基于YOLOv7-CA-BiFPN的路面缺陷检测  

Road Surface Pothole Detection Based on YOLOv7-CA-BiFPN

作　　者：高敏 李元 GAO Min;LI Yuan(College of Information Engineering,Shenyang University of Chemical Technology,Shenyang 110142,China)

机构地区：[1]沈阳化工大学信息工程学院,沈阳110142

出　　处：《计算机测量与控制》2024年第9期9-14,43,共7页Computer Measurement &Control

摘　　要：路面坑洼是主要道路缺陷,会损坏车辆,影响驾驶员的安全驾驶,严重时还会导致交通事故,针对这个问题,提出了改进YOLOv7的道路坑洼检测算法;使用Mosaic+Mixup进行内置数据增强,扩充小样本数据集,增强模型泛化能力;引入CA注意力机制,将横纵位置信息编码,保证计算量的同时又能关注大范围位置信息;采用BIFPN双向特征金字塔网络,通过特征融合多尺度语义特征提高检测效率;将损失函数SIoU替换CIoU,有效解决回归中的样本不平衡问题;实验结果表明,改进之后的算法在坑洼数据集的平精度均值和精确率达到了89.42%和90.12%,相比于原本的YOLOv7版本提高了6.18%和1.96%,更准确更快速地应用于道路维修。Road potholes are the main road defects of roads,which can damage vehicles,affect driver safety,and even lead to traffic accidents in severe cases.To address this issue,an improved YOLOv7 road pothole detection algorithm is proposed.Mosaic+Mixup is used to to carry out the built-in data augmentation,expand the small sample datasets,and enhance the model generalization ability;By introducing an coordinate attention(CA)attention mechanism,the horizontal and vertical position information is encoded to ensure computational complexity while paying attention to the large-scale position information;BIFPN bidirectional feature pyramid network is adopted to improve detection efficiency through the feature fusion of multi-scale semantic features;By replacing the loss function SIoU with the CIoU,the sample imbalance in regression is effectively solved.Experimental results show that the improved algorithm achieves the mean value and accuracy of 89.42%and 90.12%in pit datasets,which are 6.18%and 1.96%higher than that of the original YOLOv7 version.It can be more accurately and quickly applied to road maintenance.

关 键 词：坑洼检测 YOLOv7 注意力机制 数据增强 BiFPN 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]