SAR Ship Detection with Deep Land Detection Networks and Land Masking

We propose a new Synthetic Aperture Radar (SAR) ship detection framework using deep learning that makes SAR land mask to improve SAR ship detection performance. To overcome the disadvantage of a small number of SAR dataset, the land detection networks take a large amount of small image patches as input and effectively learn land feature for detecting land mask. We then eliminate land using the detected land mask and apply artificial noise to preserve image distribution. With these schemes, SAR ship detector can concentrate on ship feature rather than land feature. Experimental results demonstrated that the proposed method outperforms baseline.