A Comparative Analysis of CNN Models in Deep Learning for Leaf Disease Detection

: There is a huge growth in the population of India therefore there is necessity for more food. But in farming the unavoidable plant diseases causing a large problem. There is a need to figure out as to how much food produced by the farmers is affected. Because in the coming future a greater amount of people are to be fed. Plant leaf disease detection is very important, because depending on the amount of growth in crop the farmers make their money. Then here comes CNN in help. It is tool which is smart enough to identify the diseases and their types. In order to detect the disease in plant a Convolutional Neural Network(CNN) with the help of image processing beside is in use here in our paper. A Convolutional Neural Network is an artificial neural network which is specially designed to deal with image recognition[1] tasks when an image is input. Here the idea is to use CNN models to spot diseases in apple, grape, corn and potato. This idea is to use CNN models to spot diseases in apple, grape, corn, and potato plants. We proposed an algorithm. This paper mainly focused on CNN models CNN, AlexNet,VGG16 in deep learning that will be compared in the study.


Introduction:
The main food source of India is crop production. To improve the productivity of crops we are taking the help of advanced technology. Many varieties of plants and crops are cultivated by Indian farmers. Therefore a lot of research is been centred to search ways to cultivate more food which is healthy. In the present scenario depending completely on human speculation to detect efficiently the diseases in plant is not a good choice. Yet the modern expansion in computer vision provides the solutions to issues faced with plant and leaf which is rapid, consistent and accurate. In these recent years an outstanding amount of research has been done in deep learning in the fields like image recognition, sentiment analysis and speech recognition. Therefore convolutional neural network will be most effective means to detect diseases in leaf and plants and solve the problem raised [2]. An algorithm is proposed for the leaf disease detection. We will be importing libraries needed in the initial stage: OS, Tensorflow, pandas, Matplotlib, Cv2,Keras, random, NumPy pandas, etc. Neat the function is defined to label images and to load the training data. The images are been categorized based on the plant disease's code names. The resizing and is done to a group of random images while training the images and matching labels are added consequently.
Leaf images are added to do the testing and a CNN algorithm to add series of layers is for classification. These layers are like convolutional and pooling layers used for process optimization in every phase. And the required output is obtained from the dense layer. The learning rate is used to affect the rate of our models learning process. Later the data is been loaded to our built model, to designate the nature of leaf a healthy or diseased using a variable and afterwards the in this variable the model is saved. And this variable is used for the detection.

LITERATURE SURVEY
Nishant Shelar and colleagues discovered spotinfections in leaves and categorized them according to thediseased leaf categories using various learning algorithms . Network is to acquire and analyze data from leaf photosin order to determine healthy or diseased leaves ofmedical plants using image processing methods. [1] The paper introduces by sumit and colleagues a novel approach to image recognition by utilizing deep learning techniques. The researchers investigated three distinct neural network architectures: Faster R-CNN, R-CNN, and SSD. Their efforts resulted in a commendable validation accuracy of 94.6%. This proposed method is capable of identifying a range of diseases affecting leaves from apple, cherry, grape, peach, pepper, potato, strawberry, and tomato plants. [2] S.Bharath , K.Vishal Kumar , R.Pavithram, T.Malathi , "Detection of Plant Leaf Disease using CNN ". Image processing and CNN model can be used to improve plant disease detection techniques. It consisting data of 38 different plant leaf diseases which used to predict.The conclusion of this paper is to predict the pattern of plant diesease using CNN. [3] The investigation led by Murk Chohan and colleagues was centred on the identification of diseases in plant leaves. To enhance the dataset's sample size, they implemented augmentation techniques. For testing purposes, 15% of the data from the Plant Village dataset was employed. [4] Ali Arshagi and colleagues undertook a study in this paper, focusing on the analysis of five distinct categories of potato diseases, including Healthy, Black Scurf, Common scab, Black leg, and Pink dot. The researchers proceeded to evaluate the outcomes produced by various disease classification methods, including AlexNet, GoogLeNet, VGG, and R-CNN. [5] PROPOSED METHODOLOGY: We cannot feed the image data directly to our neural network. To prepare image data for input into a neural network, a series of pre-processing steps are required. This involves converting the images into a format that the network can understand, which often means transforming them into NumPy arrays.

Feature Extraction:
In Convolutional Neural Networks (CNNs), filters are employed to extract features from images. The remarkable aspect is that during training, the network autonomously detects these filters, determining both their quantities and parameter values. This intrinsic learning process adjusts filter amounts and content. As a hyperparameter, you prescribe the desired quantity of filters and their dimensions, influencing the architecture's capability to discern intricate image characteristics.

CNN Classification:
The CNN classifier for image classification is built using a special kind of neural network called CNN. Its main job is to look at pictures and decide which category they belong to. It learns to pick out important parts from the pictures and use them to figure out the right category, like telling what's in a photo. When you give the network a picture in the form of numbers (numpy array), it gives you a number between 0 and 1 that tells you how sure it is about its decision.

VGG16:
The VGG16 model attains an impressive test accuracy of 92.7% on ImageNet, an extensive dataset encompassing over 14 million training images distributed among 1000 distinct object classes. VGG16 represents an advancement over the AlexNet architecture by adopting a strategy of substituting the large filters with a series of smaller 3×3 filters. Unlike AlexNet, where the kernel size is 11 for the initial convolutional layer and 5 for the subsequent layer, VGG16 utilizes a consistent kernel size of 3×3