EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

SeqSAN: A Novel Sequencer Self-Attention Network for Hyperspectral Image Classification

Hanlin Feng, Chengcheng Zhong, Zitong Zhang, Yichen Liu

IEEE Access · 2026 · 14 · 37538-37552

DOI: 10.1109/access.2026.3671989 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Hanlin Feng und 3 weiteren Autor:innen. Die Arbeit nennt 43 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, Postdoctoral Fellowship Program and China Postdoctoral Science Foundation.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The Vision Transformer (ViT) leverages a self-attention framework to effectively learn representations, achieving state-of-the-art performance in hyperspectral image (HSI) recognition. However, this comes with the drawback of higher computation under quadratic complexity. Meanwhile, the encoding stage of ViT emphasizes global feature modeling while neglecting local feature information, which restricts the model’s performance. To overcome these drawbacks, we introduce a Sequencer Self-Attention Network (SeqSAN) designed for HSI recognition. Our method incorporates a Spectral-Spatial Sequencer (S3) to enhance local representation learning and a Two-stage Self-Attention (TSA) design to model global correlations while better accommodating the anisotropy of HSI features. Experimental results on four public HSI datasets demonstrate that SeqSAN outperforms current deep learning models in terms of classification accuracy and generalization. This highlights the effectiveness of integrating self-attention with the Sequencer paradigm and offers a novel perspective on HSI classification.

Förderung

  • National Natural Science Foundation of China
  • Postdoctoral Fellowship Program and China Postdoctoral Science Foundation

Zitieren

APA

Feng, H., Zhong, C., Zhang, Z., & Liu, Y. (2026). SeqSAN: A Novel Sequencer Self-Attention Network for Hyperspectral Image Classification. IEEE Access, 14, 37538-37552. https://doi.org/10.1109/access.2026.3671989

Vancouver

Feng H, Zhong C, Zhang Z, Liu Y. SeqSAN: A Novel Sequencer Self-Attention Network for Hyperspectral Image Classification. IEEE Access. 2026;14:37538-37552. doi:10.1109/access.2026.3671989

BibTeX
@article{Feng2026,
  author = {Feng, Hanlin and Zhong, Chengcheng and Zhang, Zitong and Liu, Yichen},
  title = {SeqSAN: A Novel Sequencer Self-Attention Network for Hyperspectral Image Classification},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {37538--37552},
  doi = {10.1109/access.2026.3671989}
}
Kostenlos starten →