See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/363883701

IoT-Assisted Crop Monitoring Using Machine Learning Algorithms for Smart

Farming

Chapter · September 2022

DOI: 10.1007/978-981-19-1412-6_1

CITATIONS READS

8 1,245

4 authors, including:

Srujan kotagiri Raju Dr. Neelamadhab Padhy

CMR Technical Campus GIET University
187 PUBLICATIONS 575 CITATIONS 140 PUBLICATIONS 1,770 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Srujan kotagiri Raju on 07 October 2022.

The user has requested enhancement of the downloaded file.

IoT-Assisted Crop Monitoring Using
Machine Learning Algorithms for Smart
Farming

Shraban Kumar Apat, Jyotirmaya Mishra, K. Srujan Raju,

and Neelamadhab Padhy

Abstract Agriculture expansion is critical to the economic prosperity of any country.

Agriculture employs more than 60% of the Indian population, either directly or
indirectly. Nowadays, monitoring the crop is the challenging task in the world. In this
article, data has been collected from various sensors to propose an IoT-assisted hybrid
machine learning approach for obtaining an effective crop monitoring system. Crop
monitoring system here means predicting as well as detecting diseases of crops. This
study is about leveraging existing data and applying regression analysis, SVM, and
decision tree to predict crop diseases in diverse crops such as rice, ragi, gram, potato,
and onion. Among the applied methods, SVM outperforms regression, DT methods.
The training and testing accuracy of Gram has 96.29% and 95.67%, respectively.

Keywords SVM · NB · Crop monitoring · IoT-assisted sensors

1 Introduction

Agriculture research is rapidly expanding as a result of technological advancements.

It provides numerous academics with the potential to address upcoming agricultural
difficulties. Humidity, temperature, rainfall, wind speed, light intensity, and soil pH
are all important environmental elements in precision agriculture. Smart farming is

S. Kumar Apat (B) · J. Mishra · K. Srujan Raju · N. Padhy

Computer Science & Engineering, School of Engineering & Technology, GIET University,
Gunupur 765022, Odisha, India
e-mail: [email protected]
J. Mishra
e-mail: [email protected]
N. Padhy
e-mail: [email protected]
K. Srujan Raju
Department of Computer Science and Engineering, CMR Technical Campus, Hyderabad,
Telangana, India

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 1
R. Kumar et al. (eds.), Next Generation of Internet of Things, Lecture Notes
in Networks and Systems 445, https://doi.org/10.1007/978-981-19-1412-6_1
2 S. Kumar Apat et al.

vital for dealing with a variety of difficulties in the agriculture area, such as produc-
tion, early disease detection, environmental effect, and climatic circumstances. Crops
that are healthy maximize productivity; otherwise, farmers will lose a lot of money.
In whole-crop production, it is critical to identify the pest. Pest control sensors keep
an eye on the pest population as well as the surrounding environment. Farmers can
follow the full history of pest attacks by using IoT-enabled pest control devices. The
appropriate and proper quantity of pesticides must be automatically sprayed in the
field based on this data. Wireless sensor networks track and detect insect popula-
tions, automatically monitor chemical delivery systems, and activate and maintain
crop health. The comprehensive study of machine learning principles, algorithms,
and methodologies is described in this chapter. This article provides an overview of
the suggested disease detection and prevention strategy in agriculture. In whole-crop
production, identifying the pest is critical. Pest control sensors keep track of both the
pest population and the environment. Farmers can keep a complete record of pest
attacks by implementing IoT-enabled pest control sensors. Based on this information,
the required and appropriate quantity of pesticides must be automatically sprayed
into the field. Wireless sensor networks can detect and monitor pest populations,
automatically monitor chemical delivery systems, and activate and maintain crop
health. This study investigate ML algorithm for obtaining an effective crop moni-
toring system. Crop monitoring systems are used to both predict and detect crop
diseases. This chapter also covers crop management operations such as crop yield
prediction and disease detection.

