摘要
针对基于深度学习的煤矸目标检测方法速度慢、参数量多、计算量大等问题,文章提出了一种基于CSR-YOLOv5s的煤矸目标检测算法。首先使用轻量化网络ShuffleNetv2作为主干网络,提升目标检测速度;其次将Neck区域的20×20特征图分支删除,降低了模型复杂度;最后SIOU损失函数替换CIOU损失函数,引入CBAM注意力机制使模型更加关注重要特征提高检测性能。实验结果表明:改进后的煤矸检测算法模型大小压缩了92.3%,参数量减少了94.3%,计算量降低了90.5%,帧数提高了34.2帧,可为煤矸的智能分选提供借鉴。
In order to solve the problems of slow speed,large number of parameters and large amount of computation,a new method based on CSR-YOLOv5s is proposed in this paper.Firstly,the lightweight ShuffleNetv2 is used as the backbone network to speed up target detection,and secondly,the 20×20 branch of the feature map in Neck region is deleted to reduce the model complexity.Finally,the SIOU loss function is replaced by the CIOU loss function,and the CBAM attention mechanism is introduced to make the model pay more attention to the important features and improve the detection performance.The experimental results show that the model size of the improved coal gangue detection algorithm is compressed by 92.3%,the parameter quantity is reduced by 94.3%,the computational amount is reduced by 90.5%,and the frame number is increased by 34.2 frames.
作者
孙志鹏
陶虹京
李熙尉
燕倩如
SUN Zhipeng;TAO Hongjing;LI Xiwei;YAN Qianru(Shanxi Datong University,College of Coal Engineering,Datong037003,China)
出处
《煤》
2023年第7期31-34,69,共5页
Coal
基金
山西省研究生教育创新项目(2021Y739,2022Y766)
山西大同大学研究生教育创新项目(21CX02,21CX37,22CX07)
山西大同大学2022年度校级揭榜招标项目(2021ZBZX3)
山西大同大学2021年度产学研专项研究项目(2021CXZ2)。
