摘要
球类运动是传统体育竞技中受关注度最高的一类运动,球类的目标检测可以用于提高体育比赛的分析、监控系统的安全性以及虚拟现实体验的真实感。YOLOv5作为优秀的单阶段检测算法,因其平台移植方便与检测步骤简易,是计算机视觉领域近年来使用频率最高的目标检测算法之一。但是YOLOv5模型参数量较大,为了减少参数量以便更快地移植到其他平台上,文章提出一种轻量化的改进YOLOv5算法,该算法以YOLOv5s为基础模型,通过将主干网络替换为改进的MobileNetv3、在颈部引入CBAM注意力机制并改进C3模块等方法,减少计算量并提升精度。对训练完成后的改进模型进行验证,实验结果表明改进后的检测算法参数量大致下降了65%,平均精度提升了0.5%,满足乒乓球实际应用场景的精度要求和实时性。
Ball games are the most popular sports in traditional sports competitions,and the target detection of balls can be used to improve the analysis of sports games,the security of surveillance systems,and the realism of Virtual Reality experiences.YOLOv5,as an excellent single-stage detection algorithm,is one of the most frequently used target detection algorithms in the field of computer vision in recent years due to its easy platform portability and simple detection steps.However,the YOLOv5 model has a large number of parameters.In order to reduce the number of parameters so that it can be ported to other platforms faster,this paper proposes a lightweight and improved YOLOv5 algorithm,which takes YOLOv5s as the base model,and reduces the amount of computation and improves the accuracy by the methods of replacing the backbone network with the improved MobileNetv3,introducing the CBAM Attention Mechanism in the neck,and improving the C3 module.The improved model is verified after the training,and the experimental results show that the number of parameters of the improved detection algorithm roughly decreases by 65%,and the average accuracy improves by 0.5%,which meets the accuracy requirements and real-time performance of practical application scenarios for table tennis.
作者
施博凯
张昕
邱天
张志鹏
SHI Bokai;ZHANG Xin;QIU Tian;ZHANG Zhipeng(Joint Laboratory of Digital Optical Chips of Institute of Semiconductors,Chinese Academy of Sciences,Wuyi University,Jiangmen 529020,China)
出处
《现代信息科技》
2024年第15期28-35,共8页
Modern Information Technology
基金
2019年广东省拨款高校建设“冲补强”专项基金
五邑大学高级人才科研启动基金2019(5041700171)
2021年江门市创新实践博士后课题研究资助项目(JMBSH2021B04)
广东省重点领域研发计划(2020B0101030002)
2020五邑大学大学生创新创业计划(202011349186)。
