JOURNAL OF ALGEBRAIC STATISTICS

Volume 13, No. 3, 2022, p. 752-763

Received 2022 March 15; Revised 2022 April 20; Accepted 2022 May 10

Abstract:

In India, the main source of income for healthy lifestyle is agriculture and it is occupying 70% of rural population. Crop
cultivation of India is highly miscellaneous. In India, the varieties crop is about 500 types. Even though there is
advancement in technology, the practices of agriculture are still manually preceded and low automation level is involved
as compared to the western countries. Plants are affected by many diseases and so the leaves are get damaged. From the
images of leaves, the affected plant will be identified. This research proposes novel technique in detection of various plant
disease stages using feature extraction and classification using deep learning techniques. here the input data has been
collected from tomato and grape leafs. This data has been processed for noise removal, image resize and normalization.
Then this image features have been extracted using graph Convolutional networks and classification of extracted features
has been done using ResNet-50. Furthermore, from the experimental model for the public dataset of grape leaf diseases,
this proposed method realizes better outcomes and 94% of average identification accuracy was obtained. This attention
module is added and certified will be extracted accurately as complex features of various diseases with some parameters.
The proposed model delivers a high-performance solution for diagnosing crop disease under the real agricultural
environment.

Keywords: plant disease, tomato, grapes, feature extraction, classification, deep learning

1. Introduction:

In India, the main source of income for healthy lifestyle is agriculture and it is occupying 70% of rural population. The
main population of country is indirectly or directly related with agricultural sector. Henceforth, the agricultural production
is in high-quality and it is essential withstand the economic growth of country. For achieving the highest crop yield with
higher productivity and quality, right products are decided by the farmers in order to decide the correct products to monitor
and to control the essential temperature and requirements of humidity and light [1]. Moreover, the industry of agriculture
is motivated towards the increment of food production in order to meet the growth of population, climatic changes and
instability of political situation. The researchers are attracted towards exclusive, resourceful and consistent technologies
used to increase the agricultural productivity. Some difficulties like initial plant disease identification where the farmers
are struggling. For the disease type of observation, the type of the disease is not identified by the leaves of plant through
the natural vision each and every time, therefore an automated expert system is utilized for the disease detection and are
highly essential. The technical advancement with the combination of image processing and machine learning are utilized
by farmers by means of plant disease in the initial stages [2]

Across the world, tomato is an important crop and 20 kg per year was consumed per year. Around 15% of tomato
consumption are accounted. Nearly 170 million tons of fresh tomato are obtained as yearly yield. The US, India, Turkey,
Egypt and China yields tomatoes in higher production rate [3]. As per the Food and Agriculture Organization survey data
of UN, the production of tomato across the globe is affected mainly by the tomato disease with the rate of annual loss are
in the range of 8%–10%. Though, most of the tomato disease begins from the leaves and it is spread across the whole
plant. Diseases of tomato leaves are identified automatically is enhanced by managing the production of tomato and
produces environmental growth. Conventional diagnosis expert on tomato leaf disease has higher cost and subjected
towards the risk of misjudgement. Computer technology, computer vision, machine learning, and deep learning are

752
JOURNAL OF ALGEBRAIC STATISTICS
Volume 13, No. 3, 2022, p. 752-763
https://publishoa.com
ISSN: 1309-3452
developed rapidly by detecting the crop disease. Conventional methods of machine learning uses RGB image
segmentation for crop disease and texture, shape features are also used in this machine learning methods. Though, various
disease characteristics are same, so it is challenging task for the judgement towards the disease types, identification
accuracy of disease is low in a complicated natural environment.

Deep learning network called Convolutional Neural Network (CNN) produces higher performance; this deserves a
complicated image pre-processing and the operation of feature extraction and assumes an end-to-end structure in which
the process of recognition is simplified as compared to its learning. Today, the recognition of crop disease is proceeded
by CNN for real agriculture environments, in which tomato leaf disease is automatically detected and improve the
diagnose accuracy and labor costs are reduced [4].