2 Literature Survey

Yang et al. [1] Machine learning is being used to uncover plant resistance genes and
classify plant diseases. To achieve the best classification accuracy, a careful selection
of preprocessing data approaches and machine learning technologies was applied.
To forecast essential plant resistance genes, more machine learning-based methods
are required. To identify bacterial pathogens with high prediction precisions, ML
methods such as SVM, Bayesian classifier, and RF were used (23–25). Deep convo-
lution neural networks (CNN), the most recent generation of machine learning tech-
nologies, were used. Jawade et al [2] The field sensors gathered real-time weather data
in order to anticipate disease in real time. The disease outbreak probability was calcu-
lated using the RF Regression model, which was trained on historical meteorological
data. In terms of disease predicting, the model produced fairly accurate findings. The
random forest algorithm has been shown to be fairly accurate in forecasting the possi-
bility of a thrips attack. This technique has aided farmers in taking preventative steps,
resulting in increased farm output and a reduction in the use of chemical pesticides on
crops. Akulwar et al. [3] presents an example of how a recommender system is used
in agriculture to detect and predict disease. Image preprocessing comprises noise
removal and region of interest extraction. After preprocessing, feature extraction
will be performed. Following the extraction of features, a ML algorithm is used, and
IoT-Assisted Crop Monitoring Using Machine Learning Algorithms … 3

the outcomes are anticipated. To make the best recommendations, the recommender
system is used.
Ramesh et al. [4] crop selection methods are used to boost crop productivity.
I used two alternative approaches. To assess datasets, the first is the Naive Bayes
approach, and the second is the K-Nearest Neighbor method. Bhawana et al. [5]
machine learning is being used in wheat crop production. The feature extraction of
that digital image is done using digital image processing techniques. With age, the
green pigment in wheat crops decreases. Supervised machine learning is used to clas-
sify crop growth stages. Technique SVM. Bondre and colleagues [6] Mahagaonkar.
ML algorithm such as SVM and RF were applied to agricultural data, and a fertilizer
recommendation for a specific crop was made. The aim of this research is to analyze
one crop forecasting model which can be used in the latter stage. Crop yield data
was gathered over the last five years from a variety of sources. The proposed work
is divided into two stages: 1. Soil classification 2. Crop Yield Recommendation for
Fertilizer Prediction Weather, temperature, humidity, and atmospheric pressure are
all factors to consider. Third-party applications are used to display press information.
Bhawana et al. [5] the feature extraction of that digital image is done using digital
image processing techniques. With age, the green pigment in wheat crops decreases.
As a result, the wheat crop maturities were investigated. They determined the crop’s
growth stages and classified it using the supervised machine learning approach SVM.
Optimizations along with fine-tuned choices of the ML techniques were used. Kumar
et al. [7] the best crop yield model was created using a random forest and a decision
tree. The suggested approach focuses on crop kind, yield, and weather predictions.
The information in the collection is based on agricultural statistics. This dataset serves
as a testing ground. After the data has been processed, it is broken into two subtypes:
i.e., testing and training. Temperature, rainfall, humidity, and ph are among the vari-
ables in the dataset [9]. The datasets were referred from the Kaggle, which contains
3101 instances or data that were derived from historical data with the following
parameters like humidity, rainfall Temp, ph, and crop.
Archana et al. [8] used Biosensing methods and DNA sequencing technologies to
detect fungal crop diseases. Pathogenic fungus causes diseases like coils, scab, gall,
leaf spot, root rot, and mildew, by making use of third-party applications. Concluded
Systemic foliar infections are a major source of production and crop losses, as well as
quality of crop degradation Kumar et al. [9]. Techniques such as SVM were used to
recommend the suitable crop to the farmer considering environmental condition and
crop disease prediction, and to suggest ways to protect them from illness Fenu et al.
[10]. Specific methodologies and procedures, along with performance parameters and
preprocessing techniques, were utilized. The study’s findings show that this approach
is still in its early stages, with significant obstacles to overcome. Data were collected
from different databases such as IEEE Xplore, Science Direct, MDPI, Hindawi, and
Web of Science. Over the last ten years, this study looked at classified forecasting
models for crop disease (2010–2020). Aggarwal et al. [11] used Traditional agricul-
tural procedures are being revolutionized by modern agricultural techniques that use
IoT and AI, and farming is becoming a profitable venture as well. The study explains
4 S. Kumar Apat et al.

