基于深度学习的下井人员安全状态快速检测算法研究

Study on the Fast Detection Algorithm for the Safety State of Downhole Workers Based on Deep Learning

在矿井生产中,作业环境复杂,安全生产风险较大,而传统的人工安全状态检测存在效率较低、漏检率较高等问题。因此,本文提出了基于卷积神经网络的下井人员安全状态快速检测方法。其间使用YOLOv3网络进行安全帽、工作服、工作鞋的多类别目标识别。试验表明,相比于传统人工检测方法,此算法具备更高的检测效率。在试验中,平均精度均值(mAP)达到了90.05%的高准确率,而且检测帧率达到28帧/s,具备了实时检测的能力。

In mine production,the operating environment is complex and the safety production risk is relatively high.However,the traditional manual safety state detection has problems such as low efficiency and high missed detection rate.Therefore,this paper proposed a fast detection method for the safety status of downhole personnel based on con⁃volutional neural network(CNN).In the meantime,YOLOv3 network was used for multi-category target recognition of helmets,work clothes and work shoes.Experiments show that compared with traditional manual detection methods,this algorithm has higher detection efficiency.In the experiment,the mean average precision(mAP)reached a high accuracy rate of 90.05%,and the detection frame rate reached 28 frames/s,which provided real-time detection capa⁃bilities.