Explainability In Hyperspectral Image Classification: A Study of Xai Through the Shap Algorithm
Amir Hosein Oveis, Elisa Giusti, Giulio Meucci, Selenia Ghio, Marco Martorella
Abstract
Hyperspectral Imaging (HSI) captures complex spectral signatures across a broad range of electromagnetic wavelengths, offering diverse applications from remote sensing to medical diagnostics. HSI classification assigns labels to pixels in HSI cubes, traditionally using handcrafted features and, more recently, by deep neural networks (DNNs). However, DNNs’ inherent lack of interpretability has driven the growing adoption of eXplainable AI (XAI) techniques, aimed at explaining the decision-making process. This study integrates the SHapley Additive exPlanations (SHAP) algorithm with a convolutional neural network (CNN) for HSI classification, facilitating a per-patch understanding of class contribution. Validated using the Indian Pines dataset, the framework enhances understanding of class roles in HSI classification through XAI.