EUVIMEDEUVIMED entdecken
Crossref

DIR-HSI: Degradation-Invariant Representation Learning for Hyperspectral Image Classification

Boxiang Yang, Ning Chen, Yingbo Fan, Zexi Shi, Haoyuan Zhang, Hui Liu, Run Niu, Shanjun Mao

IEEE Transactions on Geoscience and Remote Sensing · 2026 · 64 · 1-22

DOI: 10.1109/tgrs.2026.3687650 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Geoscience and Remote Sensing, 2026. Verfasst von Boxiang Yang und 7 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 64 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Postdoctoral Research Foundation of China, Project of Sihong County Active Fault Detection and Earthquake Hazard Assessment.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In recent years, hyperspectral image (HSI) classification has achieved remarkable progress in fine-grained surface recognition and intelligent remote sensing analysis. However, real-world degradations—such as sensor noise, compression artifacts, and atmospheric interference—often introduce severe domain shifts, leading to degraded robustness and limited generalization of existing models. To address this challenge, we propose DIR-HSI, a degradation-invariant hyperspectral classification framework designed to enhance feature stability and discriminability under multidegradation conditions. The framework comprises two key modules: the degradation-invariant representation by Fourier-domain feature extraction (DIR-FFE) module, which leverages Fourier-domain feature decomposition to extract discriminative and degradation-insensitive frequency features, and the degradation-invariant representation contrastive enhancement (DIR-CE) module, which employs cross-degradation supervised contrastive (SupCon) learning to enforce consistency among same-class representations across different degradation conditions. Together, these components enable effective disentanglement of degradation-sensitive and stable factors, maintaining feature consistency and reinforcing class separability. The key advantage of DIR-HSI lies in its ability to jointly model degradation-invariant and discriminative representations, achieving robust and generalizable performance across complex degradation scenarios. Extensive experiments on multiple benchmark datasets demonstrate that DIR-HSI consistently outperforms state-of-the-art methods in terms of accuracy, robustness, and generalization. The code will be available at https://github.com/yangboxiang1207/DIR-HSI

Förderung

  • National Natural Science Foundation of China
  • Postdoctoral Research Foundation of China
  • Project of Sihong County Active Fault Detection and Earthquake Hazard Assessment

Zitieren

APA

Yang, B., Chen, N., Fan, Y., Shi, Z., Zhang, H., Liu, H., Niu, R., & Mao, S. (2026). DIR-HSI: Degradation-Invariant Representation Learning for Hyperspectral Image Classification. IEEE Transactions on Geoscience and Remote Sensing, 64, 1-22. https://doi.org/10.1109/tgrs.2026.3687650

Vancouver

Yang B, Chen N, Fan Y, Shi Z, Zhang H, Liu H, et al. DIR-HSI: Degradation-Invariant Representation Learning for Hyperspectral Image Classification. IEEE Transactions on Geoscience and Remote Sensing. 2026;64:1-22. doi:10.1109/tgrs.2026.3687650

BibTeX
@article{Yang2026,
  author = {Yang, Boxiang and Chen, Ning and Fan, Yingbo and Shi, Zexi and Zhang, Haoyuan and Liu, Hui and Niu, Run and Mao, Shanjun},
  title = {DIR-HSI: Degradation-Invariant Representation Learning for Hyperspectral Image Classification},
  journal = {IEEE Transactions on Geoscience and Remote Sensing},
  year = {2026},
  volume = {64},
  pages = {1--22},
  doi = {10.1109/tgrs.2026.3687650}
}
Kostenlos starten →