A Comparative Study of Convolutional Neural Network Architectures for Enhanced Tomato Leaf Disease Classification Using Refined Statistical Features

Tomatoes, a staple in culinary practices, are currently in high demand yet low supply in India, rendering them unaffordable to the general population.This issue largely stems from the inability of farmers to identify and control prevalent tomato leaf diseases, leading to significant crop losses.Early detection and classification of leaf diseases are paramount to mitigate this problem, thereby boosting crop productivity.Despite extensive research in this domain, the precise localization and identification of various tomato leaf diseases present a complex task.This complexity arises from the significant overlap between the healthy and diseased portions of the leaves.The process is further complicated by the minimal contrast between the background and foreground of the specimen under investigation.To address these challenges, this study conducts a comprehensive performance analysis of several Convolutional Neural Networks (CNNs) models, namely, ResNet-152, ResNet-101, VGGNet, Alex Net, and LeNet, applied to the PlantVillage dataset.The results indicate that the ResNet-152 and ResNet-101 models yield superior accuracy rates when applied to both full-resolution source images and their background-removed counterparts.The performance outcomes reported herein surpass those documented in the existing literature, demonstrating the potential for significant advancements in the early detection and classification of tomato leaf diseases.