Research On UAV Radio Signal Identification Technology Based On Machine Learning

With the gradual maturity of UAV technology,the application of UAV in military,industrial and social life has been expanding.Civil unmanned aerial vehicles provide convenience in search and rescue,aerial photography and other application fields,but also bring various problems.In sensitive or crowded scenes such as military stations,government departments,airports,chemical industry and other major industrial institutions,large-scale events,there is no doubt that "black flying" UAVs have security risks,and it is necessary to strengthen the control of civil unmanned aerial vehicles.Radio detection and countermeasure is an important means of management and control of low-speed and small-scale civil unmanned aerial vehicles,which plays a vital role in the industry.However,there are still many technical problems to be solved in finding the "black flying" UAV signal from the surrounding complex radio environment.In order to find the "black flying" UAV from the complex radio environment,it is necessary to analyze and identify the radio signals in the complex electromagnetic environment.Thesis puts forward the process of classifying radio signals in complex electromagnetic environment,and screening samples from the classified signals for UAV signal classification and identification.The main research contents are as follows:(1