Classification of Different Diseases in Crop Leaves Using Deep CNN

Le résumé fourni par la source

Plant diseases continue to threaten global food security by severely undermining crop yield and quality. Traditional diagnostic methods based on visual inspection by experts are inherently limited in scalability and subject to human error, particularly in resource-constrained settings. In this study, we propose a lightweight Convolutional Neural Network (CNN) model for the automated classification of foliar diseases using the PlantVillage dataset, which comprises over 54,000 annotated images spanning 38 classes. The model architecture, inspired by MobileNetV2 and EfficientNet-B0, integrates batch normalization, dropout regularization, and data augmentation to enhance generalization and mitigate overfitting. Transfer learning was employed to leverage pretrained weights from established models, enabling improved performance with limited data. Experimental evaluation demonstrated outstanding classification metrics, including an overall accuracy of 98.64%, macro-averaged precision, recall, and F1-score exceeding 98%, and AUC values above 0.99 across all classes. The model achieved competitive results while maintaining a reduced computational footprint, making it suitable for deployment on mobile devices and UAVs for real-time field diagnostics. These findings highlight the potential of lightweight deep learning models to democratize plant disease detection and advance precision agriculture practices, especially in low-resource environments.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.