2022 10th International Conference on Emerging Trends in Engineering and Technology - Signal and Information Processing (ICETET-SIP-22) | 978-1-6654-6741-4/22/$31.

00 ©2022 IEEE | DOI: 10.1109/ICETET-SIP-2254415.2022.9791751

2022 10th IEEE International Conference on Emerging Trends in Engineering &

Technology Signal and Information Processing (ICETET-SIP-22)

Internet of Thing and Machine Learning Approach for Agricultural

Application: A Review

Kamlesh Kalbande
Pranita Kolhe Atul Deshmukh
Assistant Professor
PG Scholar Electronics and Telecommunication
Assistant Professor
Electronics and Telecommunication Engineering Electronics and Telecommunication
Engineering G H Raisoni College of Engineering Engineering
G H Raisoni College of Engineering Nagpur-440016, India G H Raisoni College of Engineering
Nagpur-440016, India [email protected] Nagpur-440016, India
[email protected] [email protected]
soni.net

Abstract— Plants provide a significant portion of the global food producer of pulses, rice, wheat, spices, and spice products in
supply. Plant infections are a factor in productivity loss, although the market. The quality of the things that farmers produce
they can be avoided with constant observation. Plant illness determines their economic expansion, which has an impact on
detection by hand is time-consuming and a mistake. Earlier the farmer's ecology.
diagnosis of fungal pathogens with computer vision and artificial The adoption of an automatic disease detection
intelligence (AI) can assist decrease illness severity and
technique is useful for detecting a plant illness at its earliest
overcoming the limitations of continuous human observation.
There have been significant advances in the creation of plant stage. The quality of the things that farmers produce
illness identification, and classifying systems based on “Machine determines their economic expansion, which has an impact on
Learning (ML) and Deep Learning (DL)” models, with a sample the farmer's ecology. As a result, computer approaches must
of assessment of several plant leaf illness classification techniques be developed to enable early diagnosis, and categorization
displayed with tables. In this article, we conduct a comprehensive utilizing leaves pictures automated, and time-consuming.
literature review on the applications of state-of-the-art Given the difficulties, plant illness identification is still a hot
Machine, and Deep Learning techniques for plant disease topic of study. Over the years, a variety of ways have been
categorization, including Support Vector Machine, Convolutional offered. Diseases in crops are fairly common as a result of
Neural Network, K-Nearest Neighbor, Naive Bayes, and other
changing climatic and environmental circumstances.
prevalent Machine learning algorithms, as well as AlexNet,
GoogleNet, VGGNet, and other prevalent Deep learning Infections wreak havoc on agricultural development and
techniques. Every algorithm can be defined by the processing productivity, and they're notoriously tough to eradicate. It is
techniques used, such as image segmentation and extraction and vital to have effective disease detection and treatment
classification, as well as the standardized experimental-setup activities to prevent them in time to order to assure good
metrics used, such as the total number of training/testing datasets quality and high productivity. There are many various kinds of
used, the number of diseases considered, the category of crops in India, and they can all be damaged by diseases that
classification model used, and the classifier performance. harm the leaves, stems, and fruits. Fungi, bacteria, and viruses
produce leaf illnesses, which are the early signs.
Keywords—Machine Learning (ML), IoT, Greenhouse, Plant
diseases, Convolution neural network(CNN). II. PLANT DISEASE CATEGORIZATION
CLASSIFICATION PRACTICE

I. INTRODUCTION
Traditional systems technique for disease detection and
differentiation in plants may be achieved using deep learning
Agricultural production is a significant part of the Indian with CNN and ALDD (orange Leaf Disease dataset) for real-
economy. The Indian agriculture sector approximately half of time detection of apple leaf diseases using deep learning
the country's workforce is employed by it. India is the greatest approach based on upgraded convolution neural networks,

