基于RepVGG-YOLOv4的焦罐提升机状态检测  

作　　者：李刚[1] 齐铭伟 张飞扬 吴凡 

机构地区：[1]北方工业大学电气与控制工程学院,北京100144

出　　处：《工业控制计算机》2022年第6期43-45,共3页Industrial Control Computer

摘　　要：提出一种基于视频流数据的提升机状态自动检测方法,将YOLOv4算法原有的CSPDarknet53框架替换成RepVGG框架,建立了基于RepVGG主干特征提取网络的YOLOv4模型,通过对提升机数据集的迭代训练,得到最终的检测模型,对设备状态进行实时分类,实现了设备状态的准确识别。测试结果表明,该算法的检测精度可达99.3%,检测速度26.28fps,满足提升机设备状态实时检测的需求,同时模型所占内存更小,更便于在生产现场部署,对硬件平台的要求有所降低。This paper proposes an automatic state detection method of hoist equipment based on video stream data,replaces the original CPSDarknet53 framework of YOLOv4 algorithm with RepVGG framework,established YOLOv4 model based on RepVGG backbone feature extraction network,through the iterative training of hoist equipment dataset,the final detection model is obtained,classifies the equipment state in real time,and realizes the accurate identification of equipment state.The test results show that the detection accuracy of the algorithm can reach 99.3%and the detection speed can reach 26.28fps,which can meet the needs of real-time detection of hoist equipment status.At the same time,the memory occupied by the model is smaller,which is more convenient for deployment in the production site,and the requirements for the hard-ware platform are reduced.

关 键 词：目标检测 深度学习 参数融合 模型优化 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TQ520.5[自动化与计算机技术—计算机科学与技术]