基于深度学习的桥梁裂缝病害检测算法研究  

Research on Bridge Crack Disease Detection Algorithm Based on Deep Learning

作　　者：方虹苏 石鑫雨 贺门涛 熊润莲 FANG Hongsu;SHI Xinyu;HE Mentao;XIONG Runlian(School of Automobile,Chang′an University,Xi′an 710064;School of Information Engineering,Chang′an University,Xi′an 710064)

机构地区：[1]长安大学汽车学院,西安710064 [2]长安大学信息工程学院,西安710064

出　　处：《公路交通技术》2025年第2期111-118,126,共9页Technology of Highway and Transport

摘　　要：针对复杂背景下不同裂缝检测场景中各裂缝类型易出现错检及漏检等情况,提出一种改进的裂缝检测算法YOLOv8-CES。先在基础网络YOLOv8结构中引入C2f-SWC模块代替原有C2f模块,后使用轻量级上采样算子CARAFE代替传统上采样算子,同时引入EffectiveSE Module通道特征自适应模块,再通过公开数据集及网络收集方式构建了3000张裂缝检测数据集图像,并采取准确率P及召回率R、平均精度均值AP_(50)及AP_(50-95)定量评价指标来对算法检测效果进行定量评价,最后通过对比试验、消融试验及检测结果可视化,进一步论证YOLOv8-CES在裂缝检测任务中的有效性及优越性。结果表明:YOLOv8-CES可提升检测精度,并在裂缝检测任务中可准确检测裂缝,具备较强的实际应用潜力。A improved crack detection algorithm YOLOv8-CES is proposed to address the common occurrence of false positives and false negatives in different crack detection scenarios under complex backgrounds.Firstly,the C2f SWC module was introduced to replace the original C2f module in the basic network YOLOv8 structure.Then,the lightweight upsampling operator CARAFE was used to replace the traditional upsampling operator,and the Effectiveness SE Module channel feature adaptive module was introduced.Then,3000 crack detection dataset images were constructed through public dataset and network collection methods,and the accuracy P and recall R,average precision mean AP_(50) and AP_(50-95) quantitative evaluation indicators were used to quantitatively evaluate the algorithm detection effect.Finally,through comparative experiments,ablation experiments,and visualization of detection results,the effectiveness and superiority of YOLOv8-CES in crack detection tasks were further demonstrated.The results indicate that YOLOv8-CES can improve detection accuracy and accurately detect cracks in crack detection tasks,demonstrating strong practical application potential.

关 键 词：深度学习 目标检测 桥梁裂缝 YOLOv8 裂缝检测 

分 类 号：U445.71[建筑科学—桥梁与隧道工程]