南京信息工程大学学报2025，Vol.17Issue(3)：328-339,12.DOI:10.13878/j.cnki.jnuist.20240710001
基于Crack-YOLACT的道路裂缝提取
Road crack detection based on Crack-YOLACT
摘要
Abstract
To address the issue that existing road crack detection algorithms often adopt a two-stage approach of de-tection followed by segmentation,resulting in two independent processes and inefficient performance in practical ap-plications,this paper proposes an integrated end-to-end road crack detection method.Firstly,a more lightweight backbone feature extraction network for cracks is employed to reduce computational costs and accelerate the model's inference process.Then,a crack feature fusion module,which integrates an asymptotic feature pyramid network and an adaptive spatial fusion module,is utilized to improve the model's detection capability for small target cracks in complex scenarios.Finally,the proposed method is experimentally validated on two datasets with significant differ-ences:one consists of complex urban street scene data collected by vehicle-mounted scanning systems,and the other is the public dataset Crack500.The results show that,when applied to the complex urban street scene dataset,the method achieves accuracy,recall,and F1 score of 86.3%,84.1%,and 85.2%,respectively;and on the Crack500 dataset,it achieves accuracy,recall,and F1 score of 82.4%,80.2%,and 81.3%,respectively for road crack detection tasks.These results highlight the method's accuracy in identifying fine cracks and its robustness in different practical environments.关键词
道路裂缝/实例分割/注意力机制/轻量化网络Key words
road crack/instance segmentation/attention mechanism/lightweight network分类
交通运输引用本文复制引用
袁文豪,尹珺宇,方莉娜,吴尚华,郭明华,侯海涛..基于Crack-YOLACT的道路裂缝提取[J].南京信息工程大学学报,2025,17(3):328-339,12.基金项目
福建省高校产学合作项目(2023H6032) （2023H6032）
福建省交通运输科技项目(ZH202317) （ZH202317）
国家自然科学基金(42071446) （42071446）
