摘要
目的:提出一种基于YOLOv5改进的检测算法,解决密集动态目标检测精度低及易漏检的问题。方法:在YOLOv5的主干网络中使用QARepNeXt结构提高深度学习模型训练速度;引入S^(2)-MLPv2注意力机制改善遮挡情况下检测效果差的问题;将具有动态聚焦机制的边界回归损失函数Wise-IoU替代原有损失函数提高收敛速度。结果:以密集人群场景为例,验证改进算法相较现有算法具有更高准确率、更低漏检率,保证了原算法的实时性。结论:通过替换网络结构、引入注意力机制并更新损失函数,可有效提升算法准确率,降低漏检率。
Objective:To address the problems of low detection accuracy and easy missed detection of dense dynamic targets,this paper proposed an improved detection algorithm based on YOLOv5.Methods:In order to improve the training speed of the deep learning model,the QARepNeXt structure was used in the backbone network of YOLOv5.At the same time,in order to improve the problem of poor detection results under occlusion,the S^(2)-MLPv2 attention mechanism was introduced.In addition,in order to improve the convergence speed,the boundary regression loss function Wise-IoU with the dynamic focusing mechanism replaced the original loss function.Results:Taking the dense crowd as an example,it was verified that the improved algorithm had higher accuracy and lower miss detection rate than the existing algorithm,while ensuring the real-time performance of the original algorithm.Conclusion:By replacing the network structure,introducing the attention mechanism and updating the loss function,the accuracy of the algorithm could be effectively improved and the missed detection rate could be reduced.
作者
刘懿平
何欢
彭丰
朱家旺
江本赤
LIU Yiping;HE Huan;PENG Feng;ZHU Jiawang;JIANG Benchi(School of Mechanical Engineering,Anhui Polytechnic University,Wuhu 241000,China;School of Artificial Intelligence,Anhui Polytechnic University,Wuhu 241000,China;Anhui Minggu Laser Intelligent Equipment Technology Co.,Ltd.,Wuhu 241000,China;Wuhu Crane&Conveyor Co.,Ltd.,Wuhu 241000,China)
出处
《安徽科技学院学报》
2024年第2期79-86,共8页
Journal of Anhui Science and Technology University
基金
国家自然科学基金(52005003)
芜湖市科技项目(2022ly03,2022yf31,2023pt08)
安徽未来技术研究院企业合作项目(2023qyhz02)。