how AI and IoT might boost agricultural output by presenting several recommen-
dations and implementation strategies. Singh et al. [11] this research leads to the
creation of the Crop, a smart phone app based on a trained illness prediction model.
Farmers can use the app to take crop photos and analyze the presence or absence
of disease. ResNet 50, AlexNet, and ResNet 34 are used to train the image classi-
fier. After four epochs of training with the fastai approach fit one cycle, ResNet 50
achieved a max accuracy of 98.73%. The most efficient network is ResNet 50 Mishra
et al. [12]. This study recommended a methodology for determining the suitable crop
and monitoring agricultural conditions in the field on a large scale using weather
analysis data and crop disease diagnostics for improved production using real-time
data. The data collected from sensors was input into a Decision Tree algorithm in
the in-plant monitoring model to identify the plant based on its circumstances. The
author used a CNN-based approach to identify and detect plant diseases from leaf
images… Sladojevic et al. [13]. Berkeley Vision and Learning Center created a deep
learning framework for the deep CNN training. For independent class examinations,
the experimental findings on the built model achieved precision ranging from 91 to
98%, with an average of 96.3%. A new strategy involving deep learning methods was
investigated in order to automatically categorize and detect plant illnesses from leaf
photos. In 13 different diseases, the proposed model was able to detect the presence
of leaves and distinguish between healthy and diseased leaves. The authors intend
to contribute significantly to long-term development by influencing crop quality for
future generations. 3000 unique images are gathered from various online sources
and expanded. Sujatha et al. [14] RDL is the domain’s top performer. DL works in
a similar way to the human brain’s neural structure, with layers and optimizers that
help develop a dependable model with improved accuracy. In their work, we consider
both learning methodologies, and the results of DL are noticeable when compared
to ML. Precision, F1 score, accuracy, and area under the curve are all considered. In
comparison, RF received a CA of 76.8%, SGD received 86.5%, SVM received 87%,
VGG-19 received 87%, Inception-v3 received 87.4%, VGG-16 received 89%, and
VGG-16 received 89.5%. On a regular basis, stakeholders collect and send images
to the system, allowing farmers to help decide which insecticide to apply to prevent
harm Mohanty et al. [15] After training a DCNN to detect 14 crop species and 25
illnesses, the trained model achieved an accuracy of 99.35% on a held out test set.
This paves the way for smart phone-assisted crop disease detection and prediction.
54,306 pictures of plant leaves were analyzed. Sambasivam et al. [16] this paper
proposed CNN to create a low-cost deep learning technique for detecting cassava
infections, which included the following steps: Dataset collection, labeling, model
training, and testing/model evaluation using k-fold cross-validation, where k = 3 to
achieve the desired accuracy. The steps for creating the model were as follows: (a)
Setting up the environment, loading, and preprocessing Data − 35% of the time,
(b) Model architecture definition − 10% of the time, (c) Model training − 50% of
the time, and (d) Performance estimation − 5% of the time. Ramcharan et al. [17]
the author used a DCNN model to assess the applicability of transfer learning on
cassava image datasets. The overall accuracy in classifying a leaf into the correct
group ranged from 73% (20–70 split, knn) to 91% (20–70 split, knn) (80–10 split,
IoT-Assisted Crop Monitoring Using Machine Learning Algorithms … 5

SVM). The leaflet cassava dataset had higher overall accuracy, ranging from 80%
(20–70 split, knn) to 93.0% (80–10 split, SVM) Ayub and colleagues [18]. To inves-
tigate the damages, the authors used data mining techniques and a family of ML
techniques (Table 1).

