摘要
为了预防煤矿作业中因个人防护装备缺失导致的安全事故,提升工人安全帽和安全背心佩戴情况的智能检测能力。基于YOLOv8提出1种改进的目标检测算法SMT-YOLOv8s,该算法引入尺度感知调制模块(scale-aware modulation transformer,SMT)用于增强图像特征提取,设计跨通道增强通道注意力模块(cross-channel enhanced channel attention,C2ECA)以突出目标特征的表征能力,并提出增强型完全交并比(improved enhanced complete intersection over union,IE-CIoU),用于更精确地计算预测框与真实框之间的位置偏差。研究结果表明:提出的SMT-YOLOv8s算法相较于YOLOv8s模型在自建数据集上的mAP50和mAP50-95分别提高3.7百分点和2.7百分点。与其他算法相比,SMT-YOLOv8s兼具较高精度和计算效率。研究结果可为煤矿个人防护装备智能检测研究提供参考。
To prevent the safety accidents caused by the lack of personal protective equipment in coal mining operation,and improve the intelligent detection ability on the wearing situation of safety helmet and safety vest of the workers,an improved target detection algorithm,SMT-YOLOv8s based on YOLOv8 was proposed.The scale-aware modulation transformer(SMT)module was introduced to enhance the image feature extraction,and a cross-channel enhanced channel attention(C2ECA)module was designed to emphasize the representation ability of target feature.An improved enhanced complete intersection over union(IE-CIoU)function was also proposed to calculate the positional deviation between predicted and ground truth boxes more accurately.The results show that the proposed SMT-YOLOv8s algorithm improves mAP50 and mAP50-95 on self built datasets by 3.7 percentage points and 2.7 percentage points,respectively,compared to the YOLOv8s model,respectively.Compared with other algorithms,the SMT-YOLOv8s has both high accuracy and computational efficiency.The research results can provide reference for the research on intelligent detection of personal protective equipment in coal mines.
作者
程磊
张俊展
景国勋
王蒙
CHENG Lei;ZHANG Junzhan;JING Guoxun;WANG Meng(School of Safety Science and Engineering,Henan Polytechnic University,Jiaozuo Henan 454003,China;Henan Collaborative Innovation Center of Coal Work Safety and Clean-efficiency Utilization,Henan Polytechnic University,Jiaozuo Henan 454003,China)
出处
《中国安全生产科学技术》
北大核心
2025年第2期115-121,共7页
Journal of Safety Science and Technology
基金
国家自然科学基金项目(52374196)。
