Abstract
Synthetic aperture radar (SAR) ship detection faces challenges such as scale diversity, weak target features, and strong background interference. To address these issues, we propose a cross-scale context fusion network (CSCF-Net), integrating a multiscale feature fusion (MSFF) module for cross-scale contextual aggregation and a multitask interactive detection head (MTIDH) for task-specific optimization through dynamic deformable convolution. Extensive experiments on three SAR datasets demonstrate superior performance; CSCF-Net achieves 98.5% mean average precision (mAP) on SSDD, 91.5% on HRSID, and 98.1% on SAR-Ship-Dataset, with 1.0%, 2.2%, and 0.9% improvements over baseline, respectively, outperforming state-of-the-art methods and validating the effectiveness of our proposed method.
Funding Information
  • National Key Research and Development Program of China (2023YFC2809400)