2021 International Conference on Emerging Smart Computing and Informatics (ESCI)

AISSMS Institute of Information Technology, Pune, India. Mar 5-7, 2021

Tomato Leaf Disease Detection and Classification

using Convolution Neural Network
Ananda S. Paymode Shyamsundar P. Magar Vandana B. Malode
2021 International Conference on Emerging Smart Computing and Informatics (ESCI) | 978-1-7281-8519-4/20/$31.00 ©2021 IEEE | DOI: 10.1109/ESCI50559.2021.9397001

Department of Electronics & Department of Computer Department of Electronics &

Telecommunication Engineering Science & Engineering Telecommunication Engineering
MGM’s, Jawaharlal Nehru Engineering TEC, Terna Engineering of College, MGM’s, Jawaharlal Nehru Engineering
College. Nerul, Navi Mumbai College.
Aurangabad, (M.S.), India (M.S.), India Aurangabad, (M.S.), India
[email protected] [email protected] [email protected]

Abstract— Mostly development of country depends of enhancing their operations with this technology, which
growth of agriculture sectors. Now a day agriculture is facing increases efficiency and reduced harm to environment, food
lot of challenges like unavailability of labor, drastic climate contamination and increase the profit.
change, uncertainty in rain, natural disaster, different diseases
on plant leaf and crops, no fixed prices and unavailability of A tomatoes are the important fruits vegetable of every
markets and many more. But as the world continuously human being. Tomatoes are high in many nutrients, and also
increasing the demands of food and more production needed in contain a heart disease and cancer-fighting substance called
next 50 years. There are huge numbers of threats in lycopene. The recent advanced technology Artificial
agriculture field. The use of artificial intelligence technology, intelligence and Deep learning discover new economic
found best for the all agriculture challenges. So our proposed opportunity, knowledge to enhance the quality of tomatoes.
research focus on detecting and classifying the accurate type of The proposed tools and technologies enhance the operation
diseases occurred on leaf at early stage. Our research aims to of tomatoes farming and increase profit which improves
address the problem using the Deep Learning (DL) techniques. production and marketing methods. The advance
The AgroDeep mobile application developed for collection of improvements can be achieved with precisely detection of
real database of agriculture leafs and crop. The real diseased daily changes of tomatoes leafs which would henceforth
leaf images collected and captured through our mobile improve and manage the timing for the harvest, pest control,
application. The captured images uploaded over database.
yield prediction, farm monitoring, disaster warning and
There are total six types of crops leaf images (tomato, grapes,
health of tomatoes. The agriculture research techniques play
soybean, sugarcane, cotton and onion) collected. The tomato
diseased leaf selected for detection and classification. The
important role in the growth of farmer and ultimately help
techniques supported whether diseases affected on leaf or not for development of nation, As India is agriculture country
with percentage of accuracy. The best Convolution Neural and 80% people and business depend on agriculture.
Network (CNN) algorithm suitable for these analysis. The CNN Some machine vision application developed in
based model gave the highest accuracy of 97 % which is agriculture like remote sensing for natural resource
highest forever for real captured diseased images. Our assessments, post harvesting product quality, safety
research playing exquisite role in agriculture sector and
detection, classification, sorting, process automation and
farmers. The proposed research supported to increase food
production in the agriculture. Ultimately it gives more profit in
precision farming [1]. Machine vision predict the size,
the farming sector which motivate the farmers for agriculture. shape, color and texture of object, it mostly support for
numerical attributes of the objects.
Keywords— Deep Learning, Convolution Neural Network Computer Vision play important role in the image
(CNN), Artificial Intelligence, AgroDeep. processing to extract the useful information image and to
I. INTRODUCTION enhance it. The removal of noise are the key steps in image
processing. Noise added in the image during the process of
Farmers are now very well aware of use the engineering storing, transmitting the images [2]. The various noise like
science and technology for farming management. The gaussian, salt ,speckle and pepper etc., to remove these the
agriculture related technological skills required for superior model are require to prevent the image from noise.
agricultures development. Usually crops health’s and The auto-encoder is right solution which can remove the
growth depends on environment condition, which is several kind of noise from images. Auto encoder provide the
probably uncertain at maximum times. Therefore great better accuracy and same compare with the use of PSNR,
initiatives require for fast development agriculture life cycle. SSIM and RMSE [2].
The use of technology always helps in agriculture sectors to
adopt practices that are profitable, environmentally sound, Deep Convolution Neural Network (DCNN)
and contribute to quality of farmer life. However, farmers automatically detect and classify diseases from plant fruits
face unique challenges and require technology and training in apple trees [3]. It also guarantee to find nutritional
to ensure their success. deficiency and examine damage due to herbicides [3]. Some
Machine Learning (ML) model interpret and classify
An Artificial Intelligence and Deep Learning technology disorders in apple leaves and state-of –art in model DCNN
helps farmers to incorporate the latest scientific advances attain very high performance than the human experience [3].
and design tools into their daily operations. The results of

