煤矿工作面喷雾除尘场景下的安全帽识别算法  

Safety helmet recognition algorithm in spray dust removal scenario of coal mine working face

作　　者：张婧 冯莹莹 李洪安[1] 杜思哲 莫金明 ZHANG Jing;FENG Yingying;LI Hongan;DU Sizhe;MO Jinming(College of Computer Science and Technology,Xi’an University of Science and Technology,Xi’an 710600,China;State Key Laboratory of Coal Mine Disaster Prevention and Control,Chongqing 400037,China;CCTEG Chongqing Research Institute,Chongqing 400037,China)

机构地区：[1]西安科技大学计算机科学与技术学院,陕西西安710600 [2]煤矿灾害防控全国重点实验室,重庆400037 [3]中煤科工集团重庆研究院有限公司,重庆400037

出　　处：《矿业安全与环保》2024年第4期9-16,共8页Mining Safety & Environmental Protection

基　　金：国家自然科学基金项目(61902311);陕西省自然科学基金项目(2024JC-YBQN-0065);重庆市自然科学基金面上项目(cstc2021jcyj-msxmX1153)。

摘　　要：首先针对煤矿工作面喷雾除尘场景下监控系统采集到的图片模糊、清晰度低的问题,提出了一种基于DeDi-Transformer(Density Difference-Transformer)的煤矿工作面图像去雾算法,该算法利用密度差实现密度感知,对采集的工作面监控图像进行增强,提高图像中人员安全帽的清晰度;其次针对煤矿工作面监控系统很难快速准确识别出煤矿工人是否佩戴安全帽的问题,提出了一种基于SAC-YOLOv9(Supervised Atrous Convolution-YOLOv9)的安全帽识别算法,该算法在YOLOv9主干提取网络中加入监督空洞卷积,获取不同尺度的感受野,加快特征提取,提高安全帽识别的精度。实验结果表明,DeDi-Transformer算法在Braize-Haze数据集上的PSNR为19.85 dB,比DeHamer算法提升了2.49 dB;SSIM为0.7179,比DeHamer算法提高了0.0434。SAC-YOLOv9算法在Dehaze-Helmet数据集上的mAP为95.7%,与YOLOv9算法相比提升了2.3%。Firstly,in order to solve the problem of blurry and low definition images collected by the monitoring system in the spray dust removal scenario of coal mine working face,a image dehazing algorithm of coal mine working face based on DeDi-Transformer(Density Difference-Transformer)was proposed.The algorithm used density contrast to realize density perception,enhanced the collected working face monitoring image,and improved the clarity of the personnel’s safety helmet in the image.Secondly,in view of the problem that it was difficult for the coal mine working face monitoring system to quickly and accurately identify whether coal miners were wearing safety helmets,a safety helmet recognition algorithm based on SAC-YOLOv9(Supervised Atrous Convolution-YOLOv9)was proposed.The algorithm added the supervised atrous convolution into the YOLOv9 backbone extraction network to obtain receptive fields of different scales,speeded up feature extraction,and improved the accuracy of safety helmet recognition.Experimental results show that the PSNR of the DeDi-Transformer algorithm on the Braize-Haze dataset is 19.85 dB,which is 2.49 dB higher than the DeHamer algorithm.The SSIM is 0.7179,which is 0.0434 higher than the DeHamer algorithm.The mAP of the SAC-YOLOv9 algorithm on the Dehaze-Helmet dataset is 95.7%,which is 2.3%higher than the YOLOv9 algorithm.

关 键 词：煤矿工作面 喷雾除尘 图像去雾 安全帽识别 密度差 监督空洞卷积 

分 类 号：TD76[矿业工程—矿井通风与安全] TD714