基于M3CFC-YOLOv7-tiny的矿工乘坐架空乘人装置违章行为识别研究  

Research on identifying illegal behavior of miners riding overhead passenger devices based on M3CFC-YOLOv7-tiny

作　　者：卢纪峰 杨超宇[1] LU Jifeng;YANG Chaoyu(School of Artificial Intelligence,Anhui University of Science and Technology,Huainan 232001,China)

机构地区：[1]安徽理工大学人工智能学院,安徽淮南232001

出　　处：《煤矿安全》2024年第11期250-256,共7页Safety in Coal Mines

基　　金：国家自然科学基金资助项目(61873004)。

摘　　要：针对矿工乘坐架空乘人装置的违章行为自动识别任务存在煤矿井下轻量化设备部署、复杂环境下识别精度低、数据样本不平衡3个难题,提出了一种基于M3CFC-YOLOv7-tiny的轻量化矿工违章乘车行为智能识别算法。在向YOLOv7-tiny模型中引入MobileNetV3-Small网络用于边缘端部署;通过融合CBAM注意力机制提升对矿工行为特征的感知与表达能力;改进CIOU损失函数为Focal-CIOU平衡正负样本损失贡献;并在自建的矿工违章乘车行为数据集上进行消融实验和对比实验。结果表明:改进模型相比于原始模型参数量降低30.6%,浮点计算量降低46.9%,检测精度提升2.3%,实现模型轻量化和实时检测;对比9种目标检测模型,改进模型在多项指标上的综合性能最优且不存在漏检和错检。A lightweight miner illegal behavior intelligent detection algorithm based on M3CFC-YOLOv7-tiny is proposed to address the three challenges of deploying lightweight equipment in coal mines,low recognition precision in complex environment,and imbalanced data samples in the automatic recognition task of miners riding overhead passenger devices.MobileNetV3-Small network is introduced into the YOLOv7-tiny model for edge deployment.CBAM attention mechanism is added to enhance the perception and expression ability of miners’behavior characteristics.Improve the loss function to Focal-CIOU to balance the contribution of sample loss.Conduct ablation experiments and comparative experiments on the self-built dataset.The results show that compared with the original model,the parameters of the improved model are reduced by 30.6%,the floating point calculation is reduced by 46.9%,the detection accuracy is increased by 2.3%,and the model lightweight and real-time detection are realized;compared with 9 object detection models,the improved model has the best comprehensive performance in multiple indicators,and there are no missed detections or false detections.

关 键 词：煤矿安全 架空乘人装置 违章行为识别 M3CFC-YOLOv7-tiny MobileNetV3-Small CBAM注意力机制 Focal-CIOU损失函数 

分 类 号：TD52[矿业工程—矿山机电]