978-1-7281-8519-4/21/$31.00 ©2021 IEEE 564

orized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on January 08,2024 at 05:31:25 UTC from IEEE Xplore. Restrictions ap
 Below figure shows the proposed block diagram of tomato The corn images captured and taken from plant village
leaf disease detection [Fig. 1]. datasets with 3 type namely rust (1558), northern leaf blight
(1377) and Healthy (1447). The corn images infected with
common rust and spots on the leaf surface [6].
Another category and type of images from plant village
pepper datasets. The bell pepper database with two category
healthy and disease. The total number of images are 997
having bacterial spot and 1478 images of health class [7].
More dataset from plant village potato contained the image
sets of early and blight 1000, healthy 152 and late blight
1000 [7]. The following are the different methods of plant
leaf disease detection and techniques used in for achieve
efficient performance using different algorithm.

Fig. 1. Proposed Block Diagram for Tomatoes Leaf Disease Detection A. Detection Using Faster R-CNN
One selective search method required much running time
Image processing in main fundamental aspects of for detections of object location [8]. Faster R-CNN
computer vision. Image processing converts image into overcome this challenge with introducing the Region
digital equivalent from extraction of features get enhanced Proposal Network (RPN), which handshake convolution
the original image version [2]. The main challenges in the features with the classification network and two networks are
image processing is the de-nosing images which removal and joined as one network that can trained and tested by end-to-
suppresses all kinds of noise from noisy image. Noise added end process [4].The use of this process the running time for
in the image due to unwanted signals, varying brightness, region proposal generation take small time and this
colors contrasts, which form tiny speckles, grains or multi- framework can maintain the detection rate time and better
colored pixels [2]. perform the state-of-the-art in object detection accuracy
The detection of tomato leaf disease accurately, fast and using CNN Model [4]. The below table show the category of
reliable with the use of Convolution Neural Network (CNN) dataset with images for training, testing and validation also
is novel approach. The work done has been developed with compared with network model and final columns shown the
CNN in agriculture field for yield detection and automated accuracy performance parameter for each model.
harvesting[4]. A state-of-art in object detection methods use
TABLE I. PLANT VILLAGE TOMATO DATASET
faster Region-based CNN with transfer learning approach.
Faster Region-based CNN is the combination of modalities S.
Accur
Color (RGB) and Near-Infrared (NIR). Category of Datasets Networks/ Model acy
N
(%)
II. RELATED SURVEY (https://challenger.ai/) Deep Convolutional Neural
1 93.71
2462 Images [9] Networks, DenseNet-121
The important objectives of related works to focus on
Images Captured Convolutional Neural
various techniques of disease detection which used for in 2 88.46
Dataset [6] Network
research paper. The plants, crops and fruit datasets with
Plant Village Computer- Multi-Layer Perceptron, and
sample, collection and calibration are most important steps 3 94.35
Assist Diagnosis [10] Support Vector Machine
before applied to model for detection and analysis of images.
The technique like Hyper spectral imaging system used Deep Learning, VGG16,
4 Plant Village [11] 91.2
InceptionV3 & MobileNet
camera (PIKA II, resonon, Inc., Bozeman, MT 59715, USA)
Mobile captured 1747 Computer Vision with Deep
with motion control mechanism and using motor mechanism 5 95.24
On-field Images [12] Learning
images captured. The captured images spectral range,
Convolutional Neural
spectral resolution, spectral bands, frame rate, integration Public Dataset,47363
6 Network, Inception-ResNet- 86.1
time and gain observed and noted for detection and analyzed Images 27 Diseases [13]
v2
using Convolution Neural Network Model [5]. The CNN
Convolutional Neural
always gives the better results for image and speech 7 600 Images [14] 93.7
Network
recognition as compared other deep learning model. As
4923 Images ,Camera- Convolutional Neural
compared to other model and Feed-forward Neural 8 95.75
Assisted Diagnosis [15] Networks, F-RCNN,
Networks, (FCNN) needed less parameters, which more
Small Datasets 124 ,
suitable for disease detection and disorders using hyper 9 Deep & Transfer Learning 95
[16]
spectral imaging [5].
Convolutional Neural
10 3663 GPU Tesla [17] 87.1
There are different dataset available having training, Network
validation and testing sets. The Convolutional Neural 11 14,725 images [18]
StridedNet, LeNet, &
95.4
Network (CNN) and a Multilayer Perceptron model train and VGGNet
optimized using training and validation datasets. The datasets 12 13,842 Sugarcane [19] Deep Learning 95.00
contain the labeled images of five types of diseases which AlexNet, ResNet,
13 Dataset 40 Classes [20] 95.97
always affected the apple fruits at maximum times. Test set GoogLeNet and VGGNet
used to assess the performance classifier was supported for Mobile Devices
Deep Learning, Transfer
classification for of apple disorders. [3] 14 Captured 2,756 93
Learning
Images.[21]