978-1-6654-6741-4/22/$31.00©2022 IEEE

Authorized licensed use limited to: UNIVERSIDAD TECNICA DEL NORTE. Downloaded on October 27,2022 at 17:31:09 UTC from IEEE Xplore. Restrictions apply.
which has got an accuracy of 78.80% & 23.13FPS. Another techniques & implemented with an leaves were taken in a
approach is Support Vector Machine (SVM) algorithms [2] pattern recognition SVM classifier which controlled
like LSVM, QSVM, CKNN, ESD (Ensemble Subspace [4]. gives a 90% atmosphere.
Discriminative) are used the approach was used to classify identification rate
grape leaf illnesses, as well as the efficiency, ranging from 90
BRBFNN Received high Worked only on
percent to 92 percent depending on the type and stage of the (Bacterial Foraging accuracy in identified fungal diseases, not
disease. A clustering strategy was performed with K-means optimization are fungal diseases like on bacteria/viruses
and focused on leaf pictures and employed CNNs as a method common rust, cedar
based on radial
for computerized identification and classification of plant apple rust, late blight,
basis function leaf curl, leaf spot,
diseases. There were ten secret layers in CNN. The number of Neural n/W) is early blight, etc.
results was six, which corresponded to the number of classes used for Vpc- 0.8621/0.8307
representing five disorders and a healthy leaf. identification & Vpe- 0.1118/ 0.1527
The article highlights the several Machine Learning and Classification [5].
Deep Learning techniques utilized in the construction of the
plant disease classification system, as well as the results and Plant disease Corn leaf samples are Overfitting problem
recognition is used. appears for random
limitations for various plant leaf diseases. Table 1 shows the applied to plant Achieved Accuracy: noise or errors
related previous research work of plant leaf disease detection. pathology using 76% for original
It is concluded based on the literature survey carried out on Deep Learning with images
that machine learning technique uses machine learning and CNN & 79% for background
GoogLeNet [6]. removed images
deep learning techniques suffers from slow convergence
problem. All these modifications are made to improve the
Nutrient detection, Nutritious To eliminate
convergence rate and generalization ability of machine
quantification & deficiencies in plants, inaccurate labels, the
learning. classification using plant color & annotating process
digital images & morphology can be must be enhanced.
Table 1: Summary of Related Work ML [7]. detected.

Methodology Used Parameter Drawbacks Summarizes the The quality of Problems of

considered pros & cons for training data has a overfitting or
plant disease significant impact on overtraining occurred
Deep learning with GoogLeNet Low recognition detection with the the system's with NN, SVM & GA
CNN & ALDD Inception & Apple accuracy for CV) approach [8]. efficiency.
(Apple Leaf disease leaf detection is Alternaria leaf spot
dataset) [1]. done via rainbow & Grey spot Deep CNN with Cucumber diseases Degradation
concatenation, conventional are identified using inaccuracy for
received accuracy classifiers like leaf symptom images unbalanced data.
of 78.80% & Random Forest, with DCNN, which
23.13FPS SVM, AlexNet is shows an accuracy of
used [9]. 93.41%
Multiclassifiers like Black rot, black 1. I/P images are low
LSVM, QSVM, Measles, and Leaf contrast & noisy GPDCNN (Global Found higher Shows lesion
CKNN, ESD Blight are examples which affect
Pooling Dilated) recognition rate & detection failure.
of grape leaf segmentation.
(Ensemble along with AlexNet learning rate for
diseases that can be 2. Degrade accuracy
Subspace detected. is used [10]. cucumber leaf
for complex images
Discriminative) are Segmentation was diseases.
used [2]. achieved. Its
accuracy of 90% & ML with annotation A robusta coffee leaf No. of samples &
Classification techniques like Rust infection level accuracy is less
accuracy of 92% object (1-4) identification is
segmentation & carried out using
A statistical method Morphological PCA shows high classification is OIRSA (Organismo
like PCA & Cluster Char’s are identified variation among used [11]. International
analysis is used [3]. for pomegranate by genotypes of the total Regional Sanded
classifying variance. Agropecuaria)
Germplasm method.

Image processing GA-CFS is Images of apple Unsupervised Tea leaf disease This model is
(anthracnose) is

Authorized licensed use limited to: UNIVERSIDAD TECNICA DEL NORTE. Downloaded on October 27,2022 at 17:31:09 UTC from IEEE Xplore. Restrictions apply.
 classification & detected. unstable for different (Speech-toText), for added security. sciences.
adaptive 2D Accuracy: leaves GUI (Graphical
thresholding is Pixel level- 94% User Interface)
Leaf level-98%
developed for (Graphic User
hyper spectralngg Interface) [19].
imaging [12].
Data Analytic, App for real-time It can only be
SVM is used along VGG16 DL model is Identification Prediction, Smart temperature and detected and is
with C-DCGAN used which gives accuracy of images is farming[20]. humidity display and unrelated to the
(Conditional deep identification affected with monitoring. temperature changes
convolution accuracy of 90% VGG16. that trigger the alert.
generative
adversarial n/w)
[13].

