Explainable Deep Learning for Automated Crop Disease Diagnosis Using UAV-Based Hyperspectral Imaging

Abstract

Precision agriculture increasingly relies on unmanned aerial vehicles (UAVs) equipped with hyperspectral sensors for continuous crop monitoring. However, deep learning-based disease diagnosis models often lack interpretability, limiting their adoption by agricultural experts. This paper proposes an explainable deep learning framework that combines Vision Transformers (ViTs) with attention-based saliency mapping for automated crop disease identification from hyperspectral imagery. The proposed framework utilizes spectral-spatial feature extraction to accurately distinguish healthy and diseased crops while generating visual explanations that highlight disease-affected regions. Extensive experiments demonstrate superior classification performance and improved interpretability compared to conventional CNN-based approaches. The proposed solution enables transparent decision-making, supporting precision agriculture through reliable disease diagnosis, optimized pesticide usage, and increased crop productivity.