Edge-aware feature learning for fine-grained ship detection in high-resolution satellite images

Fine-grained ship detection in high-resolution satellite imagery is critical for maritime applications such as spill prevention, traffic control, and sea rescue. However, the task is complicated by challenges like complex backgrounds, intra-class similarity, and variable ship aspect ratios. In this paper, we propose BALF, a novel object detection network that integrates edge-guided feature extraction and local feature interaction to improve detection accuracy. BALF enhances attention to fine details by combining traditional edge detection with deep learning, capturing subtle object features and interactions within ROIs for improved recognition. Experimental results on the ShipRS and MCSD datasets demonstrate that BALF achieves superior performance, significantly outperforming state-of-the-art methods in fine-grained ship detection.