Surface defects such as depressions, heaving, and irregular undulations frequently develop on aging concrete bridge decks under repeated traffic loading and environmental effects. Accurate and objective identification of such defects is essential for structural serviceability and safety, yet manual inspection remains labor-intensive and subjective. This study develops a systematic framework for surface defect identification through geometric feature augmentation with a streamlined point cloud learning strategy. In practical engineering scenarios, point cloud data of concrete bridge decks can be periodically acquired via vehicle-mounted mobile laser scanning (MLS) systems and subsequently streamlined for analysis. The proposed method heightens defect sensitivity by extracting interpretable geometric descriptors, further integrating multi-scale representations to capture surface defects across varying spatial extents. Evaluated on a public point-level annotated benchmark, the proposed method clearly outperforms the same network trained with geometric coordinates only. To improve result reliability, all experiments were repeated four times with different random seeds, and the performance is reported as mean ± standard deviation. Results show that the proposed method achieves a precision of 0.597 ± 0.021 and an accuracy of 0.933 ± 0.009 under the benchmark protocol. Overall, these results demonstrate a reproducible proof of concept under controlled benchmark conditions for bridge deck surface defect segmentation, while broader cross-site and cross-sensor validation will be pursued in future work.
Sun et al. (Mon,) studied this question.