AUTHOR=Zhang Jialin , Jin Jiucai , Ma Yi , Ren Peng 

TITLE=Lightweight object detection algorithm based on YOLOv5 for unmanned surface vehicles

JOURNAL=Frontiers in Marine Science

VOLUME=Volume 9 - 2022

YEAR=2023

URL=https://www.frontiersin.org/journals/marine-science/articles/10.3389/fmars.2022.1058401

DOI=10.3389/fmars.2022.1058401

ISSN=2296-7745

ABSTRACT=Visual detection technology is a major means for Unmanned Surface Vehicle (USV) to perceive the surrounding environment. The spatial position and category of the object can be determined by visual detection, which can provide important environmental information for path planning and preventing collisions of the USV. In the close-in reconnaissance mission, it is necessary for USV to navigate fast in the complex maritime environment, so the object detection algorithm needs to have a fast detection speed and high accuracy. In this paper, a You Only Look Once (YOLOv5) lightweight object detection algorithm using ghost module and transformer is proposed for USV. Firstly, in the backbone network, the original convolution (Conv) operation in YOLOv5 is upgraded by Conv stacking with Depth-wise Conv in the Ghost module. Secondly, to exalting feature extraction without deepening network depth, we take transformer integrated into the end of the backbone network and the FPN structure in the YOLOv5, which can improve the ability of the feature expression. Lastly, the proposed algorithm and six algorithms of deep learning have been used to test on ship datasets. The results show that the average accuracy of the proposed algorithm is higher than the other six algorithms. Especially, compared with the original YOLOv5 model, the model size of the proposed algorithm is reduced to 12.24M, the frames per second (FPS) reached 138, the detection accuracy has improved by 1.3%, and the mAP (0.5) from 95.3% reaches to 96.6%. In the experiment's verification, the proposed algorithm is tested on the ship video collected by the "JiuHang 750" USV under different sea environments. The test results show that the proposed algorithm has a significantly improved detection accuracy compared with other lightweight detection algorithms.