Research On Method And System Of Drone Detection And Localization

With the rapid development of various technologies in recent years,UAVs have been widely used in military and civilian use,and can be easily obtained,resulting in frequent occurrences of UAV "black flying" incidents,which has brought tremendous impact to lowaltitude security.Therefore,it is necessary to detect and locate UAVs in important unattended areas.Radar technology is costly and is not suitable for small UAV detection and positioning.Radio frequency technology and image detection technology also have their own limitations,while acoustic detection and positioning technology has the advantages of low cost,all-weather surveillance and passive detection,Therefore,this article takes the quadrotor UAV as the research object,and studies the methods and systems for detecting and positioning UAVs using sound signals.First,use deep learning to process the sound signals collected by the microphone to detect whether there is a drone around.After the drone is detected,use the microphone array to perform three-dimensional positioning of the drone based on deep learning,and use the traditional sound source localization algorithm to detect the presence of drones.Acoustic imaging is realized by human-machine visualization.The detailed research content is as follows:(1)Analyze the characteristics of the UAV’s sound signal and the dynamic time-frequency characteristics,and determine that the low-frequency aerodynamic noise is mainly used for the sound detection and positioning of the UAV,and the frequency band is 160～340Hz.And through time-frequency analysis of the drone sound samples collected in advance,timefrequency masking technology is used to separate the drone’s sound from the background noise,reducing the influence of noise on the detection and positioning effect.(2)Establishment of UAV detection and positioning data set.The drone detection data set is used to train the depth model of the detection task,including drone sound samples and nondrone sound samples;the drone positioning data set contains the multi-channel sound signals collected by the microphone array and the current location of the drone Location information,including semi-synthetic signal data set and real signal data set.(3)Research on UAV acoustic detection method and system based on deep learning.First extract the features of the sound signal,extract the Log-Mel sound spectrum and the MFCC(Mel Frequency Cepstrum Coefficient)features,and input the two features as the input of the CNN(Convolutional Neural Network)network into the network to obtain the softmax value;The two sets of softmax values are fused at the result level using DS(Dempster-Shafer)evidence theory,and the final detection result is obtained.The results show that the use of DS theory fusion can combine the advantages of the two features,increase the detection accuracy to 97.5%,and achieve effective detection within 50m.(4)Establishment of UAV positioning system based on traditional sound source positioning.The UAV positioning system based on the MUSIC(Multipule Signal Classification)algorithm is mainly composed of an 8-element MEMS microphone array,an NI 9234 data acquisition card and a camera.The sound source can be imaged by a software platform written in C#,but it is affected by reverberation.The impact of other noises is great,and it is not suitable for UAV positioning tasks;the UAV positioning system based on the SRP-PHAT(Steered Response Power-Phase Transform)algorithm includes a 6-element microphone array and a camera.The human-machine positioning and acoustic imaging can realize the effective positioning of the drone,and the positioning error is about 100.(5)The research of acoustic location method of UAV Based on deep learning.The 6-array microphone array synchronously collects the acoustic signal of UAV,and calculates GCC Phat characteristics for multichannel signals,which is used as input of crnn network.The two-stage training method is proposed.First,the semi synthetic UAV acoustic signal is used to train crnn network,and then the real UAV sound signal is used to fine tune.The results show that the positioning error of crnn is about 3-4m,which is smaller than that of CNN,which indicates that the long-time information of RNN plays a significant role in UAV positioning;Compared with the traditional SRP Phat method,crnn model has the advantages of small error and short running time,and can locate UAV effectively and in real time.