S-D HFFM: A Network Shallow-Deep Improvement Mechanism for Ship Detection in Low-Light Navigational Channel Scenarios

Accurately identifying ships in dim lighting is crucial for ensuring the safety of navigation channels. This paper proposes two improvement strategies for video ship detection. In the first scheme, for the shallow layer of the backbone network, we design a block to enhance the model’s attention to boundary information(SABlock). In the second scheme, for the deep layer of the backbone network, we design a block to preserve detailed features by increasing contrast (DPBlock). This enables the model to identify the ship’s category better. We present a shallow-deep hybrid feature fusion mechanism (S-D HFFM) to integrate shallow spatial and deep semantic information effectively. We validate these strategies through experiments conducted on YOLOv5s and YOLOv8s, demonstrating the feasibility of our approach. After incorporating S-D HFFM, YOLOv5s showed an increase of 2.1% in mAP50 and 11.8% in mAP50-95, while YOLOv8s showed an increase of 0.5% in mAP50 and 12.3% in mAP50-95. The experimental results clearly indicate that effective integration of shallow and deep features can significantly improve ship detection performance in low-light environments.