TY  - JOUR
AU  - Ji, Rendong
AU  - Zhang, Xiaojun
AU  - Tang, Xiu
AU  - Wang, Xiaoyan
AU  - Xu, Yunlong
AU  - Shi, Jiaxin
PY  - 2026
DA  - 2026/06/29
TI  - Road Crack Segmentation Algorithm Based on Multiple Attention Fusion and Defect Correction
JO  - ICCK Transactions on Intelligent Systematics
T2  - ICCK Transactions on Intelligent Systematics
JF  - ICCK Transactions on Intelligent Systematics
VL  - 3
IS  - 2
SP  - 126
EP  - 144
DO  - 10.62762/TIS.2025.325163
UR  - https://www.icck.org/article/abs/TIS.2025.325163
KW  - atrous residual convolution
KW  - defect correction
KW  - multiple attention fusion
KW  - road crack
KW  - semantic segmentation
AB  - This paper proposes an enhanced U-Net-based segmentation framework for road crack detection that effectively addresses issues such as incomplete segmentation, detail loss, environmental complexity, and crack-pixel imbalance. The model integrates multiple functional modules to improve segmentation performance across varying crack types and scales. Specifically, an atrous residual convolution (ARC) module is embedded in the encoder to expand the receptive field and capture large-scale features. A multiple attention fusion module (MAFM), combined with an efficient channel attention mechanism, is introduced at the bridge stage to emphasize crack-relevant features. In the decoder, a defect correction module (DCM) with deep supervision and adaptive refinement is designed to restore fine-grained crack boundaries, especially for small or subtle defects. The proposed model achieves F1-scores of 78.11%, 90.02%, and 81.60% on CFD dataset, CRACK500 dataset, and HYCrack dataset, respectively. Compared to existing state-of-the-art segmentation models, the proposed approach achieves superior accuracy and better preservation of crack detail. These results demonstrate its practical value and strong potential for widespread application in road crack detection and infrastructure maintenance.
SN  - 3068-5079
PB  - Institute of Central Computation and Knowledge
LA  - English
ER  - 

@article{Ji2026Road,
  author = {Rendong Ji and Xiaojun Zhang and Xiu Tang and Xiaoyan Wang and Yunlong Xu and Jiaxin Shi},
  title = {Road Crack Segmentation Algorithm Based on Multiple Attention Fusion and Defect Correction},
  journal = {ICCK Transactions on Intelligent Systematics},
  year = {2026},
  volume = {3},
  number = {2},
  pages = {126-144},
  doi = {10.62762/TIS.2025.325163},
  url = {https://www.icck.org/article/abs/TIS.2025.325163},
  abstract = {This paper proposes an enhanced U-Net-based segmentation framework for road crack detection that effectively addresses issues such as incomplete segmentation, detail loss, environmental complexity, and crack-pixel imbalance. The model integrates multiple functional modules to improve segmentation performance across varying crack types and scales. Specifically, an atrous residual convolution (ARC) module is embedded in the encoder to expand the receptive field and capture large-scale features. A multiple attention fusion module (MAFM), combined with an efficient channel attention mechanism, is introduced at the bridge stage to emphasize crack-relevant features. In the decoder, a defect correction module (DCM) with deep supervision and adaptive refinement is designed to restore fine-grained crack boundaries, especially for small or subtle defects. The proposed model achieves F1-scores of 78.11\%, 90.02\%, and 81.60\% on CFD dataset, CRACK500 dataset, and HYCrack dataset, respectively. Compared to existing state-of-the-art segmentation models, the proposed approach achieves superior accuracy and better preservation of crack detail. These results demonstrate its practical value and strong potential for widespread application in road crack detection and infrastructure maintenance.},
  keywords = {atrous residual convolution, defect correction, multiple attention fusion, road crack, semantic segmentation},
  issn = {3068-5079},
  publisher = {Institute of Central Computation and Knowledge}
}