摘要
为了确保电力作业人员按要求佩戴个人安全护具,保障其人身安全,提出了一种基于YOLOv3的多特征融合算法,检测变电站工作人员是否规范佩戴安全护具。在特征提取网络部分引入压缩与激励网络SENet模块,提升检测能力。在YOLOv3输出的3个特征图基础上,增加输出一个小尺度的特征图,为后续网络特征融合阶段提供更多小目标特征信息。由于数据集中安全帽目标居多而其他目标较少,引入迁移学习策略平衡目标样本数量,进一步提高安全护具的检测精度。为验证算法的有效性,构建一组关于安全护具佩戴情况的数据集。实验结果表明,该方法在保持检测速度相当的情况下,将平均精度提高了6.13%,每秒帧数可达47.4,能够较好地应用于安全护具佩戴检测任务。
To ensure that electric power operators wear personal safety helmet as required and ensure their safety,a multi-feature fusion algorithm based on YOLOv3 is proposed to detect whether substation workers wear safety protective equipment in a standard way.In the part of feature extraction network,the compression and excitation network SENet module is introduced to improve the detection capability.Based on the three feature maps output by YOLOv3,a small-scale feature map is added to provide more small target feature information for the subsequent network feature fusion stage.Since there are more helmet targets and fewer other targets in the data set,the migration learning strategy is introduced to balance the number of target samples and further improve the detection accuracy of safety protectors.To verify the effectiveness of the algorithm,a set of data sets about the wearing of safety protective equipment is constructed.The experimental results show that this method can improve the average accuracy by 6.13%and the number of frames per second can reach 47.4 while maintaining the same detection speed.It can be applied to the detection task of safety helmet wearing.
作者
陈菡菡
CHEN Hanhan(Department of Intelligent Manufacturing Engineering,Meizhouwan Vocational Technoly College,Putian,Fujian 351100,China)
出处
《闽江学院学报》
2023年第5期63-72,共10页
Journal of Minjiang University
