Drone Detection And Localization Using Acoustic Arrays:Algorithm Design And Implementation

With the rapid development of drone market,drones bring us a lot of convenience,but also lead to security and privacy threats.In recent years,the frequent occurrence of drone flying incidents have posed great threats to personal privacy,public security,aviation safety and national security.Therefore,the demand for drone surveillance has been growing,and the research on drone surveil-lance technology has been highly valued by all of the world.Audio-based surveillance method,as an effective method for drone detection and localization,has been widely studied by academia and industry.In this thesis,we firstly analyse drones,acoustic feature,then propose the drone detec-tion and localization algorithms and fina lly develop a drone surveillance system by using acoustic arrays.The ma:in content of this thesis can be summarized as follows:(1)A compressed spectrogram algorithm is propo sed to extract drones' acoustic feature,then the residual neural network is applied to recognize the feature,and finally a fusion detection algo-rithm is designed to improve the performance.In this thesis,the time-frequency method is used to analyze drones'acoustic signals.In order to solve the shortcomings of large dimension and feature sparse in the original spectrogram,the compressed sp.ectrogram reduces the frequency resolution and intercepts the effective frequency band.The fusion detection algorithm is based on Bayesian optimal decision and logistic regression model,and makes use of the classification results to get the final detection probability.(2)A direction of arrival(DOA)estimation algor:ithm based on drones' harmonic distribution features is proposed.Firstly,the algorithm uses the distribution feature of harmonics to estimate the center frequency.Then high-resolution narrow-band DOA estimation algorithm is applied to obtain the DOA of each harmonic.At last,the final DOA estimation result is obtained by weighting them together according to harmonics',energy distribution.This algorithm makes full use of the drones' acoustic feature and improves the performance of DOA estimation.(3)An acoustic source localization algorithm based on time difference of arrival(TDOA)is proposed.The algorithm makes full use of the redundant TDOA information.In order to estimate the initial value of Newton iterative solutions,this algorithm constructs some linear equations with small dimension,avoiding the nonlinear and sparse problems.Then a clustering algorithm is used to eliminate the abnormal results in these solutions.Finally,the iterative solutions are obtained by using the initial value(4)In the case of multipath effect and low signal-to-noise ratio,the Gaussian prior probabil-ity density distribution function(GPDF)is used to improve TDOA estimation.Multipath effect leads multiple spurious peaks when adopting generalized correlation function for TDOA estima-tion;meanwhile,the peak corresponding to the desired source will be drowned by the noise in low signal-to-noise ratio cases.This algorithm constructs a GPDF by using the TDOA information from last moment,and filters the generalized cross-correlation function to obtain a new cross-power spectral density function,which improves the TDOA estimation accuracy.