2026-10-11T19:17:10Z https://www.scidb.cn/oai

10.57760/sciencedb.j00240.00099 2026-01-20T08:43:45Z

2026-01-20 MSANet doi:10.57760/sciencedb.j00240.00099 en This paper proposes a multi-stage spatial perception and detail enhancement road damage detection model. A hybrid attention module for spatial perception and detail enhancement is designed. Through the collaborative mechanism of global direction perception and detail enhancement, long-range spatial dependencies are constructed, and the ability to represent the details of weak textures and blurred edges is significantly improved. A cross-scale feature cross-fusion module is constructed to optimize the network neck architecture to achieve heterogeneous cascaded fusion of cross-scale features, effectively balancing the collaborative expression of local spatial details and global semantic information. In addition, the improved C3K2 module embeds coordinate-aware convolution, effectively optimizing the spatial coupling modeling efficiency of high-dimensional features through spatial information enhancement. System experiments on the RDD2022 benchmark dataset show that the model in this paper effectively identifies various road damages, maintaining a real-time inference speed of 142 FPS while achieving an improvement of 1.9% in mAP@0.5, 4.9% in mAP@0.5:0.95, and 1.8% in F1-Score compared to the existing optimal methods. The mAP@0.5 reaches 87.7%. Ablation experiments verify the contribution of each module. Cross-dataset testing and generalization testing further confirm the excellent detection robustness and engineering applicability of this model. YOLO; Object Detection; Road damage detection wang yu sheng PUBLIC https://creativecommons.org/licenses/by-nc-nd/4.0/ dataset http://www.doi.org/10.11834/jig.250422 Science Data Bank