摘要
为了对煤矿井下变电所人员防护装备穿戴情况进行智能监测,以及避免监测视频受光照不均、粉尘干扰、遮挡等因素影响导致检测精确率降低的问题,提出了平滑幽灵卷积检测变换器(detection Transformer-smooth-L 1 ghost convolution,DETR-SGC)算法进行煤矿变电所安全防护装备检测。首先,在检测变换器(detection Transformer,DETR)算法的位置编码部分,引入幽灵组块通道缩放(ghost batchnormalization sigmoid gated linear unit-squeeze and excitation,GBS-SE)模块,增强算法空间维度特征提取能力;其次,在变换器模块中引入卷积块注意力模块(convolutional block attention module,CBAM),提高通道和空间维度特征提取能力,提升算法的检测精确率;最后,融合平滑L 1范数损失(smooth-L 1)和广义交并比(generalized intersection over union,GIoU)损失函数提升算法的回归精确率。实验表明,DETR-SGC算法的平均精确率、召回率、平均精确率均值分别达到了93.3%、87.9%、91.3%,比原始DETR算法分别提升了10.8%、4.3%、5.9%。因此,该算法能够有效解决煤矿变电所人员安全防护装备穿戴的检测问题。
In order to intelligently monitor the wearing of protective equipment of personnel in underground coal mine substations,and avoid the problem of uneven lighting,dust interference,obstruction,and other factors affecting the precision of video detection in monitoring videos,a detection Transformer smooth-L 1 ghost convolution(DETR-SGC)algorithm was proposed for the detection of safety protective equipment in coal mine substations.Firstly,in the position encoding part of the detection Transformer(DETR)algorithm,a ghost batchnormalization sigmoid gated linear unit-squeeze and excitation(GBS-SE)module was introduced to enhance the spatial feature extraction capability of the algorithm.Secondly,the convolutional block attention module(CBAM)was integrated into the Transformer module to improve the extraction of channel and spatial dimension features,thereby enhancing the detection precision of the algorithm.Finally,the fusion of smooth-L 1 norm loss and generalized intersection over union(GIoU)loss functions improved the regression accuracy of the algorithm.The experiments showed that the average precision,recall rate,and mean average precision of the DETR-SGC algorithm had reached 93.3%,87.9%,and 91.3%,respectively,which were 10.8%,4.3%,and 5.9%higher than the original DETR algorithm.Therefore,the algorithm can effectively solve the detection problem of safety protection equipment worn by personnel in coal mine substations.
作者
杨文轲
王向前
YANG Wenke;WANG Xiangqian(School of Computer Science and Engineering,Anhui University of Science and Technology,Huainan 232000,China;School of Economics and Management,Anhui University of Science and Technology,Huainan 232000,China)
出处
《湖北民族大学学报（自然科学版）》
CAS
2024年第4期528-532,581,共6页
Journal of Hubei Minzu University:Natural Science Edition
基金
国家自然科学基金项目(52374074)。
