BauXAIte: An Explainable AI Framework for UAV Multisensor-Driven Bauxite Stockpile Mapping and Classification

An explainable artificial intelligence (AI) framework for bauxite stockpile classification through uncrewed aerial vehicle (UAV)-driven mapping, BauXAIte, is introduced for efficient mineral resource assessment and monitoring. A UAV, equipped with RGB and multispectral sensors, is used to capture data within a broad spectrum range, allowing the analysis of the bauxite stockpiles spectral signatures. Two distinct datasets, RGB and multispectral, were created by utilizing photogrammetric tools to generate the orthophoto map and GIS for annotating the bauxite types. These datasets along with their combination, were used to train a pool of machine and deep learning models to classify spectral signatures into bauxite types. Experimental results demonstrated that, using a multilayer perceptron (MLP), the classification performance exceeded 91% across all standard metrics. To ensure model transparency, Kernel SHapley Additive exPlanations (SHAP) results accompanied the AI predictions, to indicate correlations between spectral bands and classification decisions, thereby supporting end-users toward practical deployment of bauxite resource management.