An Empirical Performance Analysis of Multi-Classification of Diseases of Tomato Leaf using CNN Models in the Deep Learning

Abstract

Tomato farming in India, producing tomatoes is one of the leading productions and stands second-largest producer of tomatoes in the world. Tomato farming has been facing challenges as the crop is susceptible to tomato diseases that include Bacterial_Spot, Early_Blight, Septoria_Leaf Spot, Spider_Mites and Late_Blight, that accounts to massive decline in the crop production. The significant drop in the production raises alarm in the analysis of the leaf of tomato with adoption of state of art technologies into the farming. The analysis of tomato leaf with the intent of early prediction of particular disease, includes employment of Convolutional Neural Network (CNN) Models include LetNet5, ResNet50 and AlexNet of the Deep Learning. The proposed work employed the kaggle database tomato leaf diseases dataset that contains 10,000 images that consist of healthy leaves and disease affected leaves. Deep Learning includes Convolutional Neural Networks models: LetNet5, ResNet50, AlexNet are applied on the  disease affected and healthy leaves of tomato dataset and it is performed empirical analysis of the CNN models in the prediction of diseases of leaf of tomato through metrics related to performance such as F1-Score, Accuracy, Precision, Recall. The proposed work which highlights empirical performance analysis of the CNN models: LetNet5, ResNet50, AlexNet, provided the noteworthy result that ResNet50 model is able to perform multi-classification the tomato leaf diseases with better accuracy 0.98701 and F1-score 0.98932.