2026· IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing· Vol 19, pp. 30875-30901· 0 citations· 189 references
TL;DR
This survey discusses the main challenges in current research, including cross-scene generalization, computational efficiency, multiview fusion effectiveness, robustness to uncertainty, as well as semantic interpretability, and further outlines potential directions for future research.
Abstract
Hyperspectral image (HSI) contains rich spectral information as well as spatial structural information, providing an important data source for a wide range of remote sensing applications. However, its high dimensionality, band redundancy, and the high cost of annotation limit the practical application of traditional methods that rely on supervised information. As a typical unsupervised learning task, clustering has received extensive attention. In recent years, with the rapid development of deep learning, deep clustering has significantly improved the unsupervised feature modeling capability for HSI by combining deep representation learning with clustering analysis. Moreover, it has gradually evolved from single-view modeling to multiview collaborative modeling, becoming an important research trend in this field. This article presents a systematic survey of deep clustering for HSI. It first introduces the basic concepts of HSI, deep clustering, and multiview clustering. Then, following the technical evolution from single-view to multiview paradigms, existing methods are systematically reviewed and categorized. Next, mainstream public datasets, commonly used evaluation metrics, and experimental comparisons of representative methods are summarized. Finally, this survey discusses the main challenges in current research, including cross-scene generalization, computational efficiency, multiview fusion effectiveness, robustness to uncertainty, as well as semantic interpretability, and further outlines potential directions for future research.
The hyperspectral images (HSIs) record subtle spectral responses of land-cover materials over a large number of contiguous bands, providing rich information for unsupervised land-cover analysis. However, local environmental variations, spectral ambiguity, and band redundancy often make the underlying clustering structu...
Hui-Lang Xu, Kai-Yu Lin, Pan Gao et al.· IEEE Transactions on Geoscie...· 0 citations
Hyperspectral imaging (HSI) captures rich spectral information across hundreds of contiguous narrow wavelength bands, enabling fine-grained material characterization beyond conventional imaging modalities. With the rapid development of sensor technologies and data-driven algorithms, HSI has been widely adopted in diver...
Xiao-Yan Luo, Lei Zhang, Jia-Lie Shen et al.· Journal of Electronic Imagin...· 0 citations
This work proposes a four–stage hybrid framework for hyperspectral image classification that integrates spectral modeling and spatial feature extraction within a unified pipeline and compares the model with recent baselines, including DaViT.
Although deep learning methods have achieved significant progress in hyperspectral image (HSI) classification, existing approaches typically focus on modeling a single dependency view, such as local spatial–spectral, structural relational, or global contextual view, making it difficult to coordinate complementary views...
Shu-Jie Ding, Xiao-Li Ruan, Jing Yang et al.· IEEE Transactions on Geoscie...· 0 citations
Joint classification of multimodal remote sensing data, such as hyperspectral image (HSI) and light detection and ranging (LiDAR), is important for Earth observation. Early deep learning-based methods usually adopt a single network to extract the features of HSI and LiDAR (HSI-LiDAR) data, respectively. However, this k...
Bin Zhao, Wen-Jie Cao, Lin Han et al.· IEEE Transactions on Geoscie...· 1 citation
Hyperspectral images (HSIs) usually contain hundreds of highly correlated spectral bands, which may introduce severe redundancy, high computational cost, and the Hughes phenomenon. Band selection (BS) reduces spectral dimensionality by selecting a compact subset of original bands while preserving their physical interpr...