EUVIMEDEUVIMED entdecken
Crossref

Manifold-Aware Multiview Evidence Fusion for Robust PolSAR Image Classification

Junfei Shi, Haojia Zhang, Haiyan Jin, Junhuai Li, Maoguo Gong, Weisi Lin

IEEE Transactions on Circuits and Systems for Video Technology · 2026-11 · 36 (11) · 16390-16403

DOI: 10.1109/tcsvt.2026.3716446 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Circuits and Systems for Video Technology, 2026-11. Verfasst von Junfei Shi und 5 weiteren Autor:innen. Die Arbeit nennt 58 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Ministry of Education of Singapore Tier3, State Key Laboratory of Water Engineering Ecology and Environment in Arid Area.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Deep learning excels at automatic feature extraction from Polarimetric Synthetic Aperture Radar (PolSAR) data, yet it ignores the derived multi-features (e.g., scattering angle, entropy, texture), which can offer crucial, physically interpretable information for classification. Traditional fusion methods, relying on feature concatenation or standard network fusion followed by a softmax classifier, face two major limitations: 1) Geometric Incompatibility: Feature concatenation is inaccurate because PolSAR covariance matrices reside on a Hermitian Positive Definite (HPD) manifold, while the derived multi-features are best represented on a Grassmann manifold. 2) Decision Instability: The softmax classifier treats all features equally, leading to unreliable predictions, especially under speckle noise. To overcome these challenges, we propose a novel Manifold-aware Multiview Evidence Fusion Framework (MMEFNet) to integrate these two complementary feature views in a geometry-consistent and uncertainty-aware manner. To learn geometric structures, we construct two superpixel-based graph convolutional networks—HPD-sGCN and Grassmann-sGCN—each operating on its native manifold using dedicated kernel metrics. To enhance decision reliability, the conventional softmax layer is replaced with a Dempster–Shafer Theory (DST)–based evidence fusion module, which jointly models class belief and uncertainty while aggregating evidence from both views. Experimental results on three real-world PolSAR datasets demonstrate that the MMEFNet framework achieves superior performance in terms of accuracy, robustness, and interpretability compared to state-of-the-art PolSAR classification methods.

Förderung

  • National Natural Science Foundation of China
  • Ministry of Education of Singapore Tier3
  • State Key Laboratory of Water Engineering Ecology and Environment in Arid Area

Zitieren

APA

Shi, J., Zhang, H., Jin, H., Li, J., Gong, M., & Lin, W. (2026). Manifold-Aware Multiview Evidence Fusion for Robust PolSAR Image Classification. IEEE Transactions on Circuits and Systems for Video Technology, 36(11), 16390-16403. https://doi.org/10.1109/tcsvt.2026.3716446

Vancouver

Shi J, Zhang H, Jin H, Li J, Gong M, Lin W. Manifold-Aware Multiview Evidence Fusion for Robust PolSAR Image Classification. IEEE Transactions on Circuits and Systems for Video Technology. 2026;36(11):16390-16403. doi:10.1109/tcsvt.2026.3716446

BibTeX
@article{Shi2026,
  author = {Shi, Junfei and Zhang, Haojia and Jin, Haiyan and Li, Junhuai and Gong, Maoguo and Lin, Weisi},
  title = {Manifold-Aware Multiview Evidence Fusion for Robust PolSAR Image Classification},
  journal = {IEEE Transactions on Circuits and Systems for Video Technology},
  year = {2026},
  volume = {36},
  number = {11},
  pages = {16390--16403},
  doi = {10.1109/tcsvt.2026.3716446}
}
Kostenlos starten →