Explainable ensemble deep learning for potato leaf pest and diseases identification

Abstract

Potatoes are one of the most important food crops globally, yet their susceptibility to a wide range of diseases poses a serious threat to agricultural productivity. Conventional methods for identifying potato diseases are labor-intensive and time-consuming, underscoring the need for more advanced solutions. Deep learning has emerged as a promising alternative, outperforming traditional methods with its ability to automate and enhance potato disease identification. However, previous research has primarily focused on single-dataset implementations, and the interpretability of these models remains insufficiently explored. To address this, we introduce a solution that combined the strengths of ensemble deep learning and Explainable AI (XAI) for the identification of potato leaf pests and diseases. Our ensemble model integrates MobileNetV3-Large and EfficientNetV2B3 architectures and demonstrates remarkable accuracy of 97.91%. Furthermore, the incorporation of XAI techniques greatly enhances the interpretability of the model. By improving both interpretability and predictive accuracy, these results support more informed and reliable model outputs.