Ship Detection in PolSAR Images Based on Structure Consistency-Based Graph

Polarimetric synthetic aperture radar (PolSAR), which is capable of characterizing the polarimetric scattering mechanism of targets, significantly enhances the performance of ship detection through polarization features. However, the complexity of sea clutter and the diversity of ship scattering structures can lead to overlaps in polarization characteristics, causing missed detection or false alarms. Instead of relying solely on the polarization information of individual pixels, we propose a novel approach that focuses on the polarization differences between the target and its surrounding pixels. Distinct scatters’ polarimetric differences cause variations in relative amplitude relationships among pixels within local patches, which significantly differ between clutter and target regions. Furthermore, these relative amplitude relationships are linked to the image’s topological structure, and their variations can be characterized using a K-nearest neighbors (K-NN) graph. Hence, we introduce a ship detection method based on graph structural consistency (GSC). By constructing local similarity graphs, the structures of images from different polarization channels are projected into the same domain for comparison, facilitating the extraction of graph structural differences across polarizations. These differences serve as features for ship detection. Experimental results with Gaofen-3 and Radarsat-2 PolSAR datasets demonstrate that the proposed method achieves superior detection performance than state-of-the-art ship detection methods. The graph structural feature method utilizes relative polarimetric information among pixels, offering a more robust solution for ship detection under complex sea conditions.