文献类型：会议论文

英文题名：Using Improved YOLOv5 for Underwater Target Detection

作者：Peng, Xiaohong[1]; Li, Yifan[1]; Chen, Xiaohan[1]

机构：[1] Guangdong Ocean University, College of Mathematics and Computer Science, Zhanjiang, China

会议论文集：2023 IEEE 6th International Conference on Pattern Recognition and Artificial Intelligence, PRAI 2023

会议日期：August 18, 2023 - August 20, 2023

会议地点：Haikou, China

语种：英文

外文关键词：Image enhancement - Marine biology - Oceanography - Semantics - Signal detection

外文摘要：In recent years, as the exploitation and utilization of marine resources have increased, underwater target detection has become increasingly important. Improving related technologies holds significance for resource utilization and extraction. Although existing target detection algorithms have achieved good results on land, they face challenges such as low accuracy and slow speed in underwater environments. To address this issue, this paper proposes an improved underwater target detection algorithm called YOLOv5-GCB. Firstly, the algorithm adopts the GhostNet network to reduce computational complexity. Secondly, an attention mechanism is introduced after the last C3 module in the Backbone, allowing the model to focus on extracting target features and highlighting key characteristics. Furthermore, a weighted bidirectional feature pyramid structure is employed in the Neck section, utilizing a bidirectional weighting approach that integrates position information from lower-level feature maps with semantic information from higher-level feature maps. Finally, we applied the improved model to our self-collected underwater image dataset for experimentation. The experimental results show that the YOLOv5-GCB model achieves an average precision (mAP) of 82.9% and a recall rate of 76.5% on the test dataset. Compared to the baseline model (YOLOv5), the performance of the improved model has improved by 5.5% and 5.8% respectively. ? 2023 IEEE.