Research On Acoustic Identification Technology Of UAV Based On Deep Learning

The development of aerial UAV technology has provided great convenience for people’s production and life,but at the same time,the black flight phenomenon of UAV is becoming more and more serious,so the development of UAV detection technology has become particularly important.Timely detection and type identification of black flight UAV is the key research content in the field of UAV detection.In this paper,aiming at the characteristics of large amount of radiated noise produced by uav in flight,the acoustic recognition technology of UAV based on deep learning is studied.As for the identification of UAV,this paper divides it into two parts: the detection of UAV and the type identification of UAV.In unmanned aerial vehicle(UAV)discriminant solution,combining automatic encoder and threshold detection theory,through the training to achieve the automatic encoder radiated noise MEL cepstrum features for unmanned aerial vehicle(MFCC)reconstruction,and to reconstruct the data between the input data and the mean square error(MSE)as a designated loss detection threshold,obtained in laboratory tests of high detection probability and low false alarm probability.In the solution of UAV type recognition,two deep learning models,convolutional neural network and cyclic neural network,are used for UAV type recognition.In the research of UAV type recognition using convolutional neural network,the time-frequency characteristics of UAV sound signal were extracted by short-time Fourier transform and wavelet transform,and the time-frequency feature tensor of UAV sound signal was enhanced by deep convolutional generative countermeasure network(DCGAN).In the scheme of applying cyclic neural network for UAV type recognition,the MEL cepstrum coefficient of UAV acoustic signal is taken as the input of the cyclic neural network,and the accuracy of UAV type recognition of RNN,LSTM and bi-lstm are compared respectively.