Research on integrated visual detection and dynamic orientation calculation for ship situation awareness

The main problem faced in intelligent ship navigation is the insufficient target recognition under complex sea surface conditions, making it difficult to accurately determine the target's position, which leads to difficulties in collision avoidance decision-making. This paper proposes a method based on YOLOv9 for the azimuth angle recognition of marine targets in ship navigation. First, YOLOv9 is adopted as the baseline framework, and a maritime target perception model is developed by combining a triple attention mechanism with a lightweight architecture. A triple attention module is introduced in the detection layer to enhance the representation of salient features and suppress sea clutter interference, while group convolutions are employed in the feature extraction stage to reduce redundant computations and model parameters. Experimental results show that the proposed perception model achieves an accuracy of 93.1% with a 13.4% reduction in parameters compared to the baseline model. Ablation and comparison experiments further validate its effectiveness and generalization capability. Building on this, a bearing analysis model is constructed by combining the geometric center of the detection bounding box with real-time bow direction data, enabling the automatic calculation of the target’s true bearing (TB) and quadrant bearing (Q). The proposed method generates an interpretable situational awareness representation that aligns with the navigation judgment habits of the crew. Without relying on expensive ranging equipment, it provides ships with low-cost, lightweight, and interpretable visual perception and risk warning capabilities, supporting navigation tasks and collision avoidance decisions, and laying the foundation for future integration with AIS, radar, and intelligent collision avoidance systems.