基于YOLOv5s的外脚手架隐患智能动态检测

Intelligent dynamic detection of hidden dangers of exterior scaffolding based on YOLOv5s

针对脚手架工程隐患人工巡检效率低、实时排查难的问题,提出一种基于深度学习的隐患实时检测方法。利用添加噪声、随机裁剪等数据增强方式扩充数据集,提高模型在复杂环境下的鲁棒性;基于YOLOv5s目标检测算法建立脚手架工程隐患图像识别模型并进行训练测试,与YOLOv4、Faster R-CNN进行对比,验证模型的有效性。结果表明,在脚手架工程隐患检测任务上,YOLOv5s模型的均值平均精度达到92.23%,较YOLOv4提升8.11百分点;检测速度达到97.01帧/s,较Faster R-CNN提升5倍。轻量的YOLOv5s网络模型适合部署于嵌入式智能监控中,实时采集现场数据并进行隐患分类识别,有效缩短隐患发现时间,研究结果可为脚手架工程监控预警平台提供研究基础。

Aiming at the problems of low efficiency of manual inspection and real-time troubleshooting of hidden dangers in scaffolding projects,a real-time detection method of hidden dangers based on deep learning is proposed.Through construction site investigation and accident case statistics,the types of hidden dangers are determined and images are collected.The data set is expanded by data enhancement methods such as adding noise and random crop-ping to improve the robustness of the model in complex environments.Based on YOLOv5s target detection algorithm,the hidden dangers of scaffolding projects are established.The image recognition model is trained and tested,and com-pared with YOLOv4 and Faster R-CNN to verify the effectiveness of the model.The results show that,in the task of scaffolding hidden danger detection,the average accuracy of YOLOv5s reaches 92.23%,which is 8.11 percentage pionts higher than that of YOLOv4,while the detection speed reaches 97.01 frames/s,which is five times faster than Faster R-CNN.The lightweight YOLOv5s network model is suitable for deployed on the embedded side.Real time collection of on-site data and identification of hidden dangers can effectively shorten the discovery time of hidden dangers.The research results can provide a research basis for the scaffolding engineering monitoring and early warning platform.