The Turbo-YOLOv8 for Underwater Target Detection in Complex Background

Underwater target detection is an important technology in marine research and application. However, due to the complexity of underwater environment and the diversity of target sizes, the existing detection models face many challenges in small target detection and model lightweight. In this paper, an underwater target detection algorithm, Turbo- YOLOv8, is proposed to improve the performance of underwater target detection. Turbo- YOLOv8 has made three improvements on the basis of YOLOv8. Firstly, an upsample path and a branch network specially used for small target detection are added to the 80×80 feature map, so that the detection ability of the model for small targets is significantly improved. Secondly, the L WC2F module is proposed, specifically, the lightweight module shuffle block is used to replace the original Bottleneck in C2F, so as to reduce the model parameters. Finally, the dilated convolution is used to replace the depthwise convolution in shuffle block to increase the receptive field. The experimental results show that the performance of Turbo-YOLOv8 on Trash-ICRA19 Dataset is better than that of mainstream detection models, in which the precision reaches 92.53% and the mAPO.5 reaches 95.36%, and the model parameters are reduced by 6.71 % compared with YOLOv8. This study not only provides an efficient and lightweight solution for underwater target detection, but also provides an important reference for small target detection in complex background.