OMPARATIVE CNN MODELS FOR PLANT LEAF DISEASE DETECTION

Abstract

Automated plant disease detection has gained traction in precision agriculture for its potential to boost crop yields and reduce manual effort. This study explores vegetable disease classification using convolutional neural networks (CNNs), utilizing a dataset of over 20,000 images across 15 disease categories and healthy classes. Four models were evaluated: a custom CNN, VGG19, ResNet50, and Xception. The custom CNN achieved 87.50% accuracy, demonstrating the potential of lightweight models for resource-limited scenarios. VGG19 performed better with 89.52%, while ResNet50 outperformed all with 94.86% accuracy and strong precision, recall, and F1 score. Xception underperformed, emphasizing the importance of model architecture and tuning. The results highlight the effectiveness of transfer learning, especially with deep models like ResNet50, and the viability of custom CNNs in constrained environments. Future work will explore class-specific errors, edge-device optimization, and techniques to address class imbalance for improved robustness.