Jia-Wei Guo

Intelligent Pavement Recognition and Automated Repair Algorithm Based on LU-Net and DP

Crack repair has long been a critical task in highway maintenance. With the continuous expansion of the scale of roads under maintenance, traditional manual patching can no longer meet the growing demand. Current mainstream pavement crack repair technologies generally have the shortcoming that real-time performance and accuracy are difficult to balance. To address the dual problems of insufficient crack detection accuracy and low efficiency in automated repair path planning, this study is committed to developing an efficient intelligent repair method. It proposes a two-stage algorithm integrating intelligent crack detection and automated repair. In the intelligent detection stage, a lightweight U-shaped hybrid network (LU-Net), which combines U-Net and MobileNetV2, is designed to facilitate intelligent crack identification and trajectory extraction. In the path planning stage, an exact algorithm based on dynamic programming (DP) is introduced to determine the optimal repair sequence automatically. Comparative experiments verified the effectiveness and accuracy of the method: the detection accuracy of LU-Net reached 88.35% with a mean intersection over union (MIoU) of 80.39%; for scenarios with 5–14 crack paths, the proposed dynamic programming algorithm not only found the global optimal solution but also reduced the solution time by 90% compared with existing exact algorithms. The results show that the LU-Net segmentation model and DP algorithm can significantly save operation time while ensuring accuracy.

An intelligent pavement crack recognition and automatic repair algorithm based on improved DeepLabv3+ and greedy algorithm

To address the growing need for pavement maintenance, this paper presents a two-stage algorithm that integrates intelligent crack identification with automated repair path planning, aiming to boost the intelligent repair efficiency of pavement crack diseases. In the intelligent crack identification phase, a lightweight modified DeepLabv3 + model is used for pixel – level pavement crack segmentation, and operations such as thinning, spur removal and crack segmentation are applied to extract crack skeleton information. In the crack repair path planning phase, the absolute greediness of traditional greedy algorithms is overcome by adding a probability – based crack endpoint selection mechanism. Meanwhile, to boost the greedy algorithm's in – depth optimization ability, three local optimization operators, namely 2 – opt, 3 – opt, and 2 – wopt, are designed to further optimize automated repair paths. To verify the improved greedy algorithm's efficiency, the paper designed experimental scenarios with 10, 20 and 30 cracks each in the repair area. In these scenarios, it compared three algorithms: ant colony optimization (ACO), traditional greedy algorithm (GA) and the proposed improved greedy algorithm (IGA). The comparison focused on optimization ability, efficiency and solution stability. Results showed that the IGA outperformed the other two in optimization and efficiency. This paper integrates deep learning with an improved optimization algorithm to present an integrated solution for one - step road crack repair.

Source apportionment and health risk assessment of arsenic contamination in groundwater in Xinxiang city based on the PCA-PMF model and WQI method