565
orized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on January 08,2024 at 05:31:25 UTC from IEEE Xplore. Restrictions ap
 The machine learning based support vector machine infected by different diseases very rapidly, and if no solution
(SVM) algorithm used for identification of apple leaf at early stage then complete crop get damage within short
disease. The charactertrics of leaf like color, shape and span of time. The below figure [Fig. 2.] show mobile
texture extracted from leaf images. Now support vector application screenshot methods and system for images
machine (SVM) finds the automatic rapid diagnosis of apple collection.
leaf disease [22]. The algorithm is more effective and
feasible to recognition of apple leaf disease, with accuracy is
above 90% on the apple leaf disease images dataset. Image
Dataset contained 2462 images.
B. Artificial Neural Network
Artificial Neural Network is used to find and predict the
nutrition available in the crops. The ANN can predict the
complex mapping when reliable dataset available, same fed
for methods. The input or data like humidity, temperature,
cloud cover and wind direction can be useful for detection,
same data get divided into two sting input and output.
C. Imaging Object Technology (IOT)
Color region and machine vision able to inspect object in Fig. 2. Application for Agriculture on Field Leaf Images Collection
light invisible such as ultraviolet (UV), near-infrared (NIR),
and infrared (IR) [1]. The invisible light region provide the Mobile based application having two way to collect the
object information which is helpful to determining plant images, by open the camera and capture the image or upload
maturity, disease, vegetable variety, ripeness and quality. It the image from phone gallery. There are different category of
also useful in detecting quality and safety such as, defect, tomato leaf disease infected images available in the Plant
functional properties, disease and composition of plant, grain Village Dataset. There are total ten classes of images in the
and nuts, vegetable and fruits[1]. Ultimately Imaging datasets. The class of healthy (1591), spot contaminated
technology help to provide better quality and safety foods. (1404), mosaic virus (373), yellow leaf curl virus (3209),
bacterial spot (2127), early blight (1000),late blight(1909),
D. Wireless Sensor Technology
leaf mold (952), septoria leaf spot (1771) and spider mites(
Wireless Sensor Technologies based application mostly 1676) [7].
suitable for distributed data collecting and monitoring [23].
The WST record the parameters like climate influence, crop
canopy influence, environmental changes, climate
monitoring, precision agriculture, pest control and precision
irrigation [23]. Which provides new features that developed
agriculture economically viable.
E. Convolutional LSTM (Convo LSTM)
Temporal information prediction introduced from Long
Short-Term Memory (LSTM) Network. The idea represent
the sequence data in 1-D space and predicting the future
element based on the stacked LSTM layers. To carry the Fig. 3. Different Class of Tomato Leaf Disease Images
required information which can hold the sequential
information designed are important. [2] The classification method with Enhanced LeNet (ELNet)
F. Autoencoder also provide for leaf tomato disease detection. The total 2850
tomato leaf images taken for classification from Plant village
Autoencoder based on Artificial Neural Networks dataset [24].
(ANN) and machine learning unsupervised techniques use
in it. Unsupervised techniques perform various feature The dataset includes 55,000 leaf images of 14 crops and
detection and classification in the techniques. The auto- 39 healthy and diseased class of data. The only tomato 9
encoder are the feed forward neural networks, where input classed of data out of 1 class data for health leaf. The total
14529 labelled training images and 3631 testing labeled
and output are almost same are shown in the Fig.1. Two
images used for prediction [25]. The dataset with 4923
main encoder i. e. Convolution encoder and Classical
images of tomato plant leaves with 4 category labels. The
encoder, the convolution encoder are more efficient than the labelled images are resize to 256 x 256 pixel as through an
classical encoder. AlexNet architecture [15].
III. DATASETS & METHODS USED There are many Deep Learning (DL) of Convolution
Neural Network (CNN- Based) architecture such as Google
A. Mobile Application Developments
Net, AlexNet, (VGG16), ResNet, SqueezeNet, etc. The
The crucial and important steps objectives are the real different data sets i.e. Plant Village Net, ResNet34,
time dataset collection and preparation of datasets. Our DenseNet-121, ImageNet, real, small and large image
mobile developed application having six category of images datasets are available.
collection support, out of six images, the tomato selected for
analysis and for proposed research works. The tomato leafs

