Research Journal of Computer Systems and Engineering

ISSN: 2230-8571, 2230-8563

Volume 02 Issue 02 - 2021 (July to December)
Page 27:32

Fungal and Bacteria Disease Detection Using Feature Extraction with

Classification Based on Deep Learning Architectures

Dr.Neha Verma
Green Energy Technology
SSIPMT,Raipur
ORCHID ID-0000-0001-9091-4428
[email protected]

Abhilash Reddy
Computer Science & Communication Engineering
AVN institute of engineering and technology
ORCHID ID-0000-0002-0860-8606
[email protected]

Dr. Sridhar Thota

Computer Science & Communication Engineering;VLSI and Embedded Systems
Alliance University
ORCHID ID-0000-0002-2047-4882
[email protected]

Article History Abstract

Received: 15 July 2021 In India, agriculture provides a living for half the population. Food
Revised: 20 September 2021 security is significantly threatened by microbial infections, however
Accepted: 22 November 2021 due to inadequate infrastructure, it is still difficult to identify them
quickly. This paper propose novel technique in Fungal and bacteria
disease detection of tomato and grapes plant using deep learning
based on feature extraction and classification. Here the input
images of tomato and grapes has been collected and processed for
noise removal and image resize. This processed image features
have been extracted using scale-invariant feature transform (SIFT)
and classified using Convolutional neural network-based Alex Net
architecture. The suggested model has attained the highest testing
accuracy. The deep learning model utilised in this study aims to
identify disease in plant leaves. However, in the future, the model
can be combined with a drone or any other technology to identify
plant diseases in real time and inform people of the locations of the
afflicted plants so that they can be treated appropriately.
Keywords: plant diseases, Fungal disease, bacterial disease, deep
learning, feature extraction, classification.
CC License CC-BY-NC-SA

Available at: https://technicaljournals.org/RJCSE/index.php/journal 27

 Research Journal of Computer Systems and Engineering

1 Introduction:

Climate change and the economy both heavily rely on plants. Due to the fact that it has become a
worldwide concern, the 2019 UN General Assembly will also address Future climate change is
expected to occur at a rate that is 10–100 times greater than that of DE glacial warming [1]. According
to estimates in the hundreds of billions of dollars [2], plant pests and diseases generate annual losses
in food, fibre, and decorative production systems.Finding illness in plants at an early stage is a very
difficult task [3]. Common plant disease symptoms include Powdery mildew, Sclerotinia (white
mould), Leaf rust (common leaf rust in maize), Stem rust (wheat stem rust), Anthracnose, seedling
damping off, leaf spot (septoria brown spot), and chlorosis (leaf yellowing) are some of the diseases
that can affect plants [4].

2 Related works:

Researchers have come up with numerous strategies to solve the aforementioned issues. Machine
learning can classify plant diseases using many different kinds of feature sets. The classic handcrafted
and deep learning (DL)-based feature sets among them are the most well-liked feature sets. Before
effectively extracting features, preprocessing, such as picture enhancement, colour modification, and
segmentation [5], is necessary.Following feature extraction, various classifiers may be employed. K-
nearest neighbour (KNN), support vector machine (SVM), decision tree, random forest, naïve bayes
(NB), logistic regression (LR), rule generation, artificial neural networks (ANNs), and Deep CNN are
some examples of well-liked classifiers [6,7].KNN is a straightforward supervised machine learning
technique that uses similarity measurements (i.e., distance, proximity, or closeness) to solve
classification problems [8,9]. Another well-liked supervised machine learning method for
classification is SVM. Finding a hyperplane between data classes that splits each class is the goal of
SVM. Network depth, width, and resolution may improve performance accuracy with fewer
parameters, claims [10].

3 System model:

This section discuss novel technique in Fungal and bacteria disease detection of tomato and grapes
plant using deep learning based on feature extraction and classification. Here the input images of
tomato and grapes has been collected and processed for noise removal and image resize. This
processed image features has been extracted using scale-invariant feature transform (SIFT) and
classified using Convolutional neural network based AlexNet architecture.The overall proposed
architecture is shown in figure-1.

