Synthetic aperture radar ship detection in dense area based on pseudo-echo suppression

To solve the missed detection problem of synthetic aperture radar ships in dense areas due to pseudo-echo, a suppression dense pseudo-echo algorithm is proposed, which includes a multidimensional echo controller with depth dense selection mechanism, which effectively enhances detection accuracy. This algorithm improves the boundary characteristics of radar signals by adopting ultrafast attenuation of the sidelobe in the boundary layer, which effectively suppresses the interference of the sidelobe signal on the original image. In addition, the target edges are highlighted to mitigate the pseudo-echo phenomenon by segmenting the two-dimensional image with fine-grained stripes so that each stripe undergoes the phase de-entanglement process. The latter adaptively selects matched feature sensing fields and detection sizes to locate ships in dense areas through weight adjustment of spatial features, effectively avoiding missed detection. Finally, we propose the spatial scale perception loss function to adapt the ship size by quantitatively matching the bound box to the target size to further improve the detection accuracy and enhance the generalization ability in dense areas. The experimental results on the HRSID, SAR-Ship-Dataset, SSDD, and LS-SSDD-v1.0 datasets show that the average accuracy (mAP0.5) index of this method has reached 93.4%, 94.2%, 98.5%, and 58.1%, respectively, which is better than the existing representative method.