Development and challenges of hyperspectral image classification techniques

Hyperspectral image classification is a pivotal task in remote sensing, leveraging the rich spatial and spectral information contained in hyperspectral images. This paper addresses the challenges inherent in hyperspectral classification, including spectral variability, band redundancy, and data scarcity. We delineate the relationship between hyperspectral classification, semantic segmentation, and target recognition, categorizing classifiers into spectral and spatial-spectral feature types. Spectral feature classifiers, ranging from traditional statistical methods to deep learning approaches, offer varying levels of performance and computational efficiency. Spatial-spectral feature classifiers, integrating spatial information, enhance classification accuracy by addressing spectral variability and noise. We discuss the strengths and limitations of different methods, highlighting the potential of deep learning-based approaches and the importance of joint spatial-spectral feature extraction. Future research should focus on overcoming the challenges associated with data acquisition, feature engineering, and model interpretability to advance hyperspectral image classification applications.