Unsupervised deep Weed (grass) & Only when the cluster III. CHALLENGES
clustering broadleaf labeling has considerably
algorithm & image with 87% of more items of a
clustering [14]. accuracy. specific class can it • Quality of plant leaves images.
be labeled.
• Needs a large amount of data set for Training &
CNN [15] DL n/w with 5 stages The stochastic Testing.
is designed which approach is used not
gives a recognition focused on leaf • Acquired images may affect by background data and
rate of 93.09% diseases. noises.
NodeMCU Module Traditional pesticide It covers big area • Identification of the exact diseased spot in a leaf is
offers a variety of sprayers relay on
pesticide sprayers human effort to difficult.
[16]. operate.
• Depending on environmental conditions; the colour
The suggested As a result, the Its occupy more size & texture of the leaf may vary.
system also intended system electricity for cloud-
includes implementation was hosted database • Regular observation is needed for a particular
NodeMCU, a 4- tested, with relay category of plants leaf.
channel relay switching as an
module, and the output from
Blynk android instructions supplied
application. [17]. via the control unit's
IV. PROPOSED METHODOLOGY
built-in IoT cloud
interface.
After studying the journal paper, our discovery came
Data Science, Accuracy of Machine As a result, across different researches used for plant leaf disease
Machine Learning Learning Algorithms environmental and detection. Each methodology has its benefits. Major
(ML), Machine (100 percent ) 94.87 genetic factors rise in techniques are briefed in the following section.
Learning percent for Decision importance with
Algorithms, Tree, 89.00 percent time..
Parkinson's Disease for Logistic
(PD), Data-set, and Regression, 87.17
Accuracy of percent for K-Nearest
Algorithms [18]. Neighbors, and 82.05
percent for Support
Vector Machine.

CNN, ANN The Google It makes use of voice

(Artificial Neural Speechtotext API recognition, which is
Network), ASR converts speech into an important task in
(Artificial Speech pass code, which is the biological
Recognition), STT then cross-verified

Authorized licensed use limited to: UNIVERSIDAD TECNICA DEL NORTE. Downloaded on October 27,2022 at 17:31:09 UTC from IEEE Xplore. Restrictions apply.
 Android mobile: It will create a mobile app, through which we
will control the robot[20].

Table 2: Comparison Of Different Algorithms

Specificatio Teachable MobileNet Convolution Neural

n Machine Network
Defination Teachable MobileNet is a Convolutional neural
Machine is a kind of CNN networks (CNNs) are
web-based designed for neural networks with
application that mobile and one or more
allows embedded convolutional layers
Fig1: Architecture of Proposed Model for plant leaf disease detection developing machine that are commonly used
machine vision for image processing,
learning applications. It classification,
models quick, is based on an segmentation, and other
In the proposed system, the image segmentation algorithm simple, and optimized auto-correlated
is designed to capture, identify and clarify the reference accessible to architecture. data[19].
images of plants. The segmentation algorithm performs anyone.
operation by the captured frames and then detects the diseases.
First of all it will give input of different plants. Then the Frame Rate As it is frame Changeable Changeable frame rate.
rate which we frame rate.
raspberry pi camera is used for capturing the video of tomato have.
plant, and pi camera is connected to the raspberry pi on-board.
The video is captured at the frame rate of 24fps. Each frame is Resize and Not required Must required Required.
then processed to plant any frame containing the reference Rename
image of the plant. In above figure, where all the devices are
connected to the raspberry pi and it will also detect the tomato Data All dataset are Only 5000 Only 200 images are
Required used. images are used.
plant disease[18]. used.
The Raspberry Pi is connected to all of the gadgets on-
board. It is the gadget around which the entire project Dataset Dataset are not Dataset are Dataset are given to the
revolves, and it is a little computer with a wide range of given to the given to the cloud
cloud cloud
functions. The raspberry pi is further connected to the DC
motor and relay driver. DC motor driver act as an interface Accuracy Accuracy Accuracy = Accuracy = 95%,
between the motors and control circuit. It will be necessary to =91%, 64%, 92%, 85%, 92%, 90%........etc.
create an Android app that can be used to control the 62%, 25%..... 98%,……etc.
robot[17]. After interfacing the relay driver acts as a switch etc.
which can open or close. In this module led is also placed to Application It is very easy Reduced Natural language and
indicate status of a relay switch. If the tomato plant disease is to used. network size facial recognition text
and reduced digitisation
detected then the pesticide tank will sprinkle the pesticides in number of processing.
different plants[16]. The final system is also capable of parameters.
avoiding the obstacle with the help of ultrasonic sensors. Advantages They perform Weight sharing, high
It's simple to better and are accuracy in image
train and more suitable recognition
Input: It will generate a video of a different plant in the input for mobile difficulties
deploy. The
and then remove its image. model was apps. Low-
Camera: The pi camera is used to capture the video frames at trained in a latency, small
the rate of 30fps. The captured frames are then given to the short amount CNN.
raspberry pi for processing. of time and is
open source,
Raspberry pi: The Raspberry pi is used as the processing unit allowing us to
for the complete image segmentation algorithm such that each experiment
frame is given to it and applies to the image segmentation with a variety
algorithm. of classes.
Disadvantag Changing the Large amounts of
Dc motor: Two 10 rpm DC gear motors are used in the system e background / training data are
to run the autonomous vehicle setting, When required, and the
MobileNet
Pesticides: Pesticides of various kinds are put on the plants in it comes to object's position and
framing your has reduced orientation are not
this tank.

