A Partcle Swarm Optimization Based Approach for Filter Pruning in Convolution Neural Network for Tomato Leaf Disease Classification

Abstract

Since, plant diseases are clearly visible in leaf so, leaves images can be easily used for detection of the disease. Recent research work shows that several machine learning (ML) and deep learning based methods can be used for the classification of images into various classes. Hence in this paper a comparison has been made between machine learning, proposed convolution neural network (CNN) and pre-trained models for the classification of tomato diseases. However, the proposed CNN based model has also been compressed particle swarm optimization based approach so that model can be deployed on devices having less computation power and storage space. For training the model tomato leaf images have been taken from PlantVillage dataset. In PlantVillage dataset, there are 39 classes for various crop but we have used data related to Tomato crop. In tomato crop dataset, there are nine diseased and 1 healthy class. The best accuracy of model using proposed CNN is 98.4% and 94.9% using k-NN tradional ML method. Pre-trained models gives best accuracy of 93.5% using VGG16. The pre-trained CNN models were compressed using Particle Swarm Optimization technique and a compression of around 60% was obtained on VGG16 model size without loss in accuracy. Similarily, the proposed CNN model has also been compressed by 40% with a drop in accuracy of less than 1%. © 2022, Springer Nature Switzerland AG.