566
orized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on January 08,2024 at 05:31:25 UTC from IEEE Xplore. Restrictions ap
 The AI challenges 2018 competition contain 61 The power of adaptive learning rate find the individual
categories. With the use of annotation tomato images learning rates of each parameters.
separated from entire dataset [26]. The total selected images
are 4187 in number. There are four disease category of B. Data Augmentation Techniques
tomato disease (powdery mildew, blight, leaf mold fungus, The huge volume of datasets increase performance of
and ToMV). The different charactertics of images such as learning algorithm and reduced the over fitting. The
color brightness, ground textures and shading effects. collection of real time datasets as input to trained model is a
difficult and challenging task. Therefore, use of
The below figure [Fig. 4] show the agriculture dataset augmentation techniques increase the change in training data
available which is created database for all the category of at great extents. Deep learning based augmentation
leaf images and tomato leaf is selected of leaf disease. techniques are possible to use for different operation on
images. The image cropping, flipping, rotation,
transformation, noise rejection, color augmentation are the
important steps. The new approach of neural style transfer
(NST) and generative adversarial networks (GANs)
supporting for image processing [27].

Fig. 4. Real Agriculture Leaf Image Datasets Database

Deep Learning based algorithm DCGAN and WGAN is

state of-art for image identification. The techniques such as
VGG16, ResNet and InceptionV3 mostly used for building
the strong system for leaf disease analysis [27] Fig. 5. Image Augmentation 1. Reshape, 2. Contrast Stretching, 3.
Histogram Equitation, 4. Adaptive Histogram Equalization
The table shows plant village dataset available with
different class of leaf diseases infected images. The images collected through Agro Deep mobile
application over the database. The dataset are limited in size
TABLE II. PLANT VILLAGE TOMATO DATASET
of images, So with available images of tomato diseased leaf
Class/ Category Images augmented and prepared dataset for training and testing. The
Healthy 1591 data augmentation techniques to improve the performance
Spot contaminated 1404 measured parameter better. Deep Convolutional Generative
Adversarial Networks, Wasserstein GAN, and Neural Style
Mosaic virus 373
Transfer are the most common deep learning based data
Yellow leaf curl virus 3209 augmentation techniques. The DCGAN and WGAN trained
Bacterial spot 2127 with NVIDIA DGX-1 V100 deep learning and epoch 1000
Early blight 1000 with batch size 64 [28] .
Late blight 1909 The class imbalance problem handle with data
Leaf mold 952 argumentation techniques. The CNN with three convolution
Septoria leaf spot 1771 layers and max pooling layers with filter applied each layer
Spider mites 1676 [11] .

