基于机器视觉的路面病害检测技术对比研究  

Comparative Study on Pavement Disease Detection Technology Based on Machine Vision

作　　者：王华 WANG Hua(Jiangxi Jiujiang Fuyin Expressway Bridge Co.,LTD.,Jiu Jiang,Jiangxi 332000,China)

机构地区：[1]江西九江长江公路大桥有限公司,江西九江332000

出　　处：《黑龙江交通科技》2022年第12期64-66,共3页Communications Science and Technology Heilongjiang

摘　　要：道路在交通荷载、自然老化及环境影响的综合作用下,服役过程中其表面将产生种类繁杂、严重程度不一的病害,对行车舒适性和安全造成不利影响。为明确路面病害检测领域最适用模型特征,首先分析深度卷积神经网络结构及激活函数、损失函数应用效果,使用R-CNN、YOLO系列中最具代表性的网络采用路面病害数据集进行训练、验证、测试工作,对模型训练评价指标及检测效果的影响因素开展深入研究。结果表明,YOLOv5算法相比于其他算法能够有效提高日常巡检效率,根据测试集数据显示,YOLOv5算法最高检测准确率可达到90.01%,具有显著优势。Under the combined effect of traffic load,natural aging and environmental impact,the pavement surface produce complex distress with different severity during service,which will adversely affect driving comfort and safety.In order to clarify the most suitable model in the field of pavement distress detection,the structure of deep convolutional neural network was analyzed and the application effect of activation function and loss function were evaluated in this paper.The most representative network in R-CNN and YOLO series were used to train,verify and test the pavement distress data set.The training evaluation index and the influencing factors of detection effect of the model were studied in depth.The results show that YOLOv5algorithm can effectively improve the efficiency of daily inspection compared with other algorithms.According to the test set data,the highest detection accuracy of YOLOv5algorithm can reach 90.01%,which has obvious advantages.

关 键 词：路面病害 深度学习 机器视觉 目标检测 性能对比 

分 类 号：U415.5[交通运输工程—道路与铁道工程]