Molecular immunology and biology are used in this techniques for the crop disease detection at earlier stage. Still, human
experts, enormous resources are needed in these techniques. As per FOA, areas of cultivation are insignificant and
proceeded by the people in the undeveloped countries with lower income. Thus, exclusive solution are unreasonable for
them and well-organized and effective techniques are proposed and it is made available to each and every farmers [5].
Now, machine learning (ML) depending on hand-coded techniques are established in the agricultural field for enhancing
decision-making power. Because of the development of digital approaches, enormous amount of data is gathered by
means of machine learning techniques and are used for making an optimized decision. These approaches such as decision
trees (DT), support vector machines (SVM), K-nearest neighbours (KNN), and Gaussian frameworks, etc. are highly
tested to detect the crop disease. The methods of Hand-coded keypoints are easily applied and didn’t require enormous
amount of training data like some time-consuming approaches are requires human expertise. Moreover, traditional ML
computation models, there is a compromise between computational complexity and robustness of detection. Because
larger keypoints are computed and improves the economic burden whereas the small feature-set are utilized that decreases
the system’s localization efficiency. Thus, performance improvements are required, that is specifically takes the
frameworks of decision-support which can contributes thx`e enormous amount of data into appreciated recommendations
[6]. Higher performance has been achieved by the proposed model for the crop disease diagnosis in the agricultural
environment.

The contribution of the work is as follows:

➢ To detect and classify the plant diseases using GUI in the ReSNeT architecture.
➢ To compare the performance of the proposed method with the state-of-art methods.

2. Related works:

Many methods are used for using the technology of dep learning in [7] for improving the vegetable’s survival rate, field
crops and fruits by the detection of early disease and suceeding management of disease. Transfer learning is applied in
the original Alex Net network, average rate of recognition has 10 types of better tomato leaves. The original AlexNet,
called VGG16 structure of network is used with the combination of migration learning in [8] for obtaining 97% of
accuracy on 7 diseased tomato leaves which is segmented. The The belongings of deviation, weight and rate of learning
on the accuracy and speediness of detecting disease are analyzed. Different highly resulted cameras are used in [9] for
capturing images of 9 diseases in tomato and pests and Faster RCNN, R-FCN, and SSD are used for training. By the
properties of transfer learning AlexNet and GoogleNet networks were trained in [10] for the identification of camellia
oleifers. Pre-trained ResNet network was used in [11] for the classification of seven types of tomato diseases with 98.8%
as rate of accuracy. The structure of deep detection model was designed in [12] for the optimization and the improvement
of tomato leaf diseases in the residual network and the classification features of the diseases are obtained by using transfer
learning. Though better results are achieved by the transfer learning, original VGG16 architecture and AlexNet has
complicated structures and enormous parameter are used for its application and model deployment. Depending on the
prior researches, as compared in [13] with the conventional feature extraction techniques, CNNs have greater potential.
The method of detecting diseases are trained by improved CNN from the open source disease dataset and it is contrasted
with conventional systems like SVM, LBP, and GIST and it is proved that this method is greater than the some other
methods of classification. Saliency analysis was used in [14] for the pest location in ta gardens, to decrease the number
of layers in the network and AlexNet with reduced convolutional kernels and it is fused with the algorithm of Dropout

