Research On Detection And Recognition Technology For UAV Signal Based On Deep Learning

In recent years,with the rapid development of Unmanned Aerial Vehicle(UAV)technology,the incidents of illegal flights of UAV pose a serious threat to low-altitude security.Therefore,it is necessary to study the technologies for supervising UAVs.A relatively low-cost detection mean is a radio spectrum monitoring mean using a passive receiver to intercept a signal transmitted by a UAV,namely a passive detection mean.However,the electromagnetic environment is more and more complicated,and various electromagnetic environment noise,various industrial and communication electromagnetic interferences have a great influence on passive detection.In addition,the modulation method of the UAV image transmission signal is more and more complicated,and it poses a new challenge to modulation recognition.It is not only necessary to realize the traditional close-set identification of modulation mode,but also to solve the problem of open-set identification of modulation mode.In order to realize remote detection,it should be adaptive to low signal-to-noise ratio(SNR).Therefore,the research on the detection and recognition of UAV signals mainly focuses on the detection of UAV signals,the identification of UAV types and the open-set identification of UAV signal modulation modes.The main contents of the paper are as follows:(1)The detection of UAV signals is studied based on faster regional convolutional neural networks.Based on the faster regional convolutional neural network,the characteristics of the time-frequency diagram,obtained by the short-time Fourier transform of the UAV signal,are learned.The distribution characteristics of the labeled UAV signal and the background environment signal are learned.Finally,the detection of the UAV signal in the time-frequency diagram is realized,and the detection performance is still good under the condition of low SNR.(2)The type identification of UAV is studied based on Alex Net.When the measured signals of UAVs are used for UAV type identification,the neural network is designed based on Alex Net.The network model can independently learn the spectrum characteristics of the measured signals of various UAVs,and realize the identification of UAV types.And it has very good adaptability to low SNR.In order to reduce the calculation amount and improve the recognition speed,the designed fully connected network realizes the UAV model identification by using the drone sampling signal data,and still has good recognition performance at low SNR.(3)The open-set identification of the UAV signal modulation mode is studied based on the reconstruction discriminant network(RDN).When the modulation mode of UAV signal is known modulation mode,this paper solves the close-set recognition of modulation mode based on neural network algorithm and discusses the influence of SNR on close-set recognition.When the mode is unknown modulation mode,this paper proposes a reconstruction discriminant network model based on the Autoencoder and the Generative Adversarial Networks.At the same time,the model can also accurately recognize the known modulation mode and still has good recognition performance in the condition of low SNR.