Table 1 Summary of literature review

Ref. No# Year Algorithm used Scope for improvement
1 2019 SVM, Naïve Bayes, SVM Localizer To anticipate essential plant resistance
genes and contribute to agriculture,
more machine learning-based
technologies are required
2 2020 RF regression model. algorithm Image processing utilizing CNN to
predicted performance measurement analyze current plant health can be
0.9 to 0.3 incorporated with this system, and the
results will undoubtedly be very exact
forecasts
3 2020 Used regression and boosting The author concentrated on fruit
techniques diseases, but the findings can be
applied to other crops
4 2019 NB method and KNN applied on Other ML algorithms and different
rice and wheat data set crops are required
5 2020 Supervised machine learning Only experimented with one crop
wheat
6 2019 SVM&RF Only for the purpose of determining
the suitability of the fertilizer to be
used
7 2020 Image processing and ML Use of DL Techniques is possible. For
Techniques different types of crops
8 2020 RF Only UsedParameter like temperature,
rainfall, humidity, ph
9 2019 Biosensing mechanisms and DNA Experimented only on fungal leaf
sequencing technologies diseases
10 2020 SVM approach Suitable to research android apps
fore-commerce of product
11 2021 Preprocessing techniques High-quality, labeled data required
performance metrics
12 2021 IoT and AI AI and IOT used in the broader sense
need to be think specifically
13 2021 Computer Vision and DNN IOT sensor may be used to collect data
more accurately
14 2021 IoT and DT Suggested to consider few more
parameters to check environmental
conditions
(continued)
6 S. Kumar Apat et al.

Table 1 (continued)
Ref. No# Year Algorithm used Scope for improvement
15 2016 CNN Increasing the model’s utility by
training it to recognize plant diseases
on larger land areas
16 2021 Proposed family of ML and DL Used Soft computing techniques and
(Inception-v3, VGG-16, VGG-19) have significant impact on the
system’s CA even when using a small
dataset?
17 2016 Deep convolution NN for 26 Improved accuracy by using more
diseases and 14 crop species diverse set of training data set
18 2021 SMOTE with CNN Multiple diseases co-occurring on the
same plant
19 2017 Deep CNN May be with usage of sensor and smart
phone this will be more user friendly
20 2018 Decision tree, RF, NN, GNB, SVM, hybrid approaches and DL
and KNN (evolutionary and data mining)

3 Research Questions

I. Is it possible to create your own data set for predicting crop disease?
Ans Yes, Possible with the help of deploying various types of sensors and
drones.
II. Is it possible to use ML/DL techniques to predict crop disease?
Ans Yes, it is possible once proper and accurate data were collected and then
using machine learning algorithm, CNN, Deep learning techniques.

4 Proposed Model

Different raw data acquired through various sensors such as optical sensors, MIX
sensors, motion detector sensors, and others were saved in the database in our
proposed model, as shown in Fig. 1, and after adequate refinement, a data set
was designed for feature selection. The disease was then predicted using proposed
machine learning algorithms (s).
Phase#1: Data collections through various sensors and stores in a data base.

4.1 Optical Sensors

Optical sensors, on the contrary, use light reflection to measure and record crop and
soil data in real-time crops. These crop reflectance sensors typically operate near
IoT-Assisted Crop Monitoring Using Machine Learning Algorithms … 7

Fig. 1 FF model

infrared visible regions of the spectrum and calculate vegetation indices at least by
integrating 02 wavelengths. The visible light has a direct association with the chloro-
phyll concentration from an agronomic standpoint, absorbing red and blue light while
reflecting green light. That is why we see healthy plants as green. Electrochemical
Sensors are a type of sensor that uses electricity to detect a chemical reaction. RVI is
the relationship between these two wavelengths (Ratio Vegetation Index). The NDVI
(Natural Difference Value Index) is a Normalized Difference Vegetation Index and
the suggested normalization ensures that the NDVI values are contained in the same
scale of values, ranging from −1 to 1, as indicated in the equation.

