The rapid evolution of computing technologies has enabled the development of advanced deep learning (DL) techniques, which have shown significant potential in Earth observation (EO) applications for analyzing remotely sensed data. Hyperspectral imaging (HSI), while offering rich spectral and spatial information for detailed surface characterization, presents substantial challenges due to its high dimensionality and data volume. Recently, DL-based classification methods have attracted considerable research attention for their ability to automatically learn discriminative spectral-spatial features from HSI data. This article presents a comprehensive review of current DL approaches for HSI classification, systematically examining their strengths, limitations, and application domains. Key challenges associated with DL-based HSI analysis, including limited labelled data and high computational demands, are also discussed, along with potential strategies to address them. The reviewed methods demonstrate wide applicability across diverse classifications such as pixel-wise semantic (agriculture, forestry, and environmental monitoring), scene level (disaster management), target detection (mineral exploration), and compressed sensing. Advances in DL have significantly improved hyperspectral image classification accuracy, enabling deeper insights into complex environmental systems and supporting innovative solutions for natural resource management.
Sudharsan et al. (Fri,) studied this question.