Hyperspectral image (HSI) classification plays a vital role in remote sensing,environmental monitoring, agriculture, medical imaging, and defense intelligence due to its abilityto capture detailed spectral signatures across hundreds of contiguous wavelength bands. Althoughrecent advances in machine learning and deep learning have significantly improved classificationaccuracy, most existing approaches remain predominantly machine-centric, prioritizing predictiveperformance while overlooking interpretability, transparency, and human trust. These limitationshinder real-world adoption, particularly in high-stakes domains where expert validation andaccountability are essential. This paper presents a Human-Centered Artificial Intelligence (HCAI)framework for the classification of segmented hyperspectral image data. The proposed approachintegrates spectral–spatial feature modeling, affinity propagation–based segmentation, a modifiedExtreme Learning Machine (ELM) classifier, and a human-in-the-loop feedback mechanismsupported by explainable AI (XAI) techniques. Mathematical formulations are introduced tomodel spectral–spatial fusion, segmentation similarity, classifier optimization, and expertfeedback integration.Experimental evaluation conducted on benchmark hyperspectral datasets demonstrates thatthe proposed framework achieves improved classification accuracy, robustness under limitedlabeled data conditions, and enhanced interpretability compared to conventional machine-centricbaselines. The results highlight that embedding human-centered principles not only strengthenspredictive performance but also improves transparency, trustworthiness, and deploymentreadiness. This work contributes a novel interdisciplinary perspective by unifying hyperspectralimaging, lightweight machine learning, and human-centered AI, offering a scalable andresponsible solution for complex image classification tasks.
Patel et al. (Sat,) studied this question.