NDVI = (ρ I R − ρV )/(ρ I R + ρV ) (1)

where
ρIR represent infrared reflectance;
and ρV is visible reflectance.
This sensor provides the status required for precision agriculture.
8 S. Kumar Apat et al.

4.2 HTE MIX Sensors

It is utilized to provide moisture content and temperature in the soil as well as the
surrounding environment. It is a typical environment parameter that pops up on a
regular basis, and its handling is crucial in various domains. It is an electric capacity
sensor that connects to a Smart Soil Moisture Sensor’s cell.

4.3 Motion Detector Sensors

All across the field, motion sensors are used. When the recordings are made around
the camp, those sensors can do a server-to-server transaction and then send a message
to each other tool when the data is processed, all while remaining within the farm’s
boundaries. This device can also be used to make noise to scare animals away from
potentially hazardous crops or plants.
Phase#2:
During this phase, the heterogeneous data collected from different sensors were
filtered and made ready for feature selections.
Phase#3:
The process of decreasing no. of input variable(s) during the development process of
a predictive model is said to be as feature selection. Its primary goal is to minimize
the number of input variables in order to reduce overall modeling costs, and in some
cases, it is used to improve performance.
Phase#4:
The following ML algorithms were applied to the selected features to compare and
predict the disease affected to the selected crops.
Phase#5:
Most likely disease will be predicted for the selected crops, and necessary precaution
may be taken to avoid major financial loss to the farmer.

5 Implementation Details

Naïve Bayes: NB classifiers are a subset of fundamental “probabilistic classifiers”

based on the NB theorem and strong feature independence assumptions. They are
among the most basic BN models, but they can achieve higher accuracy when
combined with kernel density estimation. This theorem computes the likelihood
of a subsequent event of the preceding event. Mathematically Bayes’ theorem can
be represented as (Table 2).
IoT-Assisted Crop Monitoring Using Machine Learning Algorithms … 9

Table 2 Crop prediction

Crops Accuracy training set Testing set
using NB
Rice 84.58 86.77
Ragi 90.32 90.12
Gram 76.59 77.77
Potato 88.22 88.19
Onion 79.77 80.25

P( A|B ) = fracP(B|A )P(A) = fracP(B|A ){P(B)} (2)

The above table mentioned the accuracy of the different crops. We have estimated
the accuracy in terms of training as well as testing set and found NB algorithm gives
89.32 training and 90.32 and 90.12 testing accuracy (Fig. 2).
Support Vector Machine (SVM): A supervised ML technique used to solve clas-
sification and regression problems. However, this is preferably used to solve prob-
lems based on categorization types. Each data item is represented as a point in
n-dimensional space (where n is the number of features chosen), with the value of
each feature representing the SVM algorithm’s value of a particular coordinate. After
that, the classification is completed by locating the hyperplane that best distinguishes
the two classes. The following is the hyperplane equation used to divide the points
(for classification):

H : wT (x) + b = 0 (3)

Here: b = the hyper plane equation’s intercept and bias term.

Fig. 2 Categorization using

Naive Bayes
10 S. Kumar Apat et al.

Table 3 Crop prediction

Crops Accuracy training set Testing set
using SVM
Rice 89.78 91.47
Ragi 94.39 93.97
Gram 96.29 95.67
Potato 92.26 90.19
Onion 90.47 72.35

Fig. 3 Categorization using

Support Vector Machine

The hyper plane in D dimensional space is always the D − 1 operator.

A hyper plane is a straight line in 2-D space, for example (1 − D) (Table 3 and
Fig. 3).

