Research And Application Of Tunnel Construction Advanced Geological Prediction GPR Measured Data Based On CNN

As one of the important parts of highway,the tunnel is especially important in geological exploration of initial stage.How to accurately understand geological information,identify and classify is the focus of research.Ground penetrating radar(GPR)has become the main method for tunnel geological detection because of its non-destructive testing characteristics and fast detection speed.This paper mainly uses wavelet packet to detect the data of ground penetrating radar.And on the basis of the convolution neural network model lenet-5,two improved models are proposed to classify the images data after de-noising.In practical engineering applications,there are many spikes or non-stationary noise signals in GPR data.Traditional signal processing methods such as Fourier transform are not ideal for this type of signal.The wavelet transform overcome the shortcomings of traditional methods and has good effect on local time-frequency analysis,which can effectively de-noising the data of GPR.Based on the theory of wavelet transform,this paper studies multi-resolution analysis and wavelet packet analysis.Finally,wavelet packet is used to the experiment of GPR data de-noising,and obtain good de-noising effect.In this paper,the convolution neural network used to classify the ground penetrating radar data after de-noising.On the basis of the classical convolutional neural network model lenet-5,two improved convolutional neural network models,CNN-1 and CNN-2,were designed.The improved model classification accuracy can be verified by the experiment,which the data collected from 5,000 with six kinds of tunnels were used as data sets,and the classification experiments were conducted under the conditions of Caffe platform and GPU.The experimental results show that when the model is iterated 2000 times,the classification accuracy rate reached 95.6%,and the accuracy rate of CNN-2 is 98.2%.The classification accuracy is 4.3% and 6.9% higher than that model of lenet-5.Research results show that the improved convolution neural network models have higher classification accuracy,and the parallel convolution neural network model CNN-2 classification accuracy is the highest.This paper further improved model parameters to ensure the optimal performance of the model.