Research On The Method Of Aircraft Acoustic Signal Type Recognition In A Non-cooperative Communications Enviroment

At present,in order to identify the type of the intercepted aircraft short-wave radio communication audio,the work is mainly to identify the different sound of different aircraft engines in the sound signal through manual listening,and then infer the type of aircraft.However,this method of recognition through manual monitoring has great difficulties in practical application.First,the sound signal can be intercepted is very short.Second,the intercepted sound signals in the aircraft cabin are mixed with various types of noise,which is difficult to identify.Based on the above two points,the traditional manual listening method often has a great error in the identification of aircraft type,and it is easy to cause physical and psychological damage to the listener.Therefore,the research on the identification of aircraft cabin sound and the background sound of the pilot’s call has very important practical significance.In this paper,the recognition of aircraft type in non-cooperative voice communication environment is studied according to the requirement that the correct recognition rate of aircraft type reaches over 85% by the partner.This paper mainly studies the recognition of aircraft types in non-cooperative voice communication environment.At present,researches in this direction mainly extract feature values based on wavelet packet energy,MFCC and other feature extraction algorithms,and classify feature vectors through BP neural network or SVM classifier to obtain classification accuracy for recognition.These recognition methods need to artificially select their internal functions and have certain conditions for the selection of features,which need a lot of experience to support.Aiming at the problems at present aircraft recognition,this paper studies the wavelet packet transform,higher order cumulant optimization three sound signal wavelet packet decomposition,MFCC feature extraction method and the simulation experiment was carried out respectively,and they respectively by the traditional BP neural network classifier-supervision and unsupervised learning since the encoder two classification methods were analyzed,and through the simulation experiment contrast since the encoder and the advantage of the traditional classification methods.This paper select the BP neural network and SVM and GA-SVM these three classification algorithm to extract four characteristic signal classification,for after many experimental results show,the input for the MFCC feature extraction of signal characteristic vector of the classification results of the encoder is a stable,its accuracy rate steady at around 95% float,input for high-order cumulant optimization eigenvector of wavelet package decomposition feature extraction of GA-SVM classification results are accurate and stable,multiple recognition accuracy of about 97%.Since the encoder to category 1 signal classification result is bad,still have optimization of space,so the signal feature extraction of category 1 results better wavelet packet decomposition and MFCC algorithm combined feature extraction,the joint feature as input vector of the encoder is optimized,get the optimized joint feature-since the encoder algorithm to classify the plane can get 97.74% classification accuracy,and the experiment many times its classification accuracy is very stable,more suitable for this article research aircraft cabin background signal recognition.