753
JOURNAL OF ALGEBRAIC STATISTICS
Volume 13, No. 3, 2022, p. 752-763
https://publishoa.com
ISSN: 1309-3452
model optimization for improving accuracy. This algorithm is efficient to identify 23 pests in tea gardens and achieved
88.1% as the rate of accuracy. The convolutional kernel size is reduced by Alex Net in [15] to enhance the accuracy of
disease detection and number of parameters for the analysis is reduced by this method. Four diseases in apple leaves are
detected by AlexNet was presented in [16] and 97.62% of recognition accuracy is achieved. An influential NN was
developed in [17] to successfully identify three various disease depending on the leaf’s morphological patterns. This
research gives enormous references for the tomato leaf disease diagnosis in the real time agricultural environment.
Discriminative features are extracted with its characteristics in recent years, and the network is started to use the machine
translation, generative adversarial networks etc. [18]. Even though there were many techniques available in agricultural
field, still it is exploratory. ShuffleNet, attention module was added by [19] that enhances the grape disease rate of
recognition with the PlantVillage dataset to 99.14%. The recognition and segmentation of grape leaf diseases was
presented in [20]. In the fusion featuring process, irrelevant and redundant features are added with noise and it is detached
by Neighborhood Component Analysis (NCA). An accuracy of segmentation with 90% was achieved and the accuracy
of classification was 92% was achieved. For the disease detection, radial basis function (RBF) kernel-based support vector
machine (SVM) learning algorithm was used in [22]. The classification techniques based on efficient machine learning
for disease identification and detection was met in this field. The identification model of plant leaf disease depend on
a deep convolutional neural network (Deep CNN) was presented in [23]. This method was trained utilizing various
training epochs, batch dropouts and sizes. The performance of this system is better as compared to the other machine
learning techniques. Reliability and consistency of this method is high. In the identification of plant disease based on
random forest was used in [24]. Implementation at different phases called creation of dataset, extraction of features,
classifier training and classification. It is suitable for the disease detection.

3. System model:

Proposed model possess several steps, in that the step1 is the process of dataset selection, in which the pre-trained weights
for transfer learning is obtained from the larger dataset and the next step was associated with relation between various
disease classes on plant leaves. The last step is to build and train the deep learning architectures based on the online
available tool called Label Img. This architecture is defined in figure 1.

Collecting pre-historic
data tomato and grapes Training image
Crop image dataset
plant with diseases (tomato and grape)

Processing for
noise removal and
resize

Feature extraction
using SIFT

Detecting the Classification using

Figure- 1 Overall Proposed architecture

754
JOURNAL OF ALGEBRAIC STATISTICS
Volume 13, No. 3, 2022, p. 752-763
https://publishoa.com
ISSN: 1309-3452
An effective image classifiers are built by the process of image augmentation. Starting from 100s to 1000s of training
samples are present in the dataset and this is not enough to construct an accurate method. Few in several options of image
augmentation are used to tune the vertically/horizontally, rotated by several angles and image scaling. In the dataset, the
relevant data is increased by these augmentations. Every image size in the dataset of plant village is in the size of 256x256
pixels. Keras deep-learning are used to process the data and augmenting images. The training procedure neds the
following options of augmentation, they are: Rotation- The training images are randomly rotated over different angles,
Brightness- Lighting variation are adapted in the process of image feeding at the time of training. Shear- Image’s shearing
angle is adjusted in the direction towards clockwise/counter-clockwise. Zoom- Input image is provided by scaling various
factors. Vertical/Horizontal Flip – Random flipping of image towards the vertical/horizontal axis.

3.1 Dataset Selection

For several operations performed in real life, so many datasets are developed having enormous number of classes. E.g; in
the research of object classification/detection, ImageNet dataset with extraordinary number of images are used. Likewise,
MS COCO dataset dataset are used for the purpose of transfer learning and selection respectively. In this dataset, the
number of plant varieties are 14 in number with 4 bacterial diseases, 17 fungal infections, 2 infectious diseases, 2 fungal
illnesses and1 mite-induced disease. 12 species of plant represent the healthy images of leaves.

3.2 Annotation of the Training Dataset

The three sub-datasets of Plant Village dataset are as follows: for the purpose of training it has 70% of total images (38017
images), for the validating purpose it uses 10% of images (5431 images) and for the testing purpose remaining 20% of
images (1517 images) are used. This training set annotation is the 1st step for identification of plant diseases by using
meta=architectures of deep learning. LabelImg was used to annotate the training images, this application of annotation is
in open-source graphic image representation. The coordinates of the bounding box like (Xmin, Ymin, Xmax, and Ymax)
were constructed as a result. The intersection of the union (IoU) and the bounding box predicted have been intersected
and its result are investigated by the ground truth in the bounding boxes. For saving the annotations, the format of XML
files, the Pascal VOC have been utilized.

