Abstract: In the evaluation of pavement wear detection, the traditional sand patch method and drainage method for measuring pavement texture depth are inefficient and susceptible to human factors. Although the laser cross-section detection equipment offers fast detection speed, it is influenced by the position of the longitudinal measurement line and the pavement alignment, leading to errors in DMP, RW, and PWI, which affects the reliability of the wear evaluation results. In order to accurately evaluate and quickly detect pavement wear, a method that utilized terrain prominence as a wear indicator for grade judgment and employed the ResNet50 model for image classification of pavement wear was proposed. Images and 3D texture data of roads with the same pavement type but different operation years were collected, and terrain prominence was determined as the evaluation index to judge the wear grade of the images. Then, the ResNet50 convolutional neural network was used as the initial training model, and a transfer learning strategy was adopted to fine-tune the model parameters, thereby improving the training speed and classification accuracy of the model. The research results show that the terrain prominence index can well reflect the pavement wear condition, and the average accuracy of the pavement wear detection model based on ResNet50 can reach 98.49%. By deploying the proposed model on mobile devices, the developed pavement wear detection APP can be run on any Android-system-based mobile phone, meeting practical application needs and achieving rapid and accurate detection of pavement wear.

Key words: highway engineering; pavement wear detection; pavement texture; terrain prominence; ResNet50; model deployment

摘要： 路面磨耗检测评价中传统铺砂法和排水法测量路面构造深度效率低且易受人为因素影响，而激光断面检测设备虽检测速度快，但受纵向测线位置和路面线形影响，导致平均断面深度（DMP）、 路面磨耗率（RW）和路面磨耗指数（PWI）产生误差，影响磨耗评价结果的可靠性。为了对路面磨耗进行准确评价与快速检测，提出以地形突出度为磨耗指标进行等级判断、并基于ResNet50模型进行路面磨耗图像分类的方法。采集了相同路面类型不同运营年限道路的图像以及三维纹理数据，确定以地形突出度为评价指标对图像进行磨耗等级判断。然后以ResNet50卷积神经网络作为初始训练模型，采取迁移学习的策略对网络模型进行参数微调，提高模型的训练速度和分类精度。研究结果表明：地形突出度指标能够良好地反映路面磨耗状况，基于ResNet50的路面磨耗检测模型平均准确率可达98.49%。将模型部署于移动端，开发完成的路面磨耗检测APP可运行在任意Android系统手机上，满足实际运用需求，实现了路面磨耗的快速准确检测。

关键词: 道路工程；路面磨耗检测；路面纹理；地形突出度；ResNet50；模型部署