Research And Application Of Audio Signal Concrete Pavement Classification Based On Deep Learning

In recent years,with the rapid development of economy,the construction of highway in our country has also developed rapidly.The cement concrete pavement is one of the main roads of highway traffic due to its high rigidity,strong diffusion load capacity and good stability.It has a large proportion in China's highway transportation system.It often threatened by pavement plate bottom,however,the bottom of the plate concrete pavement is one of the dangers of the most common concrete pavement,it has a serious impact on the safety of the transportation,so the road.the concrete pavement is necessary.This article mainly aims at traditional acoustic nondestructive tests with single microphone receives the audio signal when its anti-interference ability and low spatial resolution of faults is proposed using the microphone array receiving audio signals;The audio signal is collected through the microphone array and the wavelet is carried out.Using the theory of wideband signal,the acoustic source localization and beam forming of the collected signals are carried out.The audio signal is analyzed by wigner-ville and the effective features are extracted.By using the method of deep learning,this paper classifies the concrete pavement of cement concrete pavement and obtains the classification accuracy.At the same time,with the deep learning method for classification,this paper will use DBN network classification model and the network classification model,comparing the classification results and comparison are presented in the article the classification accuracy.Based on the measured data,this paper use the relevant software carried out on the processed data simulation,DBN classification using deep learning in MATLAB toolbox DeepLearnToolbox network model was carried out on the processed data simulation,and the CNN classification is the GPU computation under the condition of the Caffe platform of simulation experiment,the experimental results show that the use of deep learning approach for concrete road pavement when deciding its classification accuracy can reach 99.1%,11.6% higher than the traditional SVM classification accuracy,demonstrate the feasibility of this method can accurately determine the concrete road pavement.However,in the method of deep learning,the classification accuracy of DBN network model can reach 97.33%,and the classification accuracy of CNN network model can reach 99.1%,which shows that the classification accuracy of CNN network model is higher.