融合DCGAN网络与ResNet检测模型的路面摩擦系数检测  

Recognition Model for the Friction Coefficient of Pavements Integrates the Generative DCGAN Model and the ResNet Recognition Model

作　　者：闫章存 胡万欣 张浩然 李佳杰 岳李圣飒 YAN Zhangcun;HU Wanxin;ZHANG Haoran;LI Jiajie;YUE Lishengsa(Key Laboratory of Road and Traffic Engineering Ministry of Education,Tongji University,Shanghai 201804,China;School of Computer Science,South Central Minzu University,Wuhan 430074,China;China Academy of Transportation Science,Beijing 100029,China)

机构地区：[1]同济大学道路与交通工程教育部重点实验室,上海201804 [2]中南民族大学计算机科学学院,武汉430074 [3]交通运输部科学研究院,北京100029

出　　处：《综合运输》2025年第3期127-135,173,共10页China Transportation Review

基　　金：中南民族大学引进人才科研启动基金自科项目(YZZ18013);大学生创新创业训练计划省级项目(SCX2024043)。

摘　　要：道路路面摩擦系数是制定驾驶安全管控策略的关键参数。为精确获取该参数,本研究融合深度卷积生成对抗网络(DCGAN)与残差网络(Resnet)提出非接触式摩擦系数检测框架。包含三步,首先,以沥青混凝土路面为对象设计摩擦系数测量实验,用摆式仪测量不同湿度路面摩擦系数,同步器采集路面图像;其次,关联路面图像与对应摩擦系数建立摩擦系数图像标签数据集,并用DCGAN网络对数据集进行增强,提升场景覆盖度;最后,使用增强数据集训练Resnet摩擦系数检测模型,并设计对比实验验证检测性能。结果:所提检测模型精度达到93.43%,高于MLP-Mixer(90.97%)和ResNet18(92.55%),且在实际数据测试性能稳定。该方法将有效提升摩擦系数检测精度,降低检测成本,为车辆安全管控系统提供准确的路面信息。The road surface friction coefficient is a key factor for making the control strategy of safety driving.To acquire the accuracy friction coefficient of the road surface,a non-contact friction coefficient detection method that combines a Deep Convolutional Generative Adversarial Network(DCGAN)and a Residual Network(Resnet)using the forward video of the vehicle is proposed.This methos consists of three steps:first,we design an experiment to measure the friction coefficient typical urban road pavements and measure the friction coefficient of the pavement surface under different precipitation amounts using a pendulum instrument,at the same,the video sensors capture the image of the road surface.Secondly,we match the road surface image and the friction coefficient values to create an image-labeled friction coefficient dataset.To improve the coverage of the scenario,we augment the friction coeficient dataset by the DCGAN network.Finally,we train the Resnet-50 detection model with the augmented dataset and conduct a comparison experiment to verify the detection performance.The result shows that the accuracy of the ResNet50 is 93.43%,which is higher than that of MLP-Mixer(90.97%)and ResNet18(92.55%).Furthermore,the detection method is stable in real tests.This system improves the accuracy of friction coeficient detection,minimizes detection costs,and provides more accurate information of road surface for driving assistance system.

关 键 词：智能交通 驾驶辅助系统 路面摩擦系数检测 生成对抗网络 残差网络 

分 类 号：U461[机械工程—车辆工程]