A Novel Knowledge-Data Co-Driven Pol-SAR Target Identification Method

Effectively identifying targets of interest in Polarimetric Synthetic Aperture Radar (Pol-SAR) systems is crucial. Although Convolutional Neural Network (CNN) based methods have shown promising results in Pol-SAR target recognition, challenges such as poor network interpretability and decreased accuracy due to strong ground and sea clutter persist. To tackle these issues, we introduce a knowledge-data co-driven method that utilizes polarimetric scattering information to constrain network learning. It focuses the network's attention on the target, thereby reducing the impact of clutter on target recognition. Our experiments on a constructed Pol-SAR dataset demonstrate that this method significantly enhances target recognition accuracy without increasing computational complexity during inference, thus proving its effectiveness and robustness.