Three-dimensional (3D) ground-penetrating radar is an effective method for detecting internal crack damage in pavement structures. Inefficient manual interpretation of radar images and high personnel requirements have substantially restrained the generalization of 3D ground-penetrating radar. An improved Crack Unet model based on the Unet semantic segmentation model is proposed herein for 3D ground-penetrating radar crack image processing. The experiment showed that the MPA, MioU, and accuracy of the model were improved, and it displayed better capacity in the radar image crack segmentation task than current mainstream algorithms do, such as deepLabv3, PSPNet, and Unet. In the test dataset without cracks, Crack Unet is on the same level as deepLabv3 and PSPNet, which can meet engineering requirements and display a significant improvement compared with Unet. According to the ablation experiment, the MPA and MioU of Unet configured with PMDA, MC-FS, and RS modules were larger than those of Unet configured with one or two modules. The PMDA module adopted by the Crack Unet model showed a higher MPA and MioU than the SE module and the CBAM module did, respectively. The results show that the Crack Unet model has a better segmentation ability than the current mainstream algorithms do in the task of the crack segmentation of radar images, and the performance of crack segmentation is significantly improved compared with the Unet model. The Crack Unet model has excellent engineering application value in the task of the crack segmentation of radar images.

In order to improve the prediction accuracy regarding low compaction level of asphalt pavement, this paper carries out indoor tests to detect the voids and dielectric constants of AC-13, AC-16 and AC-25 asphalt mixtures, obtaining their relationship equations via linear fitting and determining the dielectric constant judgment threshold of low compaction level segregation risk points ${\epsilon }_1$. Based on the common mid-point method, three-dimensional ground-penetrating radar is used to obtain the dielectric constant of the physical engineering test section. The researcher can draw the distribution map of the low compaction level segregation risk area according to the judgment threshold ${\epsilon }_1$ of the rough segregation risk points; divide the connected risk areas; determine the regional convex hull; and calculate the regional indicators such as the regional area, the ratio of the convex risk points and the mean value of the regional dielectric constant. The response surface analysis method is used to acquire the model of risk area index and core void ratio. The model is employed to predict and verify the core void ratio in the risk area of the road section and verify the accuracy of the model. The results show that the error range between the predicted voids and the measured voids is −0.4%~+0.4%, and the mean absolute value of the error is 0.25%. Compared with the mean measured voids of 6.63%, the relative error is 3.77%, indicating that the model can accurately predict the regional low compaction level segregation degree.

The manual interpretation of ground-penetrating radar images is characterised by long interpretation cycles and high staff requirements. The automated interpretation schemes based on support vector machines, digital images, convolutional neural networks and other techniques proposed in recent years mainly detect features from B-scan slices of 3D ground-penetrating radar data, without taking full advantage of the multi-channel acquisition of data from 3D ground-penetrating radar and joint discrimination. This paper proposes a void recognition algorithm based on cluster analysis algorithm, using VRADI algorithm to process 3D ground-penetrating radar B-Scan, using DBSCAN clustering algorithm to divide clusters and remove noise; proposes correlation weighting coefficient Wi,j to quantitatively evaluate the degree of correlation of different survey channels, proposes prime relative position coefficient Pd indicator to quantitatively evaluate the position similarity, and proposes weighted homocentric overlap coefficient Pr indicator to quantify signal similarity. This paper applies the algorithm to carry out physical engineering experiments and uses binary logistic regression analysis to develop a correlation model. The experimental results show that significance of Pd and Pr are less than 0.05, both of which are important influencing indicators for the determination of the presence or absence of void. With an optimal critical probability of 0.4, the recognition accuracy of VRADI algorithm increases from 71.7% to 92.2%. The VRADI algorithm combined with the cluster analysis algorithm outperforms manual recognition in terms of accuracy (92.2% > 83.9%) and recall (90.5% > 86.9%), and the algorithm has good engineering application value.