To realize the intelligent and accurate measurement of pavement surface potholes, an improved You Only Look Once version three (YOLOv3) object detection model combining data augmentation and structure optimization is proposed in this study. First, color adjustment was used to enhance the image contrast, and data augmentation was performed through geometric transformation. Pothole categories were subdivided into P1 and P2 on the basis of whether or not there was water. Then, the Residual Network (ResNet101) and complete IoU (CIoU) loss were used to optimize the structure of the YOLOv3 model, and the K-Means++ algorithm was used to cluster and modify the multiscale anchor sizes. Lastly, the robustness of the proposed model was assessed by generating adversarial examples. Experimental results demonstrated that the proposed model was significantly improved compared with the original YOLOv3 model; the detection mean average precision (mAP) was 89.3%, and the F1-score was 86.5%. On the attacked testing dataset, the overall mAP value reached 81.2% (−8.1%), which shows that this proposed model performed well on samples after random occlusion and adding noise interference, proving good robustness.

Traditional automatic pavement distress detection methods using convolutional neural networks (CNNs) require a great deal of time and resources for computing and are poor in terms of interpretability. Therefore, inspired by the successful application of Transformer architecture in natural language processing (NLP) tasks, a novel Transformer method called LeViT was introduced for automatic asphalt pavement image classification. LeViT consists of convolutional layers, transformer stages where Multi-layer Perception (MLP) and multi-head self-attention blocks alternate using the residual connection, and two classifier heads. To conduct the proposed methods, three different sources of pavement image datasets and pre-trained weights based on ImageNet were attained. The performance of the proposed model was compared with six state-of-the-art (SOTA) deep learning models. All of them were trained based on transfer learning strategy. Compared to the tested SOTA methods, LeViT has less than 1/8 of the parameters of the original Vision Transformer (ViT) and 1/2 of ResNet and InceptionNet. Experimental results show that after training for 100 epochs with a 16 batch-size, the proposed method acquired 91.56% accuracy, 91.72% precision, 91.56% recall, and 91.45% F1-score in the Chinese asphalt pavement dataset and 99.17% accuracy, 99.19% precision, 99.17% recall, and 99.17% F1-score in the German asphalt pavement dataset, which is the best performance among all the tested SOTA models. Moreover, it shows superiority in inference speed (86 ms/step), which is approximately 25% of the original ViT method and 80% of some prevailing CNN-based models, including DenseNet, VGG, and ResNet. Overall, the proposed method can achieve competitive performance with fewer computation costs. In addition, a visualization method combining Grad-CAM and Attention Rollout was proposed to analyze the classification results and explore what has been learned in every MLP and attention block of LeViT, which improved the interpretability of the proposed pavement image classification model.

A prevalent application of ground-penetrating radar (GPR) for asphalt concrete (AC) pavements is predicting in-situ density, which is an important pavement quality parameter. Accurate estimation of the dielectric constant is required when using GPR for pavement inspections. Surface moisture is a challenge because it masks the GPR reflection amplitudes. Surface moisture could result from natural rainfall events or from sprayed water by roller compactors during AC construction. This study introduces an efficient method to filter GPR signals from surface moisture effects while maintaining density information intact. Through numerical simulations and laboratory testing data, the surface moisture effect was isolated. A singular value decomposition-based signal correction approach was proposed. Data from six different AC sections were utilized to develop the correction algorithm. Finally, for validation, the developed algorithm was applied to GPR scans collected in two extreme conditions: dry and wet, which were collected on another AC pavement before and after a rain event. Results supported the need for surface moisture correction, especially for AC density monitoring during compaction. After applying the correction, the mean absolute error (MAE) of dielectric constant estimation (used to predict AC density) dropped from 16% to 2.6% and, therefore, the AC density prediction MAE dropped from 8.9% to 1.3%.