Recognition Research Of Construction Equipment Based On Extreme Learning Machine And Convolution Neural Network

With the development of urbanization,the underground pipeline system is called the lifeblood of the city.It plays an important role in urban life for energy transmission,communication transmission,power supply,water supply and so on.However,with the urban construction,the accidents occur frequently due to damage of underground pipeline in the construction process.After media reports and on-the-spot investigation,we found that the treat of underground pipeline come from large construction equipment including cutting machine,electric hammer,excavator and hydraulic hammer.In order to protect of the underground pipeline,it is necessary to design intelligent surveillance system aimed at construction equipment.Analyzed the strengths and weaknesses of the different methods for recognition of construction equipment,the recognition method which is based on acoustic signal for recognition of construction equipment is proposed.In this paper,the acoustic signal features of construction equipment in time-frequency domain are studied.And then the deep learning method is adopted to extract acoustic signal features and classify for the construction equipment,so as to realize the prevention and protection for underground pipeline.The main work and achievements of this paper are as follows:1.In the traditional acoustic recognition methods,we studied that the popular acoustic feature extraction algorithms which are Mel-Frequency Cepstral Coefficients(MFCC)and Perception Linear Predication Cepstral Coefficients(PLPCC).On the basis of these two algorithms,the Mel-Frequency Perception Linear Predication Cepstral Coefficients(MF-PLPCC)is proposed.2.In the traditional methods experiments,the support vector machine(SVM)and Extreme Learning Machine(ELM)are used as the classifier.The comparative tests which are feature extract algorithms of MFCC,PLPCC and MF-PLPCC combine with classifier of SVM and ELM are designed.Compare and analyses the test results,it is concluded that the MF-PLPCC feature and RELM is more effective than other methods in the recognition rate and robustness for recognition of construction equipment.3.Deep Convolutional Neural Network(CNN)is proposed for recognition construction equipment.First,the FBank spectrums of construction equipment and environment noise are extracted.Then CNN is adopted to extract deeper acoustic features so that achieve better recognition rate.In the CNN network structure,the parameters are optimized including the number of convolution kernel,the size of input window,Dropout probability and the slope of the negative semi axis in Leaky Re LU activation function.Then compared traditional acoustic recognition methods which are MFCC+ELM,MFCC+SVM,FBank+ELM.The experiments show that the method in this paper has great advantages in recognition rate and robustness for recognition of construction equipment.Finally,Analysis of the traditional methods for mistaken recognize other acoustic categories,it is concluded that MFCC feature of construction equipment and environment noise are similarity so that lead to the phenomenon of false recognition.