改进YOLOv7的轧钢车间安全帽佩戴检测算法

佩戴安全帽能够保护生产工作者头部免受坠物撞击带来的伤害。轧钢车间存在空间跨度大、作业设备繁多、环境杂乱、昼夜光照差别大、炫光、监控目标尺度变化范围大等问题，增加了安全帽佩戴检测难度。针对上述问题，设计了基于改进YOLOv7模型的轧钢车间安全帽佩戴检测方案；算法基于NWD方法改进损失函数以提高目标检测精度，并在SPPCSPC模块上增加了BiFormer模块，使模型对小目标具有更好的检测精度，同时不会增加运算负担，优于其他注意力机制。在自建安全帽数据集上对改进的YOLOv7模型进行训练，实验表明，改进的YOLOv7模型平均精度均值为99.3%，检测速度达82FPS，与其他主流算法、改进算法对比，改进YOLOv7的mAP指标最高，大大超过了其他模型的指标，同时检测速度基本与改进模型前相差不大，并没有因为精度提高而明显降低检测速度，有较好效果。

Improved Helmet Wearing Detection Algorithm in Steel Rolling Workshop for YOLOv7

Wearing helmets can protect the head of production workers from injuries caused by falling objects. There are problems such as large span of space, many operating equipment, cluttered environment, big difference of day and night light, dazzling light, and large range of scale change of monitoring target in rolling mill, which increase the difficulty of helmet wearing detection. In response to the above problems, a helmet wearing detection scheme based on the improved YOLOv7 model is designed for the steel rolling shop. The algorithm improves the loss function based on the NWD method to improve the target detection accuracy, and adds the BiFormer on the SPPCSPC module to make the model have better detection accuracy for small targets without increasing the computational burden, which is better than other attention mechanisms. The improved YOLOv7 model is trained on the self-constructed helmet dataset, and the experiments show that the improved YOLOv7 model has a mean average accuracy of 99.3%, and the detection speed reaches 82 FPS. comparing with the other mainstream algorithms and the improved algorithms, the improved YOLOv7 has the highest mAP index, which is much more than the index of other models. At the same time, the detection speed is basically not much different from that before the improvement of the model, and does not significantly reduce the detection speed because of the improvement of accuracy, which has a better effect.