基于深度学习的冶炼工人安全着装监测系统

为解决铜冶炼作业过程中工人安全着装穿戴不规范导致的安全生产问题，设计一种基于深度学习的冶炼工人安全着装监测系统。首先，在充分对比实验的基础上，选择准确性高且满足实时性要求的YOLOv5l作为工人安全着装目标检测模型，实验结果表明mAP@0.5、F1-Score分别为84.3%、90.8%，平均单帧检测时间为13 ms；其次，设计基于时空关系分析的工人安全着装推理算法，依据监测结果进行时空关系分析，实现对工人安全着装穿戴不规范现象的智能分析报警；最后，将YOLOv5l部署到DeepStream框架中结合推理算法构建安全着装监测系统，实现对违规现象声光报警、录制违规视频、上传移动端显示报警详情。经生产现场验证，系统误检率为4.8%、漏检率为2.7%，可有效提高铜冶炼安全监管水平。

A smelter safety dress monitoring system based on deep learning

In order to solve the safety production problem caused by irregular wearing of workers'' safe dress in the process of copper smelting, a safety dress monitoring system for smelting workers based on deep learning is designed. Firstly, on the basis of full comparison experiments, YOLOv5l with high accuracy and real-time requirements is selected as the target detection model for workers'' safe dress, and the experimental results show that the mAP@0.5 and F1-Score are 84.3% and 90.8%, respectively, and the average single-frame detection time is 13 ms. Secondly, a reasoning algorithm for workers'' safe dress based on spatiotemporal relationship analysis is designed, and spatiotemporal relationship analysis is carried out according to the monitoring results, so as to realize intelligent analysis and alarm for irregular behaviors of workers'' safe dressing. Finally, YOLOv5l is deployed into the DeepStream framework and combined with the inference algorithm to build a safe dress monitoring system to realize the sound and light alarm for violations, record illegal videos, and upload mobile terminals to display alarm details. After verification by the production site, the system false detection rate is 4.8% and the missed detection rate is 2.7%, which can effectively improve the safety supervision level of copper smelting.