3.3 Graph Convolutional network based Feature extraction:

Convolutional layer: Neural network (NN) dependent on the spectral theory for the implementation of convolution
operations on topology is called as graph convolutional network. In the graph theory, the spectral convolution is defined
as the product of signal xϵℝN (each nodal scalar) and filterg θ = diag(θ); fourier domain θϵℝN is used for
parameterization and it is represented in equation (1) as shown below:

gθ ∗ x = Ugθ(Λ)U T x (1)

1 1
Where I given vector matrix of normalized graph is represented in a matrix form called U. L = IN − D−2 AD−2 = UΛU T
is the Laplacian, which is diagonal matrix with Λ eigenvalues and U T is defined as x’s graph Fourier transform. For larger
graphs, L which is the eigen decomposition, computed initially as the expensive prohibitive. For this problem avoidance,
K-localized spectral filters are suitably controlled by utilizing the truncated expansion with respect to the Chebyshev
polynomials for filter parameterization gθ in an efficient manner is represented in equation (2)

gθ(Λ) = ∑Kk=0 θk Tk (Λ) (2)

Where the Chebyshev coefficient’s parameter vector is θϵℝN and Chebyshev polynomial for kth order is TK = (Λ′ )ϵ ℝn×n
2Λ
and Λ′ = − IN was evaluated as a diagonal matrix with scaled eigenvalues in the range of [-1, 1]. The polynomial of
λmax
Chebyshev filter is Tk (𝓍) in the kth order was calculated by the relation of stable recurrence

having . The Chebyshev polynomial’s truncated

755
JOURNAL OF ALGEBRAIC STATISTICS
Volume 13, No. 3, 2022, p. 752-763
https://publishoa.com
ISSN: 1309-3452
expansion is the experiment and it is capable for the filter estimation gθ efficiently. The convolution of a signal with gθ
filter is given by the following equation(3) as follows:

gθ ∗ x = ∑Kk=0 θk Tk (L′)x (3)

2
It is noted that(UΛU T )k = UΛk U T by the verification of L′ = L − IN
λmax

Therefore, multi-layer Graph Convolutional Network was considered by the rule of layer-wise propagation as followed
in the equation (4):

H (l+1) = σ(∑Kk=0 Tk (L′ )H (l) W (l) (4)

Here, Layer-specific trainable weight matrix is given byW (l) = θk , (kϵ[0, K]Λkϵℤ) . An activation function is defined by

σ(·).the activation matrix in the layer is denoted as H (l) ϵℝN×D ; H (0) = X

Pooling layer: The graph scale is reduced by the graph signal’s fast pooling whereas preservation of local geometry and
original graph’s connected information. The Metis graph partitioning algorithm with the coarsening phase produces
graph’s fast pooling principle. Smaller graph’s set 𝐺𝑖 = (𝑉𝑖 , 𝐸𝑖 ) in the coarsening phase was obtained from G0 = (V0 , E0 )
as the original graph through the nodal fusion, in which ⃒Vi ⃒ < ⃒Vi−1 ⃒. The combination of nodal set in the graph Gi in
one node in the next-level roughening graphGi+1 is used in most of the process of coarsening. The set of nodes are given
by Viv in the Gi graph that are fused by Vi node of Gi+1 graph. The coarsened graph was made better by taking nodes and
original graph’s edge weights, the node’s weight is given by v and it will be set to the sum of each and every nodes in Viv .
If Viv with multiple nodes have connected edges for u vertex at the similar time, vertex v with edge weights are the edges
weights sum.

Single Shot MultiBox Detector (SSD): The SSD methods are simpler because of the region proposal’s elimination and
next pixel or features are resampled. This deep learning architecture incorporates each and every computation in a
network, that is called single-shot detector. From the findings of experiments, the dataset MS COCO, ILSVRC, and
PASCAL VOC are revealed and SSD is attained as better precision as compared to other deep learning methods like
ResNet-50 with lesser time of computation whereas unified training and framework of inference was provided. The SSD
key feature depends on the convolutional filters that are small like 4x4 and 8x8; category score with feature maps and the
collection with box offset prediction for default bounding boxes.

