An Optimized Plant Disease Classification System Based on Resnet-50 Architecture and Transfer Learning

Automated plant disease diagnosis is a vital task in modern agriculture to ensure timely and effective management and reduce crop losses. This paper proposes a deep learning-based plant disease classification model using ResNet-50 architecture, which offers several advantages over existing methods. ResNet-50 allows accurate identification of plant diseases from images with high precision and recall, even when the diseases have similar visual characteristics. The ResNet-50 architecture has skip connections that help in reusing the learned features from previous layers, overcoming the problem of vanishing gradients and allowing it to train on deeper networks, resulting in improved accuracy compared to shallower models. Furthermore, transfer learning is utilized in the proposed model, enabling the use of pre-trained weights from a large dataset of general images, followed by fine-tuning on the plant disease dataset. This enhances the model’s performance by leveraging the pre-trained weights’ knowledge, leading to faster convergence, improved accuracy, and reduced training time. The model is trained and tested on a comprehensive dataset from 38 different categories of plant diseases and healthy leaves. This ensures that the model is robust and can accurately identify and classify diseases across various plant species. By leveraging the advantages of the ResNet-50 architecture and transfer learning, the proposed model can accurately and efficiently diagnose plant diseases, leading to improved agricultural productivity and reduced losses due to plant diseases.