基于探地雷达的沥青路面内部病害快速识别

Rapid recognition of asphalt pavement internal diseases based on ground penetrating radar

为了提高沥青路面内部病害的检测精度及效率,开展了路面探地雷达(GPR)图像的自动识别研究.采用三维探地雷达(3D-GPR)设备高效、无损地探测沥青路面内部结构,以获取海量3D-GPR图像数据;通过多维度GPR图像辨识内部病害(横向裂缝和层间不良)的回波特征,矩形框准确标注纵断面GPR图中病害特征,进而构建内部病害回波特征GPR图像数据集(训练集、验证集和测试集);基于深度学习技术,引入YOLOv4(you only look once version 4)算法模型,首先利用训练集和验证集完成模型网络参数的迭代更新,然后利用测试集进行模型综合检测性能评估.研究结果表明:YOLOv4模型在测试集上测试的综合检测精度大于95%,并且其检测视频的每s帧数也超过30;而层间不良的回波特征相对内部横向裂缝识别更加准确;该模型可以实时、高精度自动识别出沥青路面GPR图像中病害回波特征.

To improve the detection accuracy and efficiency of internal diseases within asphalt pavement,a study on the automatic recognition of pavement ground penetrating radar(GPR)images was carried out.A massive number of three-dimensional GPR(3DGPR)images were acquired,through the efficient and non-destructive inspection of asphalt pavement structures using the 3D-GPR system.The echo features of concealed diseases(transverse crack and debonding)were identified through multi-dimensional GPR images,and they were accurately marked with rectangular boxes in the longitudinal GPR diagram.The GPR image datasets of internal disease echo features,including training set,validation set,and test set,were constructed.You only look once version 4(YOLOv4)algorithm model originating from deep learning technology,was updated iteratively in the training set and validation set.Subsequently,the comprehensive detection performance of the trained model was evaluated by the test set.Research results show that the comprehensive detection accuracy of the YOLOv4 model exceeds 95%,and the frames per second for processing a GPR video is also more than 30.The echo features of debonding are to be identified more accurately than those of internal crack.The YOLOv4 model can automatically identify the disease echo features from the GPR images of asphalt pavement in real-time with high accuracy.