nav emailalert searchbtn searchbox tablepage yinyongbenwen piczone journalimg journalInfo journalinfonormal searchdiv searchzone qikanlogo popupnotification paper paperNew
基于GAR-Net模型的隧道衬砌厚度及钢拱架识别定位
基金项目(Foundation): 中国中铁股份有限公司科技研究开发计划(2021-专项-08,2023-专项-07,2023-专项-05)
邮箱(Email): 714952479@qq.com;
DOI:
发布时间: 2025-04-30
出版时间: 2025-04-30
网络发布时间: 2025-04-30
移动端阅读
摘要:

针对传统地质雷达法在隧道衬砌厚度和钢拱架识别定位中存在的人工解译效率低和技术依赖性强的问题,提出了一种新型的群体行为识别神经网络模型(Group Activity Recognition Net,GAR-Net)。通过集成语义分割、目标检测和模板匹配三个子网络,实现了对隧道衬砌厚度和钢拱架间距的自动化识别与精准定位。其中,语义分割子网络用于从地质雷达图像中提取出衬砌厚度区域;目标检测子网络用于提取钢拱架对象;模板匹配子网络则通过与预先定义的模板进行匹配,来确定钢拱架的具体位置和衬砌的厚度。经过对特定数据集的模型训练和评价指标的严格测试,GAR-Net模型对衬砌厚度识别的平均交并比(Mean Intersection over Union,mIoU)达到了97.1%,对钢拱架识别的平均精度(Average Precision,AP)达到了96.09%。可见,GAR-Net模型在地质雷达检测隧道衬砌厚度及钢拱架识别定位方面具有良好的测试效果和较高的应用潜力。

Abstract:

This article proposed a new neural network architecture group activity recognition(GAR-Net) to address the traditional geological radar method problems of low manual interpretation efficiency and strong technical dependence in the identification of tunnel lining thickness and steel arches. This network architecture achieved automated recognition and precise positioning of tunnel lining thickness and steel arches spacing by integrating three sub networks, semantic segmentation, object detection and template matching. The semantic segmentation sub network was responsible for extracting the lining thickness areas from the ground penetrating radar image. The object detection sub network was responsible for extracting steel arch objects. The template matching subnetwork determined the specific position of the steel arch and the thickness of the lining by matching with pre-defined templates. After rigorous testing of model training and evaluation metrics on specific datasets, GAR-Net achieved the mIoU(Mean Intersection over Union) of 97.1% in lining thickness recognition and the AP(Average Precision) of 96.09% in steel arches recognition. It can be seen that the GAR-Net algorithm has good testing performance and high application potential in the detection of tunnel lining steel arch and thickness recognition by ground penetrating radar.

参考文献

[1]胡向东,白楠,李鸿博.圣彼得堡地铁1号线区间隧道事故分析[J].隧道建设,2008,28(4):418-422.

[2]奉建军,林元铖,周斌,等.基于改进U-Net网络的地质雷达图谱隧道衬砌厚度量化识别技术[J].铁道建筑,2024,64(11):107-112.

[3]MAAS C,SCHMALZL J. Using Pattern Recognition to Automatically Localize Reflection Hyperbolas in Data from Ground Penetrating Radar[J]. Computers and Geosciences,2013,58:116-125.

[4]TERRASSE G,NICOLAS J M,TROUVéE,et al. Automatic Localization of Gas Pipes From GPR Imagery[C]//IEEE.201624th European Signal Processing Conference(EUSIPCO).Budapest:IEEE,2016:2395-2399.

[5]高新闻,王龙坤.基于深度学习的隧道病害图像检测[J].计算机测量与控制,2022,30(2):58-64.

[6]沙爱民,蔡若楠,高杰,等.基于级联卷积神经网络的公路路基病害识别[J].长安大学学报(自然科学版),2019,39(2):1-9.

[7]贺丰收,何友,刘准钆,等.卷积神经网络在雷达自动目标识别中的研究进展[J].电子与信息学报,2020,42(1):119-131.

[8]冯德山,杨子龙.基于深度学习的隧道衬砌结构物探地雷达图像自动识别[J].地球物理学进展,2020,35(4):1552-1556.

[9]丁屹霖.基于探地雷达的隧道内部典型缺陷仿真与识别技术研究[D].北京:北京交通大学,2023.

[10]张永恒,徐海泽,王耀东,等.基于探地雷达的隧道衬砌空洞检测方法[J].铁道建筑,2024,64(2):98-101.

基本信息:

中图分类号:U456

引用信息:

[1]林元铖,奉建军,周斌,等.基于GAR-Net模型的隧道衬砌厚度及钢拱架识别定位[J].铁道建筑().

基金信息:

中国中铁股份有限公司科技研究开发计划(2021-专项-08,2023-专项-07,2023-专项-05)

发布时间:

2025-04-30

出版时间:

2025-04-30

网络发布时间:

2025-04-30

检 索 高级检索

引用

GB/T 7714-2015 格式引文
MLA格式引文
APA格式引文