Ship Detection via Multi-Scale Deformation Modeling and Fine Region Highlight-Based Loss Function

Ship detection in optical remote sensing images plays a vital role in numerous civil and military applications, encompassing maritime rescue, port management and sea area surveillance. However, the multi-scale and deformation characteristics of ships in remote sensing images, as well as complex scene interferences such as varying degrees of clouds, obvious shadows, and complex port facilities, pose challenges for ship detection performance. To address these problems, we propose a novel ship detection method by combining multi-scale deformation modeling and fine region highlight-based loss function. First, a visual saliency extraction network based on multiple receptive field and deformable convolution is proposed, which employs multiple receptive fields to mine the difference between the target and the background, and accurately extracts the complete features of the target through deformable convolution, thus improving the ability to distinguish the target from the complex background. Then, a customized loss function for the fine target region highlight is employed, which comprehensively considers the brightness, contrast and structural characteristics of ship targets, thus improving the classification performance in complex scenes with interferences. The experimental results on a high-quality ship dataset indicate that our method realizes state-of-the-art performance compared to eleven considered detection models.

Automated summary

This paper proposes a novel ship detection method that combines multi-scale deformation modeling and fine region highlight-based loss function. It extracts visual saliency using multiple receptive fields and deformable convolution to improve the ability to distinguish the target from complex backgrounds. A customized loss function for the fine target region highlight is used, considering the brightness, contrast, and structural characteristics of ship targets to improve classification performance in complex scenes.