AIS-FCANet: Long-Term AIS Data Assisted Frequency-Spatial Contextual Awareness Network for Salient Ship Detection in SAR Imagery
Weibao Xue, Jiaqiu Ai, Yanan Zhu, Jiayi Chen, Shuo Zhuang
Abstract
Salient ship detection in Synthetic Aperture Radar (SAR) imagery plays a crucial role in maritime surveillance. However, traditional single-modal SAR methods are constrained by limited ship-relevant information and weak feature extraction robustness, which hinders detection performance. To address this issue, this correspondence proposes a long-term Automatic Identification System (AIS) data assisted Frequency-spatial Contextual Awareness Network, named AIS-FCANet, for Salient Ship Detection in SAR Imagery. This is the first attempt to incorporate ship distribution density information derived from long-term AIS data to enhance the feature representation of ships in SAR images. We incorporate the Segment Anything Model 2.1 (SAM2.1) as the backbone feature extractor, leveraging its capability to capture both fine-grained details and global context. To further optimize the input features, an Adapter module is designed for SAM2.1, which utilizes a learnable prompting mechanism to dynamically refine feature representations, thereby significantly improving the robustness of ship feature extraction. Furthermore, to fully exploit the complementary characteristics of SAR and AIS data, a frequency-spatial domain context-aware fusion module (FSCM) is developed. This module enhances both frequency and spatial domain attention mechanisms to enhance multi-level feature fusion and facilitate cross-modal interaction. Experimental results show that AIS-FCANet outperforms existing methods, achieving significant improvements in both detection accuracy and robustness.