Plant Disease Classification

Abstract

This project develops a deep learning-based solution for detecting and classifying plant diseases using a dataset of 4000 leaf images across eight classes: Anthracnose, Bacterial Canker, Cutting Weevil, Die Back, Gall Midge, Healthy, Powdery Mildew, and Sooty Mould. Addressing the urgent need for early disease detection to improve agricultural yield and sustainability, various models were evaluated, with the proposed CNN+Transfer Learning model achieving the best accuracy. The dataset was divided into training (3,200 images), validation (400 images), and test (400 images) sets, with data augmentation apply to enhance robustness. Models tested include CNN (0.75 accuracy), CNN+SVM (0.35), CNN+Fusion (0.94), CNN+Attention (0.93), CNN+Transfer Learning (0.98), DenseNet (0.96), and ResNet (0.31). The proposed CNN+Transfer Learning model, implemented using TensorFlow and Keras, leverages

pre-trained architectures, optimized with the Adam optimizer and categorical cross- entropy loss, achieving a test accuracy of 98%. Training involved 15 epochs with early

stopping and learning rate reduction to mitigate overfitting. This approach demonstrates superior performance in automating plant disease diagnosis, offering a scalable tool for precision agriculture. Future work could explore larger datasets and advanced architectures to further enhance generalization, supporting sustainable farming practices by enabling timely disease management.