In the table ten different classes of tomato images, the

total 12000 thousand images. The training dataset prepared
with the available images. The dataset split into train and test
set. The parameter like epochs, steps size, batch size, width,
height and depth consider for training the model. The
activation function rectified linear activation function (ReLu)
and softmax. The activation function added into the neural
network, to learn the complex pattern from images. The non-
linearity of images increased with activation functions. The
image contains lot of non-linear features like color, pixel and
borders. The Relu function activate all neuron not the same
time. The ReLu easily back-propagate the errors. The
adaptive learning optimizer train the deep neural networks.
Fig. 6. Different methods Image Augmentations

567
orized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on January 08,2024 at 05:31:25 UTC from IEEE Xplore. Restrictions ap
 The image augmentation techniques gives better B. Performance Parameter
performance than original dataset. The classification The important performance parameter measured for
accuracy of pre-trained model for VGG16, ResNet, and images segmentation are the F1-Score, accuracy metrics
InceptionsV3 are 90.5%, 89.5% and 91.5 respectively. The
(AC) and receiver operating characteristic curve (ROC), The
combined dataset performance measured higher than the
classification evaluation performance techniques are the
original dataset [27] .
area under the curve (AUC) better the AUC higher the
classification performance [7].

Ac = (TP + TN) / (TP + FP + FN + TN) (1)

Where, TP is Positive Samples, TN Negative Samples, FN

False Negative and FP. The classifier performance measure
using evaluation metrics, are gives as;

(2)

Fig. 7. Augmenting Images of Different Sizes

(3)
IV. EXPERIMENTAL RESULT & DISCUSSION
False Positive, TNR True Negative Rate, and FPR False
A. Convolution Neural Network Model Positive Rate.
Convolution Neural Network is a Feedforward Neural
network, which is always better in the field of image (4)
recognition. Networks consists of three layers i.e.
convolution layers, pooling layers, and fully connected layers
[5].The batch normalization algorithm used to normalize (5)
input of each layers to make and ensure that the input data
distribution of each layer is stable. In the process of training Where m is number of classes, G-means accuracy ratio of
network, training updated parameter caused the distribution TNR and FPR.
of input data of each layer changed constantly, which adapt
new data distribution [5]. Below figure shows the proposed The Algorithm evaluation standard used is mean average
CNN model for tomato leaf disease detection and precision(mAP) which consist of precision, recall and mean,
classification. The mAP are the important for image processing and
detection. The accuracy is correct classified form the total
results. The recall is correct classified form the total
results.[26]

(6)

ROC : The receiver operating characteristic diagnosis

Fig. 8. Proposed Convolution Neural Network (CNN) Structure for
Tomato Leaf Disease Detection
tools helps for interpretation of probabilistic diagnostic for
binary classification predictive modeling problem are the
In the CNN, Convolution layer learn the input features, ROC curve. The ROC curve is plot between false positive
which consists of several features maps. The pooling layer is (FP) versus true positive (TP) at various threshold level.
designed to control the spatial information of data. The
pooling layer always available in between two convolution
layers. The input data passes through several convolution and
pooling layer, then one or more fully connected layers are to
be set. Fully connected layer acts as classifier in the complete
networks. The CNN model sharing weight local connected
neural networks. The upper layers neurons only sparsely
connected to the lower layers neurons, some neurons share
the weight and biases. In this way the reduce the number of
training parameters and get the simplified network structures
and extraction of more abstract feature is possible [5]. Fig. 9. Receiver Operating Characteristics FP versus TP

