An Anchor-Free Detection Algorithm for SAR Ship Targets with Deep Saliency Representation

Target detection in synthetic aperture radar (SAR) images has a wide range of applications military and civilian fields. However, for engineering involving edge deployment, it is difficult to find suitable balance accuracy speed anchor-based SAR image target algorithms. Thus, an anchor-free algorithm ship targets with deep saliency representation, called SRDet, proposed this paper improve performance against complex backgrounds. First, we design data enhancement method considering semantic relationships. Second, the state-of-the-art framework CenterNet2 used as benchmark, new feature-enhancing lightweight backbone, LWBackbone, designed reduce number model parameters while effectively extracting salient features targets. Additionally, mixed-domain attention mechanism, CNAM, suppress interference from land backgrounds highlight area. Finally, construct receptive-field-enhanced head module, RFEHead, multiscale perception head. Experimental results based on three large-scale datasets, SSDD, HRSID SAR-ship-dataset, show that our achieves better between exhibits excellent generalization performance.