ResNet-50 based Classification: Grape and tomato leaf disease image is provided as input ResNet-50 with the structure
of network is provided in figure 2, convolutional layer is passed first, activation layer, BN layer and the feature map is
obtained is maximized pooling. The Stage 1–4 is mainly included in this ResNet50 model every stage comprised of
sampling module and numerous mapping modules are identified. The AVG pooling operation has proceeded by the output
feature map and then flatten layer was passed for making the multi-dimensional feature’s output and one-dimension and
the output is obtained finally by the fully connected layer.

Figure 2 The structure of the ResNet-50

756
JOURNAL OF ALGEBRAIC STATISTICS
Volume 13, No. 3, 2022, p. 752-763
https://publishoa.com
ISSN: 1309-3452
As the number of network layeruoujpipu6y]t, o90s are deepened, when the network’s internal characteristics have attained
certain layer’s optimal level, the next superimposed layer of the network doesn’t alter the characteristics. ResNet-50’s
residual module will have utilized efficiently in the problem solving for the mapping identification and network
parameters and computations are reduced.

Figure 3. The residual block

Residual units form the basis for residual networks [25]. Every residual unit will be shown in equation (5) and (6):

yi = h(xi ) + F(xi , wi ) (5)

xi+1 = f(yi ) (6)

Where, the residual function is given by F, ReLU function is denoted by f, weight matrix is given by wi and layer i with
input and output is given by xi and yi respectively. The h function has a mapping identity expressed as

h(xi ) = xi (7)

The F as residual function is described in equation (8) is as follows:

F(xi , wi ) = wi . σ(B(wi′ ). σB(xi ))) (8)

Where, the batch normalization is given by B(xi ) and convolution is denoted by “·” and σ(x) = max(x, 0). The necessary
impression behind the residual learning in the path’s branching for gradient propagation. Some similarities are shared in
the residual networks having highway networks like residual blocks and shortcut connections. Every path’s output in the
highway network is measured by the function of and it is learnt in the phase of training. ResNet’s residual units are not
stacked together with convolutional layers in CNN. Rather every convolutional layer’s input to its output is introduced
by the shortcut connections. Shortcut connections with mapping identification reduces the residual network complexity
in deep network which are trained faster. Many path’s ensemble was used in ResNets rather than noticing DL architecture.
Each and every network paths in the ResNets are with similar length. A single path goes off with each and every residual
units. Furthermore, each and every signal path doesn’t propagate the gradient that accounts for fast optimization and
ResNet’s training.

The 224×244×3 is the input image size. Five convolution blocks are present in this network, every block comprising
convolutional layers and pooling layer with two FC hidden layer (4096 neurons with every layer), output layer is end with
softmax activation with 1000 classes.

In between the deep layers, skip connection is present depending on the ResNet. More than one non-linear transformation
layers have skipped by these skip connections. These connection’s output and the network stacked layer’s output are
added.

In this final block output is given by H(x), the connected layer’s output is given by x and the stacked network layer’s
output in the similar block is represented by F(x).

757
JOURNAL OF ALGEBRAIC STATISTICS
Volume 13, No. 3, 2022, p. 752-763
https://publishoa.com
ISSN: 1309-3452
3.4 Stochastic Gradient Descent (SGD) with Momentum

One of the Neural network’s optimization algorithm is called as gradient descent. Ability of faster convergence are present
in the momentum version than the standard algorithm. The

The elementary impression is used for the calculation that is weighted average of the gradients are weighted exponentially
and the weights are updated by the gradient used. For the cost function optimization, minimum loss is obtained by the
gradient descent, that slows the gradient descent and large rate of learning are avoided. Higher learning rate are utilized
and the problems of overshooting and diverging are ended. SGD optimizer are contrasted with the calculation of weight’s
gradient and biases with calculated gradient, dw and db are th weighted averages are exponential a momentum algorithm
is considered [26] by the following equation (9) and (10) as follows:

Vdw = β ∗ Vdw + (1 − β) ∗ dw (9)

Vdb = β ∗ Vdb + (1 − β) ∗ db (10)

4. Performance analysis:

Python tool is used for the implementation of proposed disease classification and PC with Ubuntu, 4GB RAM, and Intel
i3 processor are the configurations of the system.

