An Improved YOLOv5 Approach for Multi-object Detection of Unmanned Surface Vehicles

This paper aims to address the challenge of multi-object detection for unmanned surface vehicles (USVs) due to the partially occluded and blurred nature of water surface targets in captured images. To this end, we propose an enhanced version of the YOLOv5 network to improve detection accuracy and efficiency. Firstly, the Ghost module is introduced to reduce the computational complexity when obtaining feature maps, then the SE module is introduced to enhance the extraction of feature informations, and finally the loss function of YOLOv5 is optimized by CIOU to improve the training effect of the model. Results of our experiments indicate that the proposed YOLOv5s-GS model is lightweight and has significantly improved detection accuracy and efficiency over the original YOLOv5s for multi-object detection tasks.