Multi-Level Energy Decomposition For Polsar Radiation Information With 2D-C-VMD

Polarimetric Synthetic Aperture Radar (PolSAR) imagery is widely used in land cover classification and target recognition due to its rich scattering information. However, the nonlinear, non-stationary nature of PolSAR data, along with its complex scattering characteristics, makes feature extraction a challenging task. To address this, we propose a novel feature extraction method for PolSAR data based on 2D Compact Variational Mode Decomposition (2D-C-VMD). This approach decomposes the original features into modes with narrow Fourier bandwidths and enhances feature representation by reconstructing each mode using a binary support function. Specifically, 2D-C-VMD is first applied to decompose the PolSAR data, extracting modal components from different frequency ranges. A binary support function is then used to reconstruct each mode, ensuring that each mode contributes to signal reconstruction only within its support region, thus preventing interference between modes.To validate the effectiveness of this method,classification experiments were conducted on Radarsat-2 Flevoland dataset and UAVSAR Yucatan Lake dataset, and the experimental results show that this method can significantly improve the classification accuracy.