Article Details

Abstract

To address the challenge of mixed-pixel decomposition for hyperspectral imagery (HSI) in complex backgrounds, this paper proposes a Swin Transformer-based generative adversarial unmixing network (SWGU-Net). Based on joint spatial-spectral feature extraction of hyperspectral images, the Swin Transformer architecture serves as the backbone of the generator in the generative adversarial network, enhancing the model's multi-scale feature extraction capability. Additionally, this paper designs a dual-factor enhanced discriminator integrating spectral and abundance information to improve the discriminative ability during the adversarial game. Experimental results demonstrate that compared with other methods, the proposed algorithm achieves an average RMSE reduction of 53% and an average SAD reduction of 54.9% across datasets, effectively improving the accuracy of mixed-pixel decomposition. Ablation studies validate the rationality and effectiveness of the proposed architecture.

Keywords

hyperspectral imagery; Swin Transformer; generative adversarial unmixing network; dual-factor enhanced discriminator; mixed-pixel decomposition

References

1. [1] Ghosh P，Roy S K，Koirala B，et al. Hyperspectral Unmixing Using Transformer Network [J].IEEE Transactions on Geoscience and Remote Sensing，2022，60：1-16.
2. [2] Scilla D，Angulo V，Johansen K，et al. S3ViT: Self-supervised Spectral vision Transformer Framework for Hyperspectral Unmixing [J/OL].Frontiers in Remote Sensing，2026，7：1812755.https://doi.org/10.3389/frsen.2026.1812755.
3. [3] Wang C T，Gao J C，Zhu F，et al. DTU-Net: A Multi-Scale Dilated Transformer Network for Nonlinear Hyperspectral Unmixing [J].IEEE Transactions on Geoscience and Remote Sensing，2026，64：1-17.
4. [4] Chen J Y，Jiang M Y，Su Y C，et al. A Swin Transformer-Based Hybrid U-Shape Network for Nonlinear Hyperspectral Unmixing [J].IEEE Geoscience and Remote Sensing Letters，2026，23：1-5.
5. [5] Chen J F，Yang C，Zhang L，et al. Tccu-net: Transformer and CNN collaborative unmixing network for hyperspectral image [J].IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing，2024，17：8073-8089.
6. [6] Ge Y，Han L，Paoletti M E，et al. Transformer-Enhanced CNN based on Intensive Feature for Hyperspectral Unmixing [J].IEEE Geoscience and Remote Sensing Letters，2024，21：1-5.
7. [7] Meng F L，Sun H X，Li J，et al. CTNet: an Efficient Coupled Transformer Network for Robust Hyperspectral Unmixing [J].International Journal of Remote Sensing，2024，45（17）：5679-5712.
8. [8] Wang Y，Shi S K，Chen J. Efficient Blind Hyperspectral Unmixing with Non-Local Spatial Information Based on Swin Transformer [C]//IGARSS 2023-2023 IEEE International Geoscience and Remote Sensing Symposium，2023：5898-5901.
9. [9] Yang Z R，Xu M M，Liu S W，et al. Spatial-spectral attention bilateral network for hyperspectral unmixing [J].IEEE Geoscience and Remote Sensing Letters，2023，20：1-5.
10. [10] Gao L R，Han Z，Hong D F，et al. CyCU-Net: Cycle-Consistency Unmixing Network by Learning Cascaded Autoencoders [J].IEEE Transactions on Geoscience and Remote Sensing，2021，60：1-14.
11. [11] Ozkan S，Kaya B，Akar G B. EndNet: Sparse AutoEncoder Network for Endmember Extraction and Hyperspectral Unmixing [J].IEEE Transactions on Geoscience and Remote Sensing，2019，57（1）：482-496.
12. [12] Drumetz L，Meyer T R，Chanussot J，et al. Hyperspectral Image Unmixing With Endmember Bundles and Group Sparsity Inducing Mixed Norms [J].IEEE Transactions on Image Processing，2019，28（7）：3435-3450.
13. [13] Zhuang L N，Chia-Hsiang L，Figueiredo M A T，et al. Regularization Parameter Selection in Minimum Volume Hyperspectral Unmixing [J].IEEE Trans. Geoscience and Remote Sensing，2019，57（12）：9858-9877.

Funding

Key Laboratory Project of the Ministry of Natural Resources (NRMSSHR2023Y16)

About the authors

Li Jian (1985-), male, Han ethnicity, from Huanggang, Hubei, is a senior engineer and doctoral student. His research direction is pattern recognition and intelligent systems.