SCATTERING MECHANISM-GUIDED CLASSIFICATION OF POLSAR IMAGE VIA MULTI-BAND FUSION

Deep learning methods have shown significant success in single-band PolSAR image classification. However, single-band methods are limited by insufficient target representation, causing scattering mechanism ambiguity. Multi-band fusion addresses this issue by leveraging complementary polarimetric information across bands. In this paper, we propose the Scattering Mechanism-Guided Fusion Network (SMGFN) dedicated to PolSAR, which integrates scattering mechanisms as the guiding principle for multi-band fusion. Specifically, we introduce a Band-Aware Module (BAM) that performs multiband feature extraction across spatial and channel dimensions, and adaptively allocates weights. This enables the effective capture of complementary scattering behaviors. Additionally, the Multi-Band Fusion Module (MBFM) is designed to fuse multiband features through diverse strategies and scales, generating comprehensive and discriminative scattering representations for classification. Experimental results on two widely used PolSAR datasets demonstrate that SMGFN outperforms existing methods, achieving state-of-the-art classification performance.