Explainable Deep Learning for Multi-Class Plant Disease Classification Using ResNet and EfficientNet with Grad-CAM Analysis

Abstract

Plant diseases can reduce crop quality and productivity, making early detection an important aspect of modern agriculture. Recent advances in deep learning, particularly Convolutional Neural Networks (CNN), have shown promising performance in image-based plant disease classification. This study proposes an explainable deep learning approach for multi-class plant disease classification using ResNet50 and EfficientNetB0 combined with Grad-CAM visualization. The experiments were conducted using the PlantVillage dataset consisting of 15 classes of healthy and diseased plant leaves.The research process included image preprocessing, data augmentation, transfer learning, model training, performance evaluation, and explainability analysis. The dataset was divided into training and validation sets with a ratio of 80:20. Model performance was evaluated using accuracy, loss, confusion matrix, precision, recall, and f1-score metrics. Experimental results showed that ResNet50 achieved the best performance with an accuracy of 92% and a validation loss of 0.19, outperforming EfficientNetB0 which obtained 76% accuracy and 0.82 validation loss. The classification report demonstrated that ResNet50 provided more stable and consistent predictions across most disease classes. Furthermore, Grad-CAM visualization successfully highlighted disease-relevant regions such as lesions, discoloration, and damaged leaf areas, improving the interpretability of the CNN model. The findings indicate that the combination of ResNet50 and Grad-CAM is effective for plant disease classification and provides better explainability for deep learning-based agricultural applications.