Figure-1 Overall Proposed architecture

Available at: https://technicaljournals.org/RJCSE/index.php/journal 28

 Fungal And Bacteria Disease Detection Using Feature Extraction with Classification Based on Deep Learning Architectures

4 SIFT based feature extraction:

Features that are invariant to scale, rotation, and translation are produced by the SIFT algorithm.
Computing a Gaussian scale-space out of the source image is the first step in the SIFT feature
extraction process. The initial image is divided into 5 octaves, each of which comprises variations of
the original image with a lower samplerate. There are 5 photos in each octave, all with the same
sample rate but varying degrees of blur (Gaussian blur is used).The information regarding the local
spatial distribution of the gradient orientation on a specific neighbourhood must be encoded in order
to compute the descriptor. There have been publications that have used the complete image as a
neighbourhood. The original SIFT approach, in contrast, uses a square normalised patch aligned with
the point's orientation to create scale, rotation, and translation invariance by eq. (1).
min𝑦𝑒,𝑧𝑒 ∏𝑖∈𝑒 ((𝑦𝑒𝐼𝑖 + 𝑧𝑒 − 𝑞𝑖)2 + 𝑦𝑒2 )
𝐻𝑒 (𝐼𝑞 ) − 𝐻𝑒 (𝐼)𝐻𝑒 (𝑞)
𝑦𝑒 =
𝐽𝑒 (𝐼) + 𝜖 ′
𝑧𝑒 = 𝐻𝑒 (𝑞) − 𝑦𝑒𝐻𝑒 (𝐼) (1)
𝑔𝑖 = 𝐻𝑒 𝑖(𝑓𝑖0) = 𝐻𝑒 𝑖(𝑦)𝐼𝑖 + 𝐻𝑒 𝑖(𝑧)

5 C-Alex Net based Classification:

A feature extractor, which consists of a convolutional layer and a pooling layer and can efficiently
recombine extracted low-level features to extract higher-level features as illustrated in figure 2, was
introduced for convolutional neural networks, which are feedforward neural networks. Due to the
sharing of weights, automatic feature extraction significantly reduces the amount of manual work
required, making it quicker and more effective.

Figure 2 structure of a CNN, consisting of convolutional, pooling, and fully-connected layers.

Convolutional, pooling, full connection, and output layers are the typical components of a
convolutional neural network. Softmax classifier, which is suitable for solving multi-classification
issues, has a minimal computation amount and quick training times, and it is typically used in the
output layer. Assuming that equation (2) accurately describes the function:
𝑝(𝑦 (𝑖) = 1 ∣ 𝑥 (𝑖) ; 𝜃) 𝑇 (𝑖)
𝑒 𝜃1 𝑥
(𝑖) (𝑖) 𝑇 (𝑖)
𝐡𝜃 (𝑥 (𝑖) ) = 𝑝(𝑦 = 2 ∣ 𝑥 ; 𝜃) =
1 𝑒 𝜃2 𝑥 (2)
𝜃𝑇 𝑥(𝑖)
⋮ ∑𝐾j=1 𝑒
𝑗 ⋮
(𝑖) (𝑖) 𝑇 (𝑖)
[𝑝(𝑦 = K ∣ 𝑥 ; 𝜃)] [𝑒 𝑥 ]
𝜃𝐾
𝑇 (𝑖)
1 𝑒 𝜃𝑗 𝑥
𝐽(𝜃) = − [∑𝑚 ∑ 𝑘
𝑙{𝑦 (𝑖)
= 𝑗}]log⁡
𝑚 𝑖=1 𝑗=1 ∑𝑘𝑙=1 𝑒 𝜃𝑖 𝑥
𝑇 (𝑖)

𝑓(𝑥) = max(𝑥, 0)

Available at: https://technicaljournals.org/RJCSE/index.php/journal 29

 Research Journal of Computer Systems and Engineering

First, second, and fifth convolutional layers are followed by three max-pooling layers. AlexNet has
eight weight levels total, including three fully connected layers and five convolutional layers. In
conclusion, pre-trained AlexNet architecture with SS strategy functions in a way that allows for the
output performance of complete network to be attained with "satisfactory" performance from the
hidden layer classifiers. Similar to the original AlexNet architecture, pre-trained AlexNet architecture
with SS strategy uses SGD algorithm and gradient functions for the optimization process.

6 Performance analysis:

On Windows 10, we run the simulations using the PyTorch deep learning framework. The PC used for
the tests has an AMD Ryzen 5 1600X Six-Core Processor and an 8GB GeForce GTX 1070Ti GPU.
Programming is carried out using the Python language.

