Deep Learning-Based Ship Detection in Maritime Environments

With the development of shipping businesses and the increasing demand for marine monitoring, ship recognition and tracking technology have become increasingly important in the maritime field. A deep learning-based ship detection method is proposed to address the challenges of large object scale variations, background interference, and the tendency for small targets to be misdetected or undetected in satellite remote sensing images. Firstly, the DCNv2 module is integrated into the YOLOv7-tiny network to help the network adaptively identify regions of interest in images, thereby better capturing the detailed features and spatial relationships of the targets for feature optimization. Secondly, the DSConv module is introduced during the feature fusion stage to better capture the shape characteristics of the targets. Lastly, the Dyhead, which combines scale awareness, spatial awareness, and task awareness, is used as the detection head of the improved model to enhance feature expression capabilities and improve detection performance. Experimental results show that compared to YOLOv7-tiny, the improved network only increases computational load by 0.9%, but the accuracy and [email protected] are improved by 3.6% and 4.5%, reaching 83.3% and 82.6% respectively. This model achieves better detection performance while being suitable for deployment in environments with limited computing resources, demonstrating significant practical engineering value.