568
orized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on January 08,2024 at 05:31:25 UTC from IEEE Xplore. Restrictions ap
 C. Figures and Tables REFERENCES
The model train and test with following parameter. The [1] Y. R. Chen, K. Chao, and M. S. Kim, “Machine vision technology for
changes in the value gives better accuracy with fine tuning agricultural applications,” Comput. Electron. Agric., vol. 36, no. 2–3,
pp. 173–191, 2002.
the parameter EPOCHS = 25, STEPS = 100, STEPS = 100,
LR = 1e-3 BATCH_SIZE = 32, WIDTH = 256, [2] S. K. Vipparthi and B. Raman, Computer Vision and Image
Processing. 2019.
HEIGHT = 256, DEPTH = 3. Table 3 show comparative for
[3] L. G. Nachtigall and R. M. Araujo, “Classification of Apple Tree
different dataset & performance measure, algorithm and Disorders Using Convolutional Neural Networks,” pp. 472–476,
parameter value. The two performance parameter accuracy 2016.
(AC) and area under the curve (AuC). [4] I. Sa, Z. Ge, F. Dayoub, B. Upcroft, T. Perez, and C. McCool,
“Deepfruits: A fruit detection system using deep neural networks,”
TABLE III. COMPARATIVE PARAMETER WITH PROPOSED SYSTEM Sensors (Switzerland), vol. 16, no. 8, 2016.
[5] W. Yang, C. Yang, Z. Hao, C. Xie, and M. Li, “Diagnosis of Plant
Tomato Leaf Disease Analysis Cold Damage Based on Hyperspectral Imaging and Convolutional
Datasets Performance Algorithm Parameter value Neural Network,” IEEE Access, vol. 7, pp. 118239–118248, 2019.
Measures (%)
[6] S. Mishra, R. Sachan, and D. Rajpal, “Deep Convolutional Neural
Plant Village [10] Ac SVM 0.918 Network based Detection System for Real-time Corn Plant Disease
Recognition,” Procedia Comput. Sci., vol. 167, pp. 2003–2010, 2020.
AUC 0.933
[7] Y. Kurmi, S. Gangwar, D. Agrawal, S. Kumar, and H. Shanker, “Leaf
AI Challenger [26] (mAP) FRCNN 0.958 image analysis-based crop diseases classification,” Signal, Image
Plant Village [24] Ac DCNN 0.920 Video Process., 2020.
Plant Village [25] Ac CNN 0.964 [8] J. R. R. Uijlings, K. E. A. Van De Sande, T. Gevers, and A. W. M.
Smeulders, “Selective search for object recognition,” Int. J. Comput.
Plant Village [11] Ac CNN 0.912 Vis., vol. 104, no. 2, pp. 154–171, 2013.
Real Dataset [15] Ac FR-CNN 0.957 [9] Y. Zhong and M. Zhao, “Research on deep learning in apple leaf
disease recognition,” Comput. Electron. Agric., vol. 168, no. October
Proposed Dataset Ac CNN 0.970
2019, p. 105146, 2020.
Accuracy (Ac), Area under Curve (AuC), Mean Average Precision ((mAP)), The bold face values [10] Y. Kurmi, S. Gangwar, D. Agrawal, S. Kumar, and H. S. Srivastava,
show the high-performance measures of real Agro Deep Dataset
“Leaf image analysis-based crop diseases classification,” Signal,
Image Video Process., 2020.
The accuracy of with image augmentation is 97% which [11] M. Agarwal, A. Singh, S. Arjaria, A. Sinha, and S. Gupta, “ToLeD:
Tomato Leaf Disease Detection using Convolution Neural Network,”
is very higher than the original datasets. Procedia Comput. Sci., vol. 167, no. 2019, pp. 293–301, 2020.
[12] J. G. M. Esgario, R. A. Krohling, and J. A. Ventura, “Deep learning
V. CONCLUSION for classi fi cation and severity estimation of co ff ee leaf biotic
In our research paper, we have developed the mobile stress,” vol. 169, no. December 2019, 2020.
application for collections of original images of crops leaf. [13] Y. Ai, C. Sun, J. Tie, and X. Cai, “Research on Recognition Model of
Total six types of crops leaf images collected and prepared Crop Diseases and Insect Pests Based on Deep Learning in Harsh
Environments,” IEEE Access, vol. 8, pp. 171686–171693, 2020.
the dataset for tomato leafs. With use of data augmentation
[14] A. Batool, S. B. Hyder, A. Rahim, N. Waheed, M. A. Asghar, and
techniques, the dataset prepared and applied to Convolution Fawad, “Classification and Identification of Tomato Leaf Disease
Neural Network (CNN) based model. The method improved Using Deep Neural Network,” in 2020 International Conference on
the performance measured. The evaluation metrics parameter Engineering and Emerging Technologies, ICEET 2020, 2020.
accuracy increased as compared to others available datasets. [15] R. G. De Luna, E. P. Dadios, and A. A. Bandala, “Automated Image
So our research work increased accuracy is 97 %, which is Capturing System for Deep Learning-based Tomato Plant Leaf
higher with compared datasets. The better accuracy for Disease Detection and Recognition,” IEEE Reg. 10 Annu. Int. Conf.
Proceedings/TENCON, vol. 2018-Octob, no. October 2018, pp.
disease detection of leaf is crucial step in the agriculture 1414–1419, 2019.
development. The major focused of research for providing
[16] S. Coulibaly, B. Kamsu-Foguem, D. Kamissoko, and D. Traore,
the advancement in agriculture sector and increase in food “Deep neural networks with transfer learning in millet crop images,”
production. The collection of all six images and prepared the Comput. Ind., vol. 108, 2019.
real datasets is targeted work in future. The preparation of [17] M. Francis and C. Deisy, “Disease Detection and Classification in
real dataset and developed the model for agriculture field is Agricultural Plants Using Convolutional Neural Networks - A Visual
the next step approach. The future work support and Understanding,” in 2019 6th International Conference on Signal
motivate the farmer for farming, which ultimately increase Processing and Integrated Networks, SPIN 2019, 2019.
the profits of agriculture and same helps for strong nations [18] S. V. Militante and B. D. Gerardo, “Detecting Sugarcane Diseases
through Adaptive Deep Learning Models of Convolutional Neural
building. Network,” in ICETAS 2019 - 2019 6th IEEE International Conference
on Engineering, Technologies and Applied Sciences, 2019.
ACKNOWLEDGMENT
[19] S. V. Militante, B. D. Gerardo, and R. P. Medina, “Sugarcane Disease
We are thankful to farmers of nashik & Aurangabad Recognition using Deep Learning,” in 2019 IEEE Eurasia
district, MH [India] for supported to collect the real disease Conference on IOT, Communication and Engineering, ECICE 2019,
2019.
leaf images. The valuable guidance to research for Dr.
[20] K. Thenmozhi and U. Srinivasulu Reddy, “Crop pest classification
Panjabrao Deshmukh Krushi Vidyapeet, Akola always based on deep convolutional neural network and transfer learning,”
support and motivates for this works. Finally thanks to all Comput. Electron. Agric., vol. 164, 2019.
colleagues, family member, friends for co-operation and [21] A. Ramcharan, K. Baranowski, P. McCloskey, B. Ahmed, J. Legg,
passion. and D. P. Hughes, “Deep learning for image-based cassava disease
detection,” Front. Plant Sci., vol. 8, 2017.

