防灾减灾工程学报2025，Vol.45Issue(6)：1421-1433,13.DOI:10.13409/j.cnki.jdpme.20250430038
基于改进YOLOv8的隧道渗漏水轻量化检测
Lightweight Detection of Tunnel Water Leakage Based on Improved YOLOv8 Model
摘要
Abstract
To address the challenges of low efficiency in traditional methods for tunnel water leakage detection and the large number of parameters and insufficient real-time performance of existing deep learning models,this study proposed a lightweight instance segmentation model based on an improved YOLOv8n-seg.The coordinate attention(CA)mechanism was introduced to enhance feature represen-tation in target regions.The backbone network was replaced with MobileNetV4 to reduce computa-tional complexity,and the EfficientHead segmentation head was incorporated to improve the efficien-cy of feature decoding.These improvements collectively enhanced both detection accuracy and infer-ence speed in complex environments.The experiments were conducted on a tunnel water leakage data-set constructed using 3D laser scanning(including 3 140 enhanced images).Ablation experiments were employed to validate the effectiveness of each module.The CA mechanism increased the average pre-cision(AP)by 0.82%,MobileNetV4 increased the AP to 81.21%while reducing the number of pa-rameters by 43.2%,and the EfficientHead further optimized segmentation details.After joint improve-ments,the model achieved an AP of 83.21%,an F1-score of 78.53%,a number of parameters of 1.96M,and an inference speed of 355.2 FPS,representing a 6.6%increase over the original YO-LOv8n-seg.Comparative experiments demonstrated that the proposed model significantly outper-formed mainstream models such as Mask R-CNN in lightweight indicators(number of parameters and GFLOPs),while achieving accuracy comparable to that of two-stage methods,thereby meeting the re-al-time detection requirement for tunnel water leakage.This study provides an efficient and reliable lightweight solution for structural health monitoring in tunnels,offering practical value for engineering applications.关键词
隧道病害/深度学习/病害识别/实例分割/轻量化检测Key words
tunnel defect/deep learning/defect detection/instance segmentation/lightweight detection分类
交通工程引用本文复制引用
TIAN Chenghang,ZHAO Limeng,LI Yonggang,ZHANG Fangjian,ZHANG Wengang,SUN Weixin,DING Xuanming..基于改进YOLOv8的隧道渗漏水轻量化检测[J].防灾减灾工程学报,2025,45(6):1421-1433,13.基金项目
广州地铁集团有限公司科研委托项目(JT204-100111-23001)、重庆铁路投资集团有限公司科研委托项目(CSTB2022TIAD-KPX0101)、中铁二局集团有限公司科研委托项目(N2023G045)资助 （JT204-100111-23001）