7 Dataset description:

The open dataset Plant Village is what was used in this study's trials. For disease detection, we use
1,180 photos of black rot on grapevine leaves. For annotating the sick leaf portions, we employ
LabelImg. The dataset utilised in this experiment is in the coco dataset format. To hasten convergence,
we employ the model weights that have already been trained.
Though, maximum iteration number is 60, this has the development certainly in terms of accuracy for
various classes of prediction because of the errors in loss and training. Though being the trained
model which is appropriate in terms of statistics, then this model do not possess the efficient capacity
in prediction. Because of this regularization the simulation has been maximized when the simulation
for enhanced test among them ranges between the values for regularization parameter. 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 model.the overall comparative analysis of proposed
technique is shown in table-1 and figure-3 (a), (b).

Table-1 Comparative Analysis in crop disease detection

Crop Parameter k-means SVM GLCM Pro_C-AlexNet

Accuracy 95 96 93 97.25
Precision 89 90 92 97.18
Tomato
Recall 74 75 76 97.05
F-1 score 73 75 76 97.10
Accuracy 89 88 90 94.93
Precision 86 87 89 95.50
Grapes
Recall 71 72 74 94.93
F-1 score 70 86 82 94.99

(a) Tomato Fungal and bacteria disease detection

Available at: https://technicaljournals.org/RJCSE/index.php/journal 30

 Fungal And Bacteria Disease Detection Using Feature Extraction with Classification Based on Deep Learning Architectures

(a) Grapes Fungal and bacteria disease detection

Figure-3 Comparative analysis of (a) Tomato and (b) Grapes disease detection

8 Conclusion:

This research propose novel technique in Fungal and bacteria disease detection of tomato and grapes
plant using deep learning based on feature extraction and classification. Here the input images of
tomato and grapes has been collected and processed for noise removal and image resize. This
processed image features has been extracted using scale-invariant feature transform (SIFT) and
classified using Convolutional neural network based AlexNet architecture.In this study, we also put up
the idea of using occlusion techniques to pinpoint the disease-affected areas and aid in human disease
comprehension. Our long-term goal is to make deep models smaller and less computationally
intensive for mobile devices and other small computers. In addition, feature visualisation is a popular
area in deep learning and could be applied to comprehend plant illnesses.

Reference:

[1] Valenzuela, M.E.M.; Restovi´c, F. Valorization of Tomato Waste for Energy Production. In
Tomato Chemistry, Industrial Processing and Product Development; Royal Society of
Chemistry: London, UK, 2019; pp. 245–258.
[2] Elnaggar, S.; Mohamed, A.M.; Bakeer, A.; Osman, T.A. Current status of bacterial wilt
(Ralstoniasolanacearum) disease in major tomato (Solanumlycopersicum L.) growing areas in
Egypt. Arch. Agric. Environ. Sci. 2018, 3, 399–406.
[3] Boulent, J.; Foucher, S.; Théau, J.; St-Charles, P.-L. Convolutional neural networks for the
automatic identification of plant diseases. Front. Plant Sci. 2019, 10, 941.
[4] Ma, J.; Zheng, F.; Zhang, L.; Sun, Z. Disease recognition system for greenhouse cucumbers
based on deep convolutional neural network. Trans. Chin. Soc. Agric. Eng. 2018, 34, 186–192.
[5] Arya, S.; Singh, R. A Comparative Study of CNN and AlexNet for Detection of Disease in
Potato and Mango leaf. In Proceedings of the 2019 International Conference on Issues and
Challenges in Intelligent Computing Techniques (ICICT), Ghaziabad, India, 27–28 September
2019; pp. 1–6.
[6] Kaur, M.; Bhatia, R. Development of an improved tomato leaf disease detection and
classification method. In Proceedings of the 2019 IEEE Conference on Information and
Communication Technology, Baghdad, Iraq, 15–16 April 2019; pp. 1–5.
[7] Ireri D. Belal E. Okinda C. Makange N. Ji C. (2019). A computer vision system for defect
discrimination and grading in tomatoes using machine learning and image processing.Artificial
Intelligence in Agriculture, 2, 28–37. 10.1016/j.aiia.2019.06.001

Available at: https://technicaljournals.org/RJCSE/index.php/journal 31

 Research Journal of Computer Systems and Engineering

[8] Badage, A. (2018). Crop disease detection using machine learning: Indian agriculture. Int. Res.
J. Eng. Technol, 5(9), 866-869.
[9] ALEnezi, N. S. A. (2019). A method of skin disease detection using image processing and
machine learning. Procedia Computer Science, 163, 85-92.
[10] Badage, A. (2018). Crop disease detection using machine learning: Indian agriculture. Int. Res.
J. Eng. Technol, 5(9), 866-869.

Available at: https://technicaljournals.org/RJCSE/index.php/journal 32