基于深度学习的复杂作业场景下安全帽识别研究

为有效预防由于个人防护缺失所造成的事故,着力探究复杂作业情况下施工人员安全帽佩戴情况的智能化识别.提出在Faster R-CNN目标检测算法的基础上,针对小目标的安全帽识别问题通过增加锚点提升检测能力,为解决数据集中类别不平衡问题采用Focal loss替代原本的损失函数,为解决安全帽预测区域不匹配问题,引入ROI A...

Research on recognition of safety helmets under complex operation scenes based on deep learning

In order to effectively prevent the accidents caused by the lack of personal protection,the intelligent recognition of helmet wearing conditions for the construction workers under complex operation scenes was explored with focus.It was proposed to improve the detection ability by adding the anchor points to the helmet recognition problem of small targets on the basis of the Faster R-CNN target detection algorithm.To solve the problem of category imbalance in the data set,the Focal loss was used to replace the original loss function.To solve the problem of mismatch in the predicted area of helmet,the ROI Align was introduced to replace the error generated by the secondary quantization in the ROI Pooling operation,thereby improving the accuracy of the detection model.Finally,the network performance evaluation was performed on the basis of the helmet data set under the constructed complex scenes.The results showed that based on the improved Faster R-CNN network framework,the mAP increased by 15%,which provides an effective and accurate identification method for the intelligent management and control on the wearing problem of personal protective equipment at the construction sites.