ExE-Net: Explainable Ensemble Network for Potato Leaf Disease Classification

Tasnim Ahmed, Md. Bakhtiar Hasan, Sabbir Ahmed, Md. Hasanul Kabir

Abstract: Potato cultivation faces significant harvest loss from several diseases, emphasizing the importance of early and accurate disease detection. Analyzing potato leaf images using computer vision tools and machine learning algorithms offers a promising way for early disease identification and continuous plant health monitoring. However, previous approaches in potato leaf disease classification have overlooked key aspects such as feature learning potentials of recent deep learning models, benefits of ensemble learning, and explainability techniques, leading to limitations in accuracy, interpretability, and practicality. To this end, We introduce the Explainable Ensemble Network (ExE-Net) for potato leaf disease classification, which outperforms existing methods in accuracy and generalization. Our key contributions include an ensemble architecture combining Xception, DenseNet201, and InceptionResNet to capture diverse features, and integrating explainability techniques like Grad-CAM, LIME, and SHAP to enhance transparency. Data augmentation further boosts ExE-Net’s performance, resulting in superior accuracy and robustness.