Xiaohui Shi

Xiaohui Shi

Auto-generated profile, based on publicly available information

Xiaohui Shi

Beijing University of Technology

Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 October 2026 in Engineering Research Express
Verified
Journal Article
Closed access
XSXiaohui Shi
Xiaohui Shi
YBYan Bao
Yan Bao
CTChao Tang
Chao Tang
YWYong Wang
Yong Wang
KYKun Yang
Kun Yang
To address the challenges of irregular geometries, multi-scale variation, and diverse target types in road damage detection, we propose an improved deep learning model based on YOLOv8, named YOLO-DMD (Deform–Multiple–Dynamic). The model enhances detection performance through three key components: C2f-Deform, which introduces deformable convolutions to improve adaptability to irregular geometric features; a multi-scale dilated attention module that integrates multi-scale dilated convolutions with attention for enhanced feature representation; and a dynamic head that decouples scale-, spatial-, and task-aware representations to refine classification and localization. Experiments on the public SVRDD dataset show that YOLO-DMD achieves an mAP@0.5 of 79.8% and a recall of 73.3%, demonstrating competitive and often superior performance compared with several mainstream detectors, including YOLOv5 and YOLOv8. Ablation studies further verify the individual effectiveness of each module and their complementary interactions. In addition, cross-dataset evaluations on RDD2020 and GRDDC-US indicate improved robustness under domain shift. Although YOLO-DMD introduces additional computational parameters, it maintains real-time inference capability and provides a practical balance between accuracy and efficiency, making it suitable for automated road damage detection.