Advances in hyperspectral image unmixing: From algorithmic frameworks to practical applications

Hyperspectral unmixing (HU) is a cornerstone of hyperspectral image (HSI) analysis, providing both theoretical depth and practical utility. It addresses the mixed pixel problem by decomposing observed spectra into constituent endmembers and their corresponding abundances. Existing reviews largely focus on algorithmic developments, while the broader impact of HU on practical applications has received less attention. This paper provides a comprehensive survey of recent advances in HU, systematically examining representative algorithms and exploring their practical applications. The review begins with foundational concepts of the linear mixing model (LMM), followed by a structured analysis of LMM-based methods and techniques designed to handle spectral variability (SV) in HU. The effectiveness of HU in supporting downstream HSI analysis tasks, such as superresolution, reconstruction, classification, segmentation, anomaly detection, target detection, and change detection, is explored in detail. In addition, the review highlights diverse real-world applications of HU, such as terrestrial and planetary mineral mapping, land cover classification, coastal wetland monitoring, and natural disaster assessment. Finally, current challenges, future research opportunities, and the enduring theoretical and practical significance of HU are discussed, advocating for greater recognition and sustained research in this critical field. • Reviews HU from LMM-based to SV modeling approaches. • Demonstrates HU’s effectiveness in supporting seven downstream HSI analysis tasks. • Summarizes HU’s practical application in diverse real-world scenarios. • Outlines HU’s challenges and future directions.