Automatic detection and segmentation of potholes and raveling are crucial for preventive maintenance and ensuring roads structural health. However, the extraction of pavement potholes was mainly based on traditional image processing methods, which have proven to be ineffective and inaccurate. Additionally, the absence of a unified pothole and raveling dataset has resulted in the lack of benchmarks for evaluating various methods. This work makes three contributions to address these problems. Firstly, we have curated the pavement pothole and raveling distress detection and segmentation datasets. Secondly, a two-step pavement pothole and raveling detection and segmentation method was proposed. In the initial step, an automated pavement pothole and raveling detection model was developed using the modified YOLOX. Subsequently, the segmentation model, named dual self-attention segmentation network (DSASNet), was proposed to segment distress by extracting mode-sensitive features from intensity and range images using two parallel Twins-SVT self-attention branches. Moreover, we design a mid-fusion module to adaptively fuse mode-specific and scale-specific features. Finally, a pyramid pooling module (PPM) is connected to further enhance the segmentation capability for potholes and raveling of various sizes and shapes. The F1-score and Intersection over union (IoU) of the proposed DSASNet on the test set are 93.65% and 0.881, respectively, outperforming other baseline methods. Furthermore, we conduct an experiment to quantitatively compare the two-step method with the one-step method using only a single semantic segmentation model. The results demonstrated clear advantages of the proposed two-step method in terms of accuracy and efficiency for pavement pothole and raveling segmentation.

Roads are the foundation of intelligent transportation systems, yet cracks are widely present in roads and seriously affect system performance. Cracks not repaired promptly can develop into severe road defects, significantly increasing the risk of traffic accidents. Researchers in the community have started to focus on the automatic sensing of cracks in asphalt pavements, while it is still a challenging task on concrete pavements. Cracks in the concrete pavement are easily recognized as interrupted segments rather than a continuous whole due to the interference of the surface texture. This mistake can seriously mislead the judgment of cracks and subsequent road repair. In this paper, we aim to solve the challenge by enhancing contextual information about cracks within the images. We first extract the information from the local and global representations using image information and then fuse it into complete contextual information by a designed multilayer perceptron. Finally, we use the discriminative loss to constrain the edges of cracks and backgrounds using complete crack contextual information. We have collected and annotated several images of concrete pavements from several significant provinces in China. Experiments show that our method achieves the best performance compared to state-of-the-art methods, especially in edge determination.

Standard back propagation (BP) neural network has disadvantages such as slow convergence speed, local minimum and difficulty in definition of network structure. In this paper, an learning vector quantization (LVQ) neural network classifier is established, then it is applied in pattern classification of two-dimensional vectors on a plane. To test its classification ability, the classification results of LVQ neural network and BP neural network are compared with each other. The simulation result shows that compared with classification method based on BP neural network, the one based on LVQ neural network has a shorter learning time. Besides, its requirements for learning samples and the number of competing layers are also lower. Therefore it is an effective classification method which is powerful in classification of two-dimensional vectors on a plane.