
Terence, S., Purushothaman, G., 2020. Systematic review of Internet of Things in smart Yu, C., 2020. Plant spike: A low-cost, low-power beacon for smart city soil health
farming. Trans Emerging Tel technology 39, 1–34. monitoring. IEEE Internet Things J. 7 (9), 9080–9090.
Terliksiz, A.S., Altylar, D.T., 2019. Use Of deep neural networks for crop yield prediction: Yu, N., Li, L., Schmitz, N., Tian, L.F., Greenberg, J.A., Diers, B.W., 2016. Development of
a case study Of Soybean Yield in Lauderdale County, Alabama, USA. In: In 8th methods to improve soybean yield estimation and predict plant maturity with an
International Conference on Agro-Geoinformatics (Agro-Geoinformatics), pp. 1–4. unmanned aerial vehicle-based platform. Remote Sens. Environ. 187, 91–101.
Thakur, D., Kumar, Y., Vijendra, S., 2020. Smart irrigation and intrusions detection in Yu, J., Sharpe, S.M., Schumann, A.W., Boyd, N.S., 2019. Deep learning for image-based
agricultural fields using I. Proced Comput Sci 167, 154–162. weed detection in turfgrass. Eur. J. Agron. 104, 78–84.
Tiwari, D., Ashish, M., Gangwar, N., Sharma, A., Patel, S., Bhardwaj, S., 2020. Potato leaf Zaragoza, A.C., Perea, G.R., García, F.I., Poyato, C.E., Díaz, R.J.A., 2020. Open source
diseases detection using deep learning. In: 4th International Conference on application for optimum irrigation and fertilization using reclaimed water in olive
Intelligent Computing and Control Systems (ICICCS). IEEE, Madurai, India, India, orchards. Comput. Electron. Agric. 2173 (105407), 1–27.
pp. 461–466. Zhang, S., Guo, J., Wang, Z., 2019. Combing K-means Clustering and Local Weighted
Too, E.C., Yujian, L., Njuki, S., Yingchun, L., 2019. A comparative study of fine-tuning Maximum Discriminant Projections for Weed Species Recognition. Front. Comput.
deep learning models for plant disease identification. Comput. Electron. Agric. 161, Sci. 1 (4), 1–29.
272–279. Zhang, X., Han, L., Dong, Y., Shi, Y., Huang, W., Han, L., Moreno, G.P., Ma, H., Ye, H.,
NM Trendov, S Varas, and M Zeng (2019) Digital technologies in agriculture and rural Sobeih, T.A., 2019. Deep learning-based approach for automated yellow rust disease
areas - status report, Tech. Rep., Nations. Rome, Italy. Food and Agriculture detection from high-resolution hyperspectral UAV images. Remote Sens 11, 1554.
Organization of the United, 1-19. Zhang, S., Zhang, S., Zhang, C., Wang, X., Shi, Y., 2019. Cucumber leaf disease
Tu, Y.H., Johansen, K., Phinn, S., Robson, A., 2019. Measuring canopy structure and identification with global pooling dilated convolutional neural network. Comput.
condition using multi-spectral UAS imagery in a horticultural environment. Remote Electron. Agric. 162, 422–430.
Sens. 11, 269–299. Zhang, L., Zhang, Z., Luo, Y., Cao, J., Tao, F., 2020. Combining optical, fluorescence,
Turkoglu, M., Hanbay, D., 2019. Plant disease and pest detection using deep learning- thermal satellite, and environmental data to predict county-level maize yield in
based features. Turk J Electr Eng Comput Sci 27, 1636–1651. china using machine learning approaches. Remote Sens. 12 (1), 21–48.
Ubbens, J.R., Stavness, I., 2017. Deep plant phenomics: a deep learning platform for Zhou, X., Lee, W.S., Ampatzidis, Y., Chen, Y., Peres, N., Fraisse, C., 2021. Strawberry
complex plant phenotyping tasks. Front. Plant Sci. 8, 90–111. Maturity Classification from UAV and Near-Ground Imaging Using Deep Learning.
Unal, Z., 2020. Smart Farming Becomes Even Smarter With Deep Learning—A Smart Agricultural Technology 1 (100001), 1–8.
Bibliographical Analysis. IEEE Access 8, 105587–105609. Zhou, C., Ye, H., Hu, J., Shi, X., Hua, S., Yue, J., Xu, Z., Yang, G., 2019. Automated
United Nations (2019), Department of Economic Affairs Social, Division Population. counting of rice panicle by applying deep learning model to images from unmanned
World population prospects 2019: highlights. aerial vehicle platform. Sensors 19 (14), 3106–3131.
P Varga, S Plosz, G Soos, and Hegedus, “Security threats and issues in automation Io”, In Zhu, N., Liu, X., Liu, Z., Hu, K., Wang, Y., Tan, J., Huang, M., Zhu, Q., Ji, X., Jiang, Y.,
IEEE international workshop on factory communication systems proceedings, WFCS, Guo, Y., 2018. Deep learning for smart agriculture: Concepts, tools, applications, and
IEEE, Trondheim, Norway, ISBN 9781509057887, 6-19. opportunities. Int J Agric & Biol Eng 11 (4), 32–44.
Veeramani, B., Raymond, J.W., Chanda, P., 2018. DeepSort: deep convolutional Zhu, S., Zhou, L., Gao, P., Bao, Y., He, Y., Feng, L., 2019. Near-infrared hyperspectral
networks for sorting haploid maize seeds. BMC Bioinf. 19 (9), 289–319. imaging combined with deep learning to identify cotton seed varieties. Molecules 24,
Wang, P., Hafshejani, B.A., Wang, D., 2021. An improved multilayer perceptron 3268–3291.
approach for detecting sugarcane yield production in IoT-based smart agriculture.
Microprocess. Microsyst. 82 (103822), 1–7.

29