Information Bottleneck-Guided Enhancement of Decomposition-Derived Scattering Features for PolSAR Image Classification
IEEE Transactions on Geoscience and Remote Sensing · 2026 · 64 · 5218014-5218014
DOI: 10.1109/tgrs.2026.3743369 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Geoscience and Remote Sensing, 2026. Verfasst von Zhaoquan Wang und 4 weiteren Autor:innen. Die Arbeit nennt 36 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Heilongjiang Postdoctoral Fund.
Abstract (Originalsprache)
Model-based polarimetric target decomposition derives physically interpretable scattering features from predefined scattering models and has been widely used for polarimetric synthetic aperture radar (PolSAR) image classification. However, the idealized assumptions of these models may be inconsistent with real PolSAR targets, especially when man-made structures are not aligned with the radar line of sight. This mismatch can lead to scattering ambiguity in decomposition-derived features, weakening feature discriminability and physical interpretability, and thereby restricting data-driven PolSAR classification. To address this issue, an information bottleneck-guided plug-and-play framework is proposed to enhance decomposition-derived scattering features. The framework couples semantic discrimination with physics-guided scattering constraints, thereby connecting deep discriminative feature learning with physically interpretable modeling without introducing additional scattering models. Experiments on multiple airborne and spaceborne PolSAR datasets demonstrate clearer class separation, more physically consistent scattering behavior, and improved classification performance.
Förderung
- National Natural Science Foundation of China
- Heilongjiang Postdoctoral Fund
Zitieren
APA
Wang, Z., Zhuang, D., Zhang, L., Zou, B., & Takeuchi, W. (2026). Information Bottleneck-Guided Enhancement of Decomposition-Derived Scattering Features for PolSAR Image Classification. IEEE Transactions on Geoscience and Remote Sensing, 64, 5218014-5218014. https://doi.org/10.1109/tgrs.2026.3743369
Vancouver
Wang Z, Zhuang D, Zhang L, Zou B, Takeuchi W. Information Bottleneck-Guided Enhancement of Decomposition-Derived Scattering Features for PolSAR Image Classification. IEEE Transactions on Geoscience and Remote Sensing. 2026;64:5218014-5218014. doi:10.1109/tgrs.2026.3743369
BibTeX
@article{Wang2026,
author = {Wang, Zhaoquan and Zhuang, Di and Zhang, Lamei and Zou, Bin and Takeuchi, Wataru},
title = {Information Bottleneck-Guided Enhancement of Decomposition-Derived Scattering Features for PolSAR Image Classification},
journal = {IEEE Transactions on Geoscience and Remote Sensing},
year = {2026},
volume = {64},
pages = {5218014--5218014},
doi = {10.1109/tgrs.2026.3743369}
}
EUVIMED