CLASSIFICATION OF DIFFERENT PLANT LEAF DISEASES USING MULTIPLE CONVOLUTIONAL NEURAL NETWORK AND IMAGE PREPROCESSING

Abstract

Since it guarantees food stability, agriculture is a vital part of the global economy. Plants, on the other hand have recently been found to be heavily afflicted by a variety of diseases. Detecting and tracking plant diseases used to be done manually with the help of experts in the field. However this process was time consuming, unreliable, and laborious. Experience, knowledge of plant is essential for making the right decision and selecting the most appropriate method for the treatment of the diseases. This research paper discuss a variety of bacterial and fungal diseases[4] as well as how to recognize and classify them using image pre-processing techniques[1]. With this method,we will identify the type of plant, decide whether the plant has diseases, and classify the disease type. Batch normalization used in this approach is a technique for avoiding network overfitting while also increasing the model's robustness. The Relu activation function and Adam optimizer are used to improve convergence and accuracy of the model.

Purpose: The objective of this paper is to identify the type of plant, decide whether the plant has diseases, and classify the disease type. 

Methodology: Diseases and disorders affect plants in a variety of ways. Environmental factors like temperature, humidity, nutrient surplus or loses, light, and the most common diseases like bacterial, viral, and fungal[4] diseases are all potential causes. The system identification of leaf disease for different plants like cherry, tomato, potato, peach, and strawberry is done using the  plantvillage  dataset  as well as real-time datasets, as these plant  diseases can show different characteristics on the leaves, such as a change in form, scale, and colour. The recent detectors such as convolutional neural networks[2] and image pre-processing for the identification and classification of plant diseases is used. Models are trained using images from an open database containing various plants. Previously, onlya single plant disease couldbe detected at a time,butnowusing different layers of the convolutional neural network, multiple plant diseases can be detected at the same time.Image preprocessing is a technique for improving the quality of an image by eliminating noise or distortions. In image preprocessing technique the images collected from the dataset are resized into default size and labelling of the images are done. The photos collide during image pre-processing. MaxPooling2D is used to max pool the value from a given size matrix. Flatten layer   is   used   to flatten the dimensions of a picture after it has been convolved. Steps involved are getting datasets, pre-processing, labelling of images, augmentation phase, build a model and validation. Collecting the required dataset which is used as the input to the network. Data augmentation is carried out here by performing various operations on the training images, such as rotation, distance, height, shear, and zooming. The model is trained using the augmented data as input. Conv2D layer is used to divide an image into several images. Flatten layer is used to flatten the dimensions of a picture after it has been convolved and finally the trained model is obtained. A total of 3900 sample images have been used.

Findings: The Plant village dataset's sample picture arrangements and meta-information conveyance play an important role in the proposed model's efficient planning and operation. The exhibition and exactness of the profound neural organization to be generated will be directly affected by morphological highlights, colouring form, and surface-based highlights.

Originality/value: It is focused in, how picture from given dataset (prepared dataset) in field and past informational collection utilized foresee the example of plant sicknesses utilizing CNN model. This brings a portion of the following experiences about plant leaf sickness forecast. Likewise, this framework thinks about the previous creation of information which will enable the rancher to get understanding the interest and the expense of different plants in market.