基于改进YOLOv7的煤矿工人不安全行为识别方法

Method for identifying unsafe behaviors of coal mine workers based on improved YOLOv7

针对在煤矿变电所工人未穿戴绝缘手套和绝缘胶鞋容易引发安全事故,提出了基于YOLOv7的绝缘手套和绝缘胶鞋目标检测方法.使用深度可分离卷积在提高模型特征提取能力的同时降低网络计算量.通过引入SE注意力机制重构ELAN1和SPPCSPC模块,以增强网络特征提取能力,加强对小目标检测的精度.采用WIoU损失函数提高网络回归精度.实验结果表明,改进的YOLOv7算法比原始算法准确率提高3.8%,mAP提高12%.能够对未穿戴绝缘手套和绝缘胶鞋的不安全行为进行高效实时的检测.

A YOLOv7-based target detection method for insulated gloves and insulated rubber shoes was proposed to address the risk of safety accidents caused by workers in coal mine substations not wearing insulated gloves and insulated rubber shoes.Using depthwise separable convolution improved the model′s feature extraction ability while reducing network computation.By introducing the SE attention mechanism,the ELAN1 and SPPCSPC modules were reconstructed to enhance the network's feature extraction capability and improve the accuracy of small object detection.The WIoU loss function was employed to enhance the accuracy of network regression.The experimental results indicated that the improved YOLOv7 algorithm increased accuracy by 3.8%and mAP by 12%compared to the original algorithm,allowing for efficient and real-time detection of unsafe behavior related to not wearing insulated gloves and insulated rubber shoes.