Warship Object Detection in Remote Sensing Images with Improved YOLOv5

Aiming at the task of detecting ships on the sea surface in remote sensing images, there are a lot of disturbances such as variable object sizes, cloud occlusion, complex image backgrounds, and different ship orientations, etc. In this paper, we propose a ship rotating object detection network based on the improved YOLOv5, which is constructed through the strategies of introducing Swin Transformer as a feature extraction network to enhance the feature extraction performance of the network, introducing a rotating detection head to realize the detection of the rotation angle, and modifying the network loss function to accelerate the convergence of the network. The network finally achieves a 71.7% map on ShipRSImageNet validation set, which is an improvement of 2.4% compared with the original network model. The network proposed in this paper solves the problem that the YOLOv5 algorithm is unable to detect rotating objects, and the network based on the self-attention mechanism is used to further enhance the ability to detect small objects. Finally, a ship object detector that can be used in real remote sensing satellite images is obtained.