6 Conclusion and Future Scope

Machine learning is a useful tool in today’s world for analyzing massive amounts of
data and producing more accurate results and predictions. Our research demonstrated
and provided accurate results for the five crops that were chosen as a sample for yield
prediction. Hence, we were able to confirm that machine learning algorithms may also
be used to predict various illnesses impacting crops over multiple seasons and across
a variety of crops because our training set and testing data are practically identical.
To achieve the best level of classification accuracy, careful selection of preprocessing
data methodologies and machine learning technologies is required. As a result, more
machine learning-based technologies are required to predict various sorts of illnesses
impacting diverse. SVM beats the other two methods, with Gram’s training accuracy
of 96.29% and testing accuracy of 95.67% as prediction is concerned.
 IoT-Assisted Crop Monitoring Using Machine Learning Algorithms … 11

More advanced technologies, such as deep learning and CNN algorithms, will
be used in our future research to detect with more accuracy yield prediction and
agricultural illnesses.

References

1. Yang X, Guo T (2019) Machine learning in plant disease research. Machine learning 34
2. Jawade PB, et al. (2020) Disease prediction of mango crop using machine learning and IoT.
Advances in decision sciences, image processing, security and computer vision. Springer,
Cham, pp 254–260
3. Akulwar P (2020) A recommended system for crop disease detection and yield prediction
using machine learning approach. Recommender system with machine learning and artificial
intelligence: practical tools and applications in medical. Agricultural and other industries, pp
141–163
4. Medar R, Rajpurohit VS, Shweta S (2019) Crop yield prediction using machine learning
techniques. 2019 IEEE 5th international conference for convergence in technology (I2CT).
IEEE
5. Sharma, Bhawana, Jay Kant Pratap Singh Yadav, and Sunita Yadav. (2020) Predict crop produc-
tion in India using machine learning technique: a survey. 2020 8th international conference on
reliability, Infocom technologies and optimization (Trends and future directions) (ICRITO).
IEEE
6. Bondre DA, Mahagaonkar S (2019) Prediction of crop yield and fertilizer recommendation
using machine learning algorithms. Int J Eng Appl Sci Technol 4(5):371–376
7. Kumar YJN, et al (2020) Supervised machine learning approach for crop yield prediction in
agriculture sector. 2020 5th international conference on communication and electronics systems
(ICCES). IEEE
8. Jain A et al (2019) A review of plant leaf fungal diseases and its environment speciation.
Bioengineered 10(1):409–424
9. Kumari P, et al (2020) Predicition of crop yeild using SVM approch with the facility of E-MART
system. Easychair
10. Fenu G, Malloci FM (2021) Forecasting plant and crop disease: an explorative study on current
algorithms. Big Data Cogn Comput 5(1):2
11. Aggarwal N, Singh D (2021) Technology assisted farming: implications of IoT and AI. IOP
Conf Ser: Mater Sci Eng 1022(1). IOP Publishing
12. Mishra D, Deepa D (2021) Automation and integration of growth monitoring in plants (with
disease prediction) and crop prediction. Mater Today: Proc 43:3922–3927
13. Sladojevic S, et al (2016) Deep neural networks based recognition of plant diseases by leaf
image classification. Comput Intell Neurosci 2016
14. Sujatha R et al (2021) Performance of deep learning vs machine learning in plant leaf disease
detection. Microprocess Microsyst 80:103615
15. Mohanty SP, Hughes DP, Salathé M (2016) Using deep learning for image-based plant disease
detection. Front Plant Sci 7:1419
16. Sambasivam G, Opiyo GD (2021) A predictive machine learning application in agriculture:
cassava disease detection and classification with imbalanced dataset using convolution neural
networks. Egypt Inf J 22(1):27–34
17. Ramcharan A, et al (2017) Deep learning for image-based cassava disease detection. Front
Plant Sci 8:1852
18. Ayub U, Moqurrab SA (2018) Predicting crop diseases using data mining approaches: clas-
sification. 2018 1st International conference on power, energy and smart grid (Icpesg).
IEEE

View publication stats