基于深度学习的水工混凝土结构表面缺陷边框级检测方法研究  

Research on Border-level Detection Method for Surface Defects of Hydraulic Concrete Structures Based on Deep Learning

作　　者：王洪钰 苏超[1] 王文君 WANG Hong-yu;SU Chao;WANG Wen-jun(College of Hydrology and Water Resources,Hohai University,Nanjing 210098,China)

机构地区：[1]河海大学水利水电学院,江苏南京210098

出　　处：《水电能源科学》2025年第3期101-105,151,共6页Water Resources and Power

摘　　要：为提高水工混凝土结构表面缺陷检测的效率及准确性,提出了一种边框级缺陷检测模型DetDamage。该模型首先采用轻量化主干网络进行特征提取,降低整体计算成本;然后通过多级特征融合网络对特征图进行融合和增强;最后通过解码头合并得到预测结果。测试结果表明,DetDamage对于四类缺陷分别可以实现77.24%(裂缝)、87.43%(剥落)、82.61%(露筋)、83.67%(渗漏)的检测精度,平均检测精度为82.74%。这表明该模型在水工混凝土表面缺陷的检测中具有良好的效果和可行性。In order to improve the efficiency and accuracy of surface defect detection of hydraulic concrete structures,a frame-level defect detection model DetDamage was proposed.Firstly,the lightweight backbone network was used for feature extraction to reduce the overall computing cost.Then,the feature map was fused and enhanced through a multi-level feature fusion network.Finally,the prediction results were obtained by undocking and merging.The test results show that the DetDamage can achieve 77.24%(cracks),87.43%(spalling),82.61%(exposed ribs)and 83.67%(leakage)detection accuracy for the four types of defects,with an average detection accuracy of 82.74%.This shows that the model has good effect and feasibility in the detection of surface defects in hydraulic concrete.

关 键 词：水工混凝土结构 缺陷检测与量化 深度学习 智慧管理 

分 类 号：TV431[水利工程—水工结构工程]