Classification of Healthy and Unhealthy Leaf Disease Using Modified MobileNetV2

A study employing Convolutional Neural Network (CNN) architectures, specifically MobileNetV2 band ResNet-50, compares their performance in classifying tomato leaf diseases, aiming to address the substantial challenges faced by farmers due to these diseases. Previously, various kinds of diseases belonging to tomatoes were identified using the model, which leads to lesser accuracy and would require more computational energy. The dataset utilized for the research consists of 1900 healthy and 1800 unhealthy tomato plant images which includes all the diseases, with an additional 500 healthy and 1080 unhealthy plant images set aside for validation. MobileNetV2, renowned for its lightweight design, achieves an impressive accuracy of 95.3% after customizing the model with various layers makes it unique from the other existing models. While ResNet-50, known for its deeper architecture, attains a lower accuracy of 51%. The selection of MobileNetV2 as the preferred model is based on its superior accuracy and computational efficiency. The study includes analysis utilizing backpropagation to understand weight update trends within the chosen models. These results highlight the potential of deep learning approaches, specifically MobileNetV2 and ResNet-50, in automated disease diagnosis in agriculture, offering solutions to mitigate crop losses and enhance productivity in tomato cultivation. Overall, the study contributes significantly to advancing the field of automated disease diagnosis in agriculture by leveraging deep learning techniques, particularly MobileNetV2 and ResNet-50, for the classification of tomato leaf diseases. The insights gained from this research have profound implications for improving crop yield and sustainability in tomato cultivation practices, ultimately aiding in the advancement of agricultural technology and practices.