4.1 Dataset description:

Plant Village is an open dataset used for the experimental analysis. Grapevine leaf black rot with 1180 images are used
for detecting disease. LabelImg are used for the disease annotation. The number of diseases are there in images that can
detect about 17000 targets of detection in total. Previous to the training process, 1180 images are divided into training
and test sets. The network training of 1072 was selected for network training and test set with 108 images are used for the
network evaluation named as test_pv. Additionally, grape leaves with 108 images with spotted black rot in the orchard
environment were gathered as an extra test set named test_orchard. Two parts of training set in the network training
process is called as training set and validation set. The training and validation division ratio is 9:1.In CNN, the training
set is utilized for fitting model and the validation set is a distinct sample set in the model training process, that was utilized
towards the adjustment of super parameters in this method and the capability of the model is preliminarily evaluated.

4.2 Results:

The stages disease detection for tomato and grapes has been analysis based on the confusion matrix given in figure 4 (a)
and (b). Here the disease has been detected from the true and false positive rate by actual and predicted class. It provides
the efficiency of the proposed model in comparison with actual and predicted values. This evaluation depending on the
estimating True Positive (TP), False Negative (FN), True Negative (TN), and False Positive (FP) as represented in
equation (15)

TP FP
Confusion Matrix = [ ] (15)
FN TN

here, TP is the predicted value is predicted as positive. FP is the predicted value which is initially projected as negative
and later as positive. TN is the value predicted as negative and expected as unfavourable. FN is the predicted value that
is initially projected as positive and negative later.

758
JOURNAL OF ALGEBRAIC STATISTICS
Volume 13, No. 3, 2022, p. 752-763
https://publishoa.com
ISSN: 1309-3452
Result 1:

(a) (b)

Figure-4 Confusion matrix in stages of disease detection for tomato and grapes plant

The below figure-5 (a), (b) shows the precision- recall (PR) and ROC curve for grape diseases detection using proposed
feature extraction and classification techniques.

Result 2:

(a) (b)

Figure-5 PR and ROC curve for grape disease detection

(a) (b)

Figure-6 PR and ROC curve for tomato disease detection

759
JOURNAL OF ALGEBRAIC STATISTICS
Volume 13, No. 3, 2022, p. 752-763
https://publishoa.com
ISSN: 1309-3452
The above figure-6 (a), (b) shows the precision- recall (PR) and ROC curve for tomato diseases detection using proposed
feature extraction and classification techniques. Hence from above confusion matrix the prediction class has been
validated along with the parameters of precision and recall of a deep neural network classification method. The entire
comparative analysis of proposed technique is represented in table-1 and figure-7 (a), (b).

Table-1 Comparative Analysis in crop disease detection

Crop Parameter k-means SVM GLCM Pro_GCN-ResNet50

Accuracy 95 96 93 93.86

Precision 89 90 92 94.70

Tomato Recall 74 75 76 93.86

F-1 score 73 75 76 93.94

AUC 89 84 87 99.01

Accuracy 89 88 90 96.78

Precision 86 87 89 96.73

Grapes Recall 71 72 74 96.52

F-1 score 70 86 82 96.60

AUC 99.80

(a) Tomato stages of disease detection

760
JOURNAL OF ALGEBRAIC STATISTICS
Volume 13, No. 3, 2022, p. 752-763
https://publishoa.com
ISSN: 1309-3452

(b) Grapes stages of disease detection

Figure-7 Comparative analysis of (a) Tomato and (b) Grapes stages of disease detection

