基于深度线结构提取的点线联合光-SAR图像配准方法OA

In order to address problems of unstable point features,fractures and mismatching of line struc-tures caused by noise interference,geometric distortions,and modal discrepancies in optical and synthetic aperture radar(SAR)image registration,this paper proposes a registration method of point-line graph-matching attention network(PLG-MAN).The method develops a few-shot SAR line structure extraction module,which integrates self-supervised pretraining,multi-scale contextual modeling,and attention mechanisms to enhance the connectivi-ty of long-line skeleton features,and furthermore,introduces a line-endpoint cross-attention mechanism into the point-line joint matching stage to explicitly model the geometric correspondences between line-endpoints.Experi-mental results demonstrate that the method outperforms OS-SIFT,SuperGlue and ALIKED methods in terms of the number of point matching,the reprojection root-mean-square error(RMSE)of point matching,spatial cover-age and other indicators,and that PLG-MAN enables more stable and more accurate optical-SAR registration un-der scenarios with high-noise and geometric distortions.