Research On Detection And Recognition Of UAV Based On Audio Characteristic

In the mode of passive acoustic detection,to identify unmanned aerial vehicle(UAV)type by using UAV acoustic signals,which is a new topic in passive acoustic target recognition field.On the basis of the research and analysis of the mechanism and characteristics of the representative UAV acoustic signals,the related research work on target detection and recognition system based on UAV acoustic signal is conducted.Firstly,According to the characteristics of UAV noise and environmental background noise,the signal preprocessing technique is applied to study the threshold de-noising algorithm based on wavelet transform.The algorithm filters out the noise signal whose wavelet coefficient is less than the given threshold to achieve the effect of noise reduction.The experimental results show that the noise reduction effect of wavelet global threshold is better than that of layered threshold.Secondly,two feature extraction techniques of UAV target acoustic signal are studied.According to the perspective of energy distribution,the feature based on EMD and IMF power ratio is extracted.Normalized the power ratio of IMF spectrum and origin signal spectrum,and making the power ratio as eigenvector of signals.The MFCC features are extracted from auditory perception and combined with the ?MFCC features with time-varying characteristics to represent the UAV features.Experiment shows that both eigenvectors can reflect the characteristics of UAV and make preparations for UAV type identification.Thirdly,VQ and SVM are designed for the UAV acoustic target classification.These two classifier and the features of four types of UAVs of real acoustic signals are used to do the experiments of target classification.Results show that the two feature extraction techniques canreflect the different characteristics of different types of UAVs.Finally,the PCA is used to reduce the dimensionality of eigenvectors.Experiments show that the dimensionality reduction can reduce the high-order redundancy of eigenvectors under the condition of ensuring good recognition performance,and reduce the computation time and space.The brief and multi-angle mixed features,which produce by the PCA,have batter classification performance than signal feature.