"A Comparative Analysis of Deep Learning Algorithms for Fruit Disease Classification"

Abstract

Agriculture plays a pivotal role in the Indian economy, serving as the primary source of income for a substantial portion of the population. Enhancing fruit output is crucial for sustaining agricultural livelihoods. However, fruit diseases, mainly caused by fungal and bacterial pathogens, significantly impact fruit quality and overall yield. Timely identification of these diseases is essential for forecasting and mitigating their occurrence, leading to cost savings for farmers. Researchers have developed fruit disease identification systems to safeguard agricultural investments. This study aims to conduct a comprehensive comparative analysis of deep learning classification approaches for fruit disease detection. We evaluate the performance of VGG16, InceptionV3, MobileNetV2, ResNet50, NasNetV2, and a Convolutional Neural Network (CNN) model. Additionally, we incorporate optimization techniques such as Stochastic Gradient Descent with Momentum (SGDM), Adaptive Moment Estimation (Adam), and Root Mean Square Propagation (RMSProp), along with baseline learning techniques and transfer learning methods. In this study, we evaluate the performance of the ResNet50 architecture in fruit disease detection, achieving an impressive accuracy of 93.25%. This performance is compared to other deep learning architectures, including VGG16, InceptionV3, MobileNetV2, and NasNetV2. Our findings highlight the effectiveness of ResNet in accurately detecting fruit diseases, showcasing its potential as a valuable tool for farmers in mitigating the impact of diseases during the early stages of infestation.