基于残差神经网络的路面类型辨识模型  

Pavement Type Recognition Model Based on ResidualNeural Network

作　　者：袁永辉 陈震 王国 李昌亮 陈威 郎洪 蒋愚明 YUAN Yonghui;CHEN Zhen;WANG Guo;LI Changliang;CHEN Wei;LANG Hong;JIANG Yuming(China Overseas Construction Limited,Shenzhen,Guangdong 518057,China;The Key Laboratory of Road and Traffic Engineering of the Ministry of Education,Tongji University,Shanghai 201804,China)

机构地区：[1]中海建筑有限公司,广东深圳518057 [2]同济大学道路与交通工程教育部重点实验室,上海201804

出　　处：《公路工程》2025年第1期113-120,共8页Highway Engineering

基　　金：上海市科委创新行动计划软科学项目(23692118200);国家自然科学基金项目(62206201)。

摘　　要：现有的路面状况检测系统在道路类型辨识上存在局限性,不能准确、自动地辨识道路类型,这影响了道路状况检测系统的自动化程度与工作效率。提出了一种改进残差神经网络ResNet101的路面类型辨识模型,能够准确、快速地自动识别沥青、水泥路面类型。采用三维结构光扫描设备采集中国5个省、市(即浙江、江苏、四川、上海、山西)具有不同路面特征的沥青路面和水泥路面图像,总计35 500张2D与3D路面图像,其中70%、20%和10%分别被随机抽取出用于模型的训练、验证和测试。模型在测试数据集中的识别准确率达到0.973,F1-score达到0.971,其中沥青路面和水泥路面的识别准确率分别达到0.980与0.969,同时模型推理速度为10.06 ms/张。提出的方法能够为提高基于三维结构光扫描的路面状况评定系统的自动化水平带来实质性的帮助。Existing pavement condition detection systems have limitations in road type recognition and cannot accurately and automatically recognize road types,which affects the automation degree and work efficiency of road condition detection systems.This paper proposes an improved residual neural network ResNet101 pavement type recognition model,which can accurately and quickly recognize asphalt and cement pavement types automatically.A 3D structured light scanning device was used to collect asphalt and cement pavement images with different pavement characteristics from five provinces and cities in China,namely,Zhejiang,Jiangsu,Sichuan,Shanghai and Shanxi,totaling 355002D and 3D pavement images,of which 70%,20%and 10%were randomly selected for training,validation and testing of the model,respectively.The model achieves a recognition accuracy of 0.973 and F 1-score of 0.971 in the test dataset,with recognition accuracies of 0.980 and 0.969 for asphalt and concrete pavements,respectively,while the model inference speed is 10.06 ms/picture.The proposed method can provide substantial assistance in improving the automation level of the road condition assessment system based on three-dimensional structured light scanning.

关 键 词：道路工程 路面类型 残差神经网络 沥青路面 水泥路面 

分 类 号：U416.2[交通运输工程—道路与铁道工程]