Authorized licensed use limited to: UNIVERSIDAD TECNICA DEL NORTE. Downloaded on October 27,2022 at 17:31:09 UTC from IEEE Xplore. Restrictions apply.
 examples, size, reduced encoded. [2] Alishba Adeela, Muhammad Attique Khan, Muhammad
Changing parameters. Sharifa, Faisal Azama, Jamal Hussain Shaha, Tariq Umera,
microphones
and locations Shaohua Wanc, “Diagnosis and recognition of grape leaf
Taking audio diseases: An automated system based on a novel saliency
samples with a approach and canonical correlation analysis based multiple
lot of noise, features fusion”, Sustainable Computing: Informatics and
Becoming
aware of your Systems 2210-5379/© 2019 Elsevier Inc.
own bias. [3] Ali Khadivi, Daneyal Ayenehkar, Milad Kazemi, Alireza,
Result
In teachable In MobileNet In convolution neural
“Phenotypic and pomological characterization of a
machine the the accuracy network the accuracy pomegranate (Punica granatum L.) germplasm collection and
accuracy of level is very level is above 90%. It identification of the promising selections”, Elsevier Scientia
diseases is accurate, but will not vary in Horticulturae 238 (2018) 234–245
good, but not it working different
showing in process is diseases[19].
[4] Zhang Chuanlei, Zhang Shanwen, Yang Jucheng, Shi
continuous very lengthy. Yancui, Chen Jia, “Apple leaf disease identification using
manner. It will genetic algorithm and correlation-based feature selection
be vary in method”, International Journal of Agriculture & Biol Eng Vol.
different
diseases.
10 No.2 March 2017
[5] Siddharth Singh Chouhan, Ajay Kaul, Uday Pratap Singh,
Sanjeev Jain, “Bacterial Foraging Optimization Based Radial
Basis Function Neural Network (BRBFNN) for Identification
and Classification of Plant Leaf Diseases: An Automatic
V. CONCLUSIONS
Approach Towards Plant Pathology”, 2169-3536 2018 IEEE
Translations
This study compares and contrasts several machine [6] Jayme G.A. Barbedo, “Factors influencing the use of deep
learning and deep learning techniques for identifying and learning for plant disease recognition” Science Direct
categorizing plant leaf diseases. Furthermore, the foregoing Biosystem Engg, 2018 Published by Elsevier
are some prospective areas for additional research: It would be [7] Jayme Garcia Arnal Barbedo, “Detection of nutrition
crucial to be able to diagnose a specific stage of a plant leaf deficiencies in plants using proximal images and machine
disease. The analysis could be used to limit the number of learning: A review”, Computers and Electronics in Agriculture
compounds used in precise identification. Most plant diseases 162 (2019) 482–492 2019 Published by Elsevier
should be identified and classified using an online platform. [8] Sukhvir Kaur, Shreelekha Pandey, Shivani Goel, “Plants
More study research is required to examine the backside of the Disease Identification and Classification Through Leaf
leaf to output the examination on the backside of the leaf in Images: A Survey”, Published by Springer Archives of
order to establish a method to detect an infected leaf owing to Computational Methods in Engineering (2019) 26:507–530
nutrient insufficiency. For effective identification of illnesses [9] Juncheng Maa, Keming Dua, Feixiang Zheng, Lingxian
of the plant leaf, real-time photos should be employed. It is Zhang, Zhihong Gong, Zhongfu Suna, “A recognition method
possible to take images in a variety of lighting settings. It is for cucumber diseases using leaf symptom images based on
capable of measuring the severity of diseases that have been the deep convolutional neural network”, Published by Elsevier
discovered automatically. Further study initiatives are required Computers and Electronics in Agriculture 154 (2018) 18–24
to identify plant diseases of many other sections, such as [10] Shanwen Zhanga, Subing Zhang, Chuanlei Zhang,
stems, in order to design a system to extract the necessary Xianfeng Wanga, Yun Shia, “Cucumber leaf disease
damaged area from the image's complicated background. The identification with global pooling dilated convolutional neural
pesticide step involves identifying more investigation. network”, Published by Elsevier Computers and Electronics in
Agriculture 162 (2019) 422–430
[11] Jorge Parraga-Alava, Kevin Cosme, Ang_elica Loor,
Esneider Santander, “RoCoLe: A robusta coffee leaf images
VI. REFERENCES
dataset for evaluation of machine learning-based methods in
plant diseases recognition”, Published by Elsevier Data in
[1] Peng Jiang, Yuehan Chen, Bin Liu, Dongjian he, brief 25 (2019) 104414
Chunquan Liang, “Real-Time Detection of Apple Leaf [12] Lin yuana, Peng Yang, Wenyan Hanb, Yanbo Huang, Bin
Diseases Using Deep Learning Approach Based on Improved Wangd, Jingcheng Zhang, Haibo Zhanga, Zhiyan Baoa,
Convolutional Neural Networks", IEEE journal Vol. 7, “Detection of anthracnose in tea plants based on hyperspectral
Access. 2914929 2019. imaging”, Published by Elsevier, Computers, and Electronics
in Agriculture 167 (2019) 105039

