Study On Intelligent Identification Of Quality Defects In Tunnel Lining Based On Deep Learning Model

In recent years,with the rapid development of transportation construction in the mountainous areas of western China,the scale of construction and operating mileage of China’s traffic tunnels rank first in the world.At the same time,tunnel development has moved from construction to equal emphasis on construction and maintenance,coupled with the quality defects and diseases of tunnels under construction and operation,making tunnel quality inspection and maintenance increasingly heavy.The geological radar method is an important means of tunnel lining quality detection,and the existing geological radar image interpretation mostly relies on manual experience to determine and identify,and the identification workload is large,the efficiency is low and the experience of the inspectors is greatly affected.By introducing deep learning intelligent algorithms and combining with geological radar method,the research on intelligent identification technology of tunnel quality defects is of great engineering practical significance and application value for improving tunnel detection efficiency and saving manpower and time costs.At present,most of the research on tunnel intelligent detection focuses on the recognition of tunnel lining apparent diseases or simulated radar images,and there are few studies on the application of deep learning to the interpretation of measured geological radar images.Therefore,based on geological radar detection theory and deep learning target detection algorithm,this paper carries out research on radar feature map and intelligent recognition of lining typical quality defects.The main research work and conclusions are as follows:(1)The basic principle of geological radar detection and the filtering method and process of geological radar image are elaborated.The theoretical framework of deep learning based on convolutional neural network model is introduced,and the feasibility of geological radar image recognition based on deep learning model is analyzed by combining the model transfer method.(2)The full-scale detection model of tunnel lining under three typical surrounding rock conditions: grade III.,grade IV.and grade V.was produced.The lining detection model is divided into a complete and defect-free lining model and a typical quality defect lining model with insufficient lining thickness and the cavity behind the lining,lining emptying,and insufficient number of secondary lining steel bars.Based on the analysis of geological radar detection data,the radar detection standard map of lining typical quality defects is established,and the radar image characteristics of typical quality defects are determined,so as to verify and assist the establishment of detection data samples.(3)Based on the actual detection data of geological radar in a large number of tunnel projects,the preprocessing processes such as gain,de-averaging and normalization of geological radar data are completed,and the data sample set of typical quality defects of tunnel lining based on real radar images is established.(4)Combined with the geological radar image dataset and the idea of transfer learning,an intelligent identification method for tunnel lining quality defects based on SSD model and YOLOv4 model is established.Firstly,the dataset is labeled and divided,and the training set and test set are divided.Then,the VOC data is used to complete the weight pre-training,and then the initial weight parameters and model are determined.On this basis,the training and recognition effect evaluation of the two types of deep learning models are carried out,and the applicability of the two types of algorithms in lining quality detection and recognition is comparatively analyzed.The results showed that:1)The deep learning algorithm has a significant effect in tunnel defect identification,and the accuracy rate of SSD algorithm is 86.58% and that of YOLOv4 algorithm is 86.05% in terms of void defects;In terms of rebar defects,the accuracy of the SSD algorithm reached 97.7%,and the accuracy of the YOLOv4 algorithm reached 98.18%.It shows that tunnel quality defect detection based on deep learning model has strong practicability.2)In terms of comprehensive performance,the m AP value of the SSD algorithm is 92.14%,and the m AP value of the YOLOv4 algorithm is 92.12%,which is almost the same,and both have strong generalization performance.The YOLOv4 algorithm is superior to the SSD algorithm in terms of training time and detection speed.The results show that the YOLOv4 algorithm has better applicability and application potential in the application of lining quality defect detection.