Securing the Seas, Strengthening the State: Leveraging Multi-Scale Joint Attention Networks for Intelligent Ship Detection

With the advancement of economic globalization and international exchange, the demand for the automatic detection of ships is increasing daily. Surface ship detection faces numerous technical challenges. Firstly, the water surface environment is complex; changes in lighting can lead to image overexposure or low contrast, while adverse weather conditions such as haze, rain, and snow significantly reduce visibility. Secondly, the characteristics of ship targets themselves vary greatly and are prone to deformation due to different shooting angles. These issues collectively constitute the technical difficulties of intelligent ship detection. In this paper, inspired by the attention mechanism and multi-dimensional object detection design, we propose an improved model based on the YOLO framework. Additionally, considering that images of the same target at different distances can be vastly different, a ship target replication augmentation method based on segmentation is adopted. This method involves first segmenting the target, then scaling it according to various proportions and placing it against water surface background images to increase the diversity of target sizes. Without employing complex training techniques, our model demonstrates significantly improved performance compared to other detection models on the constructed ship dataset.