基于YOLOv5的水下生物检测算法研究  

Research on Underwater Biological Detection Algorithm Based on YOLOv5

作　　者：赵学军[1] 曹天宇 李琳[1] 

机构地区：[1]中国矿业大学(北京)人工智能学院,北京100083

出　　处：《工业控制计算机》2025年第2期95-96,99,共3页Industrial Control Computer

摘　　要：针对水下生物目标检测的挑战,提出了一种基于YOLOv5改进的高效模型。通过引入BiFormer注意力机制、创新性的Focal-ShapeIoU损失函数,模型在水下生物特征提取、类别不平衡处理、细节恢复和计算效率方面均得到显著提升。实验结果显示,相较于原始YOLOv5,改进后的YOLOv5在水下生物目标检测任务上的准确率提高了4.1%,验证了改进策略的有效性。This paper proposes an efficient model based on YOLOv5 improvement for underwater biological target detection.By introducing the BiFormer attention mechanism and innovative Focal-ShapeIoU loss function,the model has achieved significant improvements in underwater biological feature extraction,class imbalance processing,detail recovery,and computational efficiency.Experimental results show that the accuracy of the improved YOLOv5 model on underwater biological target detection tasks has increased by 4.1%compared to the original YOLOv5,validating the effectiveness of the improved strategy.

关 键 词：目标检测 BiFormer 注意力机制 损失函数 YOLOv5 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程]