基于机器视觉技术的井下人员安全检测系统  

Underground Personnel Safety Detection System Based on Machine Vision Technology

作　　者：崔德伟 CUI Dewei(Xiadian Coal Mine,Cilinshan Coal Industry Co.,Ltd.,Shanxi Lu'an Mining Group,Changzhi 046200,Shanxi,China)

机构地区：[1]山西潞安矿业集团慈林山煤业有限公司夏店煤矿,山西长治046200

出　　处：《能源与节能》2025年第3期61-63,66,共4页Energy and Energy Conservation

摘　　要：研究了基于机器视觉技术的井下人员安全检测系统。通过向数据集逐步引入不同比例(10%、20%、30%、40%)的高斯噪声,以模拟矿井内图像质量的逐渐下降。实验发现,10%的噪声对原图影响有限,而30%和40%的噪声则导致图像严重失真,无法有效反映井下真实环境。采用方差为0.01、强度为20%的高斯噪声对安全帽与口罩数据集进行加噪处理实验,结果显示,尽管这一处理方式导致YOLOv5模型在检测任务中的精确度、召回率及mAP@0.5(mean AveragePrecision atIoU 0.5,交并比阈值为0.5时的平均精度)略有下降,但仍保持了较高的识别精度,证明了其在矿井复杂环境下的鲁棒性和适应性。A safety detection system for underground personnel based on machine vision technology was studied.By gradually introducing Gaussian noise of different proportions(10%,20%,30%,40%)into the dataset,the gradual decline in image quality inside the mine can be simulated.The experiment found that 10%noise had limited impact on the original image,while 30%and 40%noise caused severe distortion of the image and cannot effectively reflect the real underground environment.The experiment of adding noise to the dataset of safety helmets and masks using Gaussian noise with variance of 0.01 and intensity of 20%showed that although this processing method resulted in higher accuracy,recall rate,and mAP@0.5(mean Average Precision at IoU 0.5)had slightly decreased of YOLOv5 model in detection tasks,but still maintained a high recognition accuracy,demonstrating its robustness and adaptability in complex mining environments.

关 键 词：机器视觉技术 井下人员 安全检测 图像识别 

分 类 号：TD79[矿业工程—矿井通风与安全]