Research On Audio Based Low Altitude Unmanned Aerial Vehicle Detection And Positioning System

Unmanned Aerial Vehicles(UAVs)are becoming more and more common in our lives.While UAVs bring us convenience,they also bring us more and more hidden dangers in terms of security and privacy.The demand for UAVs surveillance is growing.The research of UAVs supervision technology is a hot research issue at present.The audio detection method has been widely studied in academia and industry due to its advantages of simple operation and low cost.In view of the urgent need for supervision of UAVs,this paper builds a UAV audio data set based on the audio detection method.We analyzed the UAV audio signal characteristics and the audio characteristics under various flight attitudes in detail.This paper verifies the effectiveness of the UAV recognition method based on audio features,builds the ADS-CNN network model to recognize the UAV’s flight attitude,builds a microphone array and studies the audio localization algorithm based on TDOA,and establishes a UAV detection system.The main work of this paper is as follows:(1)This paper builds the UAV audio data set and feature library.We used a single microphone to record the audio information of the DJI MAVIC AIR2 flying in a quiet playground.The recorded UAV audio clips are 3 seconds long,for a total of 6300 seconds.During the recording process,each piece of audio is marked with a flight attitude label for use in the UAV flight attitude recognition task.Furthermore,to extend the UAV audio dataset,we downloaded the publicly available noise dataset FSD50 K to augment our dataset with a total of 12600 seconds after augmentation.We extracted the audio features of UAV audio signals and noise signals through MFCC and STFT and MFCC-STFT fusion methods,and built three audio feature libraries.(2)This paper verifies the effectiveness of the UAV recognition method based on audio features.Based on the UAV audio feature library,we use the SVM,GMM,and CNN models commonly used in existing UAV audio recognition tasks to identify UAV audio features.We experimentally verify that the audio recognition method is effective in drone recognition.In addition,the experiment also shows that the recognition effect of the CNN method is better than that of the SVM and GMM in the UAV audio recognition task.(3)This paper realizes UAV flight attitude recognition based on audio features.Based on the method of UAV recognition based on audio features,this paper hopes to obtain more useful information from UAV audio signals,so it proposes the research of UAV flight attitude based on audio features.Aiming at the problem of UAV flight attitude recognition,an audio recognition network ADS-CNN is proposed.First of all,we add the focus module of interest area in ADS-CNN to enable the network to quickly and accurately locate the interest area in the feature map,and improve the accuracy and efficiency of recognition;Secondly,the residual structure is designed to solve the problem of gradient disappearance caused by the increase of network depth;Finally,in view of the small number of samples in the UAV audio dataset recorded by ourselves,we can deepen the number of network layers as much as possible and effectively restrain over fitting by using the lightweight structure of depth separable convolution.Compared with other classical network models,the effectiveness of ADS-CNN in UAV flight attitude recognition is verified.(4)This paper improves the time delay estimation algorithm and improves the TDOA location accuracy based on audio signal.Aiming at the problem that the existing time delay estimation methods have large errors when the signal to noise ratio of UAV audio signal is low,this paper proposes an improved time delay estimation algorithm.Based on the generalized quadratic cross-correlation time delay estimation algorithm,this algorithm adds exponential operation to the cross-correlation transform function and inverse Fourier transform function,and introduces a new generalized weighting function.so that when the signal to noise ratio is low,the time delay estimation error of UAV audio signal is reduced.The experimental results show that the algorithm can improve the accuracy of UAV audio location.