Authorized licensed use limited to: UNIVERSIDAD TECNICA DEL NORTE. Downloaded on October 27,2022 at 17:31:09 UTC from IEEE Xplore. Restrictions apply.
[13] Hu Genshenga, Wu Haoyua, Zhang Yana, Wan [17] Rishi Mathur, Kamlesh Kalbande, “Internet of Things
Mingzhub, “A low shot learning method for tea leaf’s disease (IoT) based Energy Tracking and Bill Estimation System”
identification”, Published by Elsevier, Computers, and Publish by Fourth International Conference on I-SMAC ISBN:
Electronics in Agriculture 163 (2019) 104852 978-1-7281-5464-0.
[14] Alessandro dos Santos Ferreira, Daniel Matte Freitas, [18] Surekha Tadse, Muskan Jain, Pankaj Chandankhede,
Garcinia Gonçalves da Silva, Hemerson Pistorib, Marcelo “Parkinson’s Detection Using Machine Learning”, Publish by
Theophilo Folhesc, “Unsupervised deep learning and semi- International Conference on Intelligent Computing and
automatic data labeling in weed Discrimination”, Published by Control Systems (ICICCS 2021) ISBN: 978-0-7381-1327-2.
Elsevier, Computers, and Electronics in Agriculture 165 [19] Pankaj H. Chandankhede, Abhijit S. Titarmare, Sarang
(2019) 104963 Chauhvan, “Voice Recognition Based Security System Using
[15] Vinit Bodhwania, D. P. Acharjyaa, Umesh Bodhwania, Convolutional Neural Network”, Publish by 2021
“Deep Residual Networks for Plant Identification”, Published International Conference on Computing, Communication, and
by Elsevier, Procedia Computer Science 152 (2019) 186–194 Intelligent Systems (ICCCIS).
[16] Swapna Choudhary, Kamlesh Kalbande, Nitin Dhote, [20] Mukul Kolhe, Roshan Paturkar, Umesh Sahu, Sofia
“IoT based Multi-point Pesticide Spraying Machine”, Publish Pillai, Abhijit Titarmare, “Analytic for Temperature and
by Sixth International Conference on Inventive Computation Humidity – Cloud based Forecasting and Dashboard”, Publish
Technologies [ICICT 2021]. by, International Conference on Intelligent Computing and
Control Systems (ICICCS 2020) ISBN: 978-1-7281-4876-2.

Authorized licensed use limited to: UNIVERSIDAD TECNICA DEL NORTE. Downloaded on October 27,2022 at 17:31:09 UTC from IEEE Xplore. Restrictions apply.