5. Conclusion:

This research proposes novel technique in detection of various plant disease stages using feature extraction and
classification using deep learning techniques. here the input data has been collected from tomato and grape leafs. This
data has been processed for noise removal, image resize and normalization. Then this image features have been extracted
using graph Convolutional networks and classification of extracted features has been done using ResNet-50. Seven
disease classes are used for the identification of minimum layer’s set are used in this technique. Plant Village dataset is
used for training the neural network. This system is designed by using Graphical User Interface. The image dataset is
chosen by the user with GUI permission. Any image from the dataset can be selected by the user and the images are
loaded that is followed by the disease prediction and it is shown by the user interface. Identification and recognition of
disease in the plant leaf is trained by ResNet-50 and it is used in the classification and prediction of correct disease for
nearly with some anomalies. Furthermore, the experimental comparison of grape leaf diseases in the public dataset and
better results are obtained with 94% of average identification accuracy. Attention module is added and it is certified there
is an accurate complex feature extraction based on the disease variety and some parameters. Higher performance was
achieved by the proposed model for the crop disease diagnosis in the agricultural environment.

References:

1. Shruthi, U., Nagaveni, V., & Raghavendra, B. K. (2019, March). A review on machine learning classification
techniques for plant disease detection. In 2019 5th International conference on advanced computing &
communication systems (ICACCS) (pp. 281-284). IEEE.
2. Jogekar, R. N., & Tiwari, N. (2021). A review of deep learning techniques for identification and diagnosis of plant
leaf disease. Smart Trends in Computing and Communications: Proceedings of SmartCom 2020, 435-441.
3. Venkataramanan, A., Honakeri, D. K. P., & Agarwal, P. (2019). Plant disease detection and classification using deep
neural networks. Int. J. Comput. Sci. Eng, 11(9), 40-46.
4. Ramesh, S., Hebbar, R., Niveditha, M., Pooja, R., Shashank, N., & Vinod, P. V. (2018, April). Plant disease detection
using machine learning. In 2018 International conference on design innovations for 3Cs compute communicate
control (ICDI3C) (pp. 41-45). IEEE.

