Comparative Analysis of Fruits Disease Detection using Different types of CNN Architecture

Abstract

Fruit diseases faces significant challenges to agricultural productivity, leading to substantial economic losses globally. Early and accurate detection of these diseases is crucial for effective management and prevention. In recent years, Deep Learning, particularly Convolutional Neural Networks (CNNs), has emerged as a powerful tool for image classification and disease detection in agriculture. This project focuses on a comparative analysis of various CNN architectures for detecting diseases in fruits such as guava, mango, and orange. The study involves seven specific fruit diseases, including anthracnose in guava and mango, black mould rot in mango, black spot in orange, greening in orange, fruit fly in guava, and stem rot in mango. A dataset comprising 11 classes with 5,091 images was utilized to train and evaluate six CNN architectures: a 5-layer CNN, LeNet, VGG16, MobileNet, Inception V3, and ResNet. The experimental results reveal that Inception V3 achieved the highest test accuracy of 96.71%, followed by VGG16 with 93.41% and LeNet with 90.47%. MobileNet, ResNet, and the 5-layer CNN recorded accuracies of 88.21%, 88.91%, and 84.06%, respectively. These findings highlight the superior performance of Inception V3 for fruit disease detection tasks and underline the potential of deep learning in developing efficient and automated solutions for agricultural applications. This research contributes to the field by identifying the strengths and limitations of different CNN architectures, providing valuable insights for future advancements in fruit disease detection systems.