Acoustic Signal Recognition Of Construction Equipments Based On Statistical Features

Underground pipeline network provides essential conditions for normal city life,as it is an important infrastructure working for water supply,drainage,energy transmission,electric power supply and telecommunication.However,with the advancement of urbanization process,construction activities are continuously increasing in the city,which leads to serious damages and threatens to the underground pipelines by external earthmoving devices.The security precautions and protections from external destructions are hard to be achieved because the pipelines are widely distributed and buried under the earth.Thus,designing a round-clock and intelligent surveillance system with advanced technology is urgently needed.After field investigations we found that the destroying accidents are often caused by destructive equipments including electric hammer,hydraulic hammer,excavator and cutting machine.This paper proposed a novel acoustic feature extraction method with statistical parameters based on the study on working process of excavation devices and the analysis of characteristics of their acoustic signals,then proposed a recognition algorithm using statistical features combined with support vector machine(SVM)and extreme learning machine(ELM)as intelligent classifiers.The proposed recognition algorithm is able to effectively detect and recognize destructive devices threatening the safety of underground pipelines,and used in the surveillance system it can be intelligent and useful for pipelines maintenance.The main work and achievements of this paper are as follows:1.Microphone array was used in this paper to collect acoustic signals to constitute sample database and test signal set for experiments,and the acoustics collection were conducted in actual construction field under different environment and from different distances.2.This paper did research on Mel-frequency Cepstral Coefficients(MFCC)and Linear Prediction Cepstral Coefficients(LPCC)and based on their shortage and limitation on environmental acoustic recognition proposed a statistical feature extraction method,which describes signals and calculates parameters from both time and frequency domain.And also the related thresholds values were set according to distribution range of the signal parameters.3.SVM and ELM algorithms were studied and adopted as intelligent classifiers to establish training model and realize classification correctly.4.To validate the contribution and discuss the influence of statistical parameters from each time or frequency domain,experiments are conducted respectively using each set of parameters as feature vector to examine the recognition accuracy.5.Comparison experiments on recognition performance were also designed and conducted respectively using statistical feature,MFCC and LPCC features.Both SVM and ELM were adopted as classifiers and validated the reliability of proposed feature extraction method.Compared to the MFCC and LPCC,the statistical feature shows the advantages on recognition accuracy.Besides,it also achieves better performance in application for signals from long distance and jammed with Gaussian white noise.