Rotated Object Detection of Ship in UAV Aerial Images Based on Improved YOLOv7

Aiming at the difficulty of ship detection in UA V aerial images and the problem of ignoring its direction, a rotated object detection algorithm based on improved YOLOv7 is proposed. By adding the angle prediction dimension and introducing the Gray Coded Label representation method to discretize the angle, the accuracy of the angle prediction is improved. The backbone network is replaced by Swin Transformer, and its attention mechanism is used to improve the feature extraction ability of the model. Then the K-means++ algorithm is used to cluster the anchor boxes, so that they are more suitable for the overall distribution to improve the detection performance of the model. The experimental results show that the improved rotated object detection method improves the detection accuracy by 9.12 percentage points compared with the original YOLOv7 model, and it can provide effective technical support for the task of ship detection.