基于深度学习的隧道病害快速检测方法研究  

Research on Fast Detection Method of Tunnel Disease Based on Deep Learning

作　　者：陆志溢 LU Zhi-yi(China Design Group Co.,Ltd.)

机构地区：[1]华设集团股份有限公司

出　　处：《智能建筑与智慧城市》2025年第2期174-176,共3页Intelligent Building & Smart City

摘　　要：针对隧道渗漏水和掉块病害的快速目标检测问题,论文提出了一种基于深度学习的隧道病害快速检测方法,能够实现快速、准确的自动识别。首先,搭建YOLOv3模型进行不同尺度的特征提取;其次,收集病害图片并进行病害标注,制作足够的数据集;然后,借鉴了迁移学习的思想,通过冻结和解冻两阶段训练方法获得网络权重;最后,输入测试集图片,进行预测。实验结果表明,YOLOv3模型对于渗漏水和掉块病害的目标检测AP值分别为90.91%和100%,在一阶段方法快速识别的基础上具有较高的准确率。Aiming at the problem of rapid target detection of tunnel water leakage and block-drop disease,this paper proposes a rapid detection method of tunnel disease based on deep learning,which can realize rapid and accurate automatic identification.Firstly,it constructs YOLOv3 model for feature extraction at different scales;secondly,it collects disease pictures and label the disease to make enough data sets;then,it uses the idea of migration learning for reference,and obtains the network weights through the two-stage training method of freezing and thawing;finally,it inputs the test set picture and make predictions.The experimental results show that the YOLOv3 model has 90.91%and 100%AP values for the target detection of water leakage and block disease,respectively.It has a high accuracy rate based on the rapid identification of the one-stage method.

关 键 词：隧道 病害识别 深度学习 YOLOv3 多尺度融合 

分 类 号：U456.3[建筑科学—桥梁与隧道工程]