569
orized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on January 08,2024 at 05:31:25 UTC from IEEE Xplore. Restrictions ap
 [22] Y. Zhong and M. Zhao, “Research on deep learning in apple leaf [26] Y. Zhang, C. Song, and D. Zhang, “Deep Learning-Based Object
disease recognition,” Comput. Electron. Agric., vol. 168, 2020. Detection Improvement for Tomato Disease,” IEEE Access, vol. 8,
[23] L. Ruiz-Garcia, L. Lunadei, P. Barreiro, and J. I. Robla, “A review of 2020.
wireless sensor technologies and applications in agriculture and food [27] J. Arun Pandian, G. Geetharamani, and B. Annette, “Data
industry: State of the art and current trends,” Sensors (Switzerland), Augmentation on Plant Leaf Disease Image Dataset Using Image
vol. 9, no. 6, pp. 4728–4750, 2009. Manipulation and Deep Learning Techniques,” Proc. 2019 IEEE 9th
[24] “Enhanced Convolution Neural Network for Tomato Leaf Disease Int. Conf. Adv. Comput. IACC 2019, pp. 199–204, 2019.
Classification,” Int. J. Recent Technol. Eng., vol. 8, no. 6, 2020. [28] J. Arun Pandian, G. Geetharamani, and B. Annette, “Data
[25] M. Agarwal, S. K. Gupta, and K. K. Biswas, “Development of Augmentation on Plant Leaf Disease Image Dataset Using Image
Efficient CNN model for Tomato crop disease identification,” Manipulation and Deep Learning Techniques,” in Proceedings of the
Sustain. Comput. Informatics Syst., vol. 28, p. 100407, 2020. 2019 IEEE 9th International Conference on Advanced Computing,
IACC 2019, 2019.

570
orized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on January 08,2024 at 05:31:25 UTC from IEEE Xplore. Restrictions ap