Quantitative Identification Technology of Tunnel Lining Thickness in Ground Penetrating Radar Images Based on Improved U-Net Network
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
维普
万方数据
传统地质雷达图谱隧道衬砌厚度识别采用射线追踪法与人工判识,自动化程度不高.本文分析了隧道衬砌地质雷达图谱特征,制作地质雷达图谱衬砌厚度数据集,利用数据增强技术实现样本扩容,基于改进的U-Net语义分割目标识别网络对模型进行训练与评估.改进的U-Net网络对衬砌厚度自动识别的平均交并比(mean Intersection Over Union,mIOU)和平均像素准确率(mean Pixel Accuracy,mPA)分别为94.2%与96.7%,且检测效率为原始U-Net网络的3.75倍.提出了一种模板匹配算法,对自动识别的衬砌厚度雷达图谱实际坐标系进行重建,建立像素坐标与实际坐标间的映射关系,完成隧道地质雷达图谱衬砌实际里程与厚度的量化识别,并实现了衬砌欠厚信息的快速统计与输出,具有一定的工程应用价值.
The traditional ground radar spectrum tunnel lining thickness recognition used ray tracing method and manual identification,and the degree of automation was not high.This article analyzed the characteristics of tunnel lining geological radar spectra,created a geological radar spectrum lining thickness dataset,used data augmentation to achieve sample expansion,trained and evaluated the model based on an improved U-Net semantic segmentation target recognition network.The results show that,the improved U-Net network automatically recognizes the lining thickness with mIOU(mean Intersection Over Union)and mPA(mean Pixel Accuracy)values of 94.2%and 96.7%,respectively,and the detection efficiency is 3.75 times that of the original U-Net network.This article proposes a template matching algorithm to reconstruct the actual coordinate system of the radar map after automatic recognition of lining thickness,establishes the mapping relationship between pixel coordinates and actual coordinates,and finally achieves quantitative recognition of the actual mileage and thickness of the tunnel geological radar map lining.It also realizes the rapid statistics and output of under thickness information,which has certain engineering practical value.
维普
万方数据