761
JOURNAL OF ALGEBRAIC STATISTICS
Volume 13, No. 3, 2022, p. 752-763
https://publishoa.com
ISSN: 1309-3452
5. Patidar, S., Pandey, A., Shirish, B. A., & Sriram, A. (2020, July). Rice plant disease detection and classification
using deep residual learning. In International Conference on Machine Learning, Image Processing, Network
Security and Data Sciences (pp. 278-293). Springer, Singapore.
6. Arsenovic, M., Karanovic, M., Sladojevic, S., Anderla, A., & Stefanovic, D. (2019). Solving current limitations of
deep learning based approaches for plant disease detection. Symmetry, 11(7), 939.
7. Jasim, M. A., & Al-Tuwaijari, J. M. (2020, April). Plant leaf diseases detection and classification using image
processing and deep learning techniques. In 2020 International Conference on Computer Science and Software
Engineering (CSASE) (pp. 259-265). IEEE.
8. Loey, M., ElSawy, A., & Afify, M. (2020). Deep learning in plant diseases detection for agricultural crops: a
survey. International Journal of Service Science, Management, Engineering, and Technology (IJSSMET), 11(2), 41-
58.
9. Trivedi, J., Shamnani, Y., & Gajjar, R. (2020, February). Plant leaf disease detection using machine learning.
In International conference on emerging technology trends in electronics communication and networking (pp. 267-
276). Springer, Singapore.
10. Ashok, S., Kishore, G., Rajesh, V., Suchitra, S., Sophia, S. G., & Pavithra, B. (2020, June). Tomato leaf disease
detection using deep learning techniques. In 2020 5th International Conference on Communication and Electronics
Systems (ICCES) (pp. 979-983). IEEE.
11. Tulshan, A. S., & Raul, N. (2019, July). Plant leaf disease detection using machine learning. In 2019 10th
International Conference on Computing, Communication and Networking Technologies (ICCCNT) (pp. 1-6). IEEE.
12. Badage, A. (2018). Crop disease detection using machine learning: Indian agriculture. Int. Res. J. Eng.
Technol, 5(9), 866-869.
13. Wang, G., Sun, Y., & Wang, J. (2017). Automatic image-based plant disease severity estimation using deep
learning. Computational intelligence and neuroscience, 2017.
14. Li, L., Zhang, S., & Wang, B. (2021). Plant disease detection and classification by deep learning—a review. IEEE
Access, 9, 56683-56698.
15. Panchal, A. V., Patel, S. C., Bagyalakshmi, K., Kumar, P., Khan, I. R., & Soni, M. (2021). Image-based Plant
Diseases Detection using Deep Learning. Materials Today: Proceedings.
16. Ouhami, M., Hafiane, A., Es-Saady, Y., El Hajji, M., & Canals, R. (2021). Computer vision, IoT and data fusion for
crop disease detection using machine learning: a survey and ongoing research. Remote Sensing, 13(13), 2486.
17. Bedi, P., & Gole, P. (2021). Plant disease detection using hybrid model based on convolutional autoencoder and
convolutional neural network. Artificial Intelligence in Agriculture, 5, 90-101.
18. Roy, A. M., & Bhaduri, J. (2021). A deep learning enabled multi-class plant disease detection model based on
computer vision. AI, 2(3), 413-428.
19. Tiwari, D., Ashish, M., Gangwar, N., Sharma, A., Patel, S., & Bhardwaj, S. (2020, May). Potato leaf diseases
detection using deep learning. In 2020 4th International Conference on Intelligent Computing and Control Systems
(ICICCS) (pp. 461-466). IEEE.
20. Adeel, A., Khan, M. A., Sharif, M., Azam, F., Shah, J. H., Umer, T., & Wan, S. (2019). Diagnosis and recognition
of grape leaf diseases: An automated system based on a novel saliency approach and canonical correlation analysis
based multiple features fusion. Sustainable Computing: Informatics and Systems, 24, 100349.
21. Pukkela, P., & Borra, S. (2018). Machine learning based plant leaf disease detection and severity assessment
techniques: State-of-the-art. In Classification in BioApps (pp. 199-226). Springer, Cham.
22. Kumar, S. (2022). A quest for sustainium (sustainability Premium): review of sustainable bonds. Academy of
Accounting and Financial Studies Journal, Vol. 26, no.2, pp. 1-18
23. Allugunti V.R (2022). A machine learning model for skin disease classification using
24. convolution neural network. International Journal of Computing, Programming and Database Management 3(1),
141-147
25. Geetharamani, G., & Pandian, A. (2019). Identification of plant leaf diseases using a nine-layer deep convolutional
neural network. Computers & Electrical Engineering, 76, 323-338.
26. Liu, B., Zhang, Y., He, D., & Li, Y. (2017). Identification of apple leaf diseases based on deep convolutional neural
networks. Symmetry, 10(1), 11.

762
JOURNAL OF ALGEBRAIC STATISTICS
Volume 13, No. 3, 2022, p. 752-763
https://publishoa.com
ISSN: 1309-3452
27. Ramesh, S., Hebbar, R., Niveditha, M., Pooja, R., Shashank, N., & Vinod, P. V. (2018, April). Plant disease detection
using machine learning. In 2018 International conference on design innovations for 3Cs compute communicate
control (ICDI3C) (pp. 41-45). IEEE.
28. Anand, D., Arulselvi, G., Balaji, G.N. (2022). An Assessment on Bone Cancer Detection Using Various Techniques
in Image Processing. In: Deepak, B.B.V.L., Parhi, D., Biswal, B., Jena, P.C. (eds) Applications of Computational
Methods in Manufacturing and Product Design. Lecture Notes in Mechanical Engineering. Springer, Singapore.
https://doi.org/10.1007/978-981-19-0296-3_48.

763