Target-Region-Aware Refocusing of Moving Ship in SAR Image

Abstract

Maritime ship targets in synthetic aperture radar (SAR) images are prone to azimuth defocusing, which causes edge blurring and scattering-structure degradation due to target motion and residual phase errors. The conventional refocusing methods are physically interpretable but often rely on explicit motion estimation, phase-error compensation, or iterative optimization, which limits their robustness in complex scenarios. This letter proposes a target-region-aware multiscale refocusing method for moving ship targets in SAR images. The pairs of focused-reference and defocused images are constructed from complex-valued SAR data by introducing azimuth phase perturbations, and a MIMO-UNet-based refocusing network is developed with adaptive cross-scale attention fusion (ACAF), interactive feature merging (IFM), and mask-supervised gate constraints. The proposed method guides the network to recover scattering structures in defocused ship regions while reducing unnecessary background modification. On the validation set, the proposed method achieves a peak signal-to-noise ratio (PSNR) of 32.21 dB, a structural similarity index measure (SSIM) of 0.757, an entropy reduction of 0.48, and a correlation-based edge preservation index (EPI) of 0.495. The comparisons with phase gradient autofocus (PGA), MIMO-UNet, and CV-CFUNet on four naturally defocused measured Umbra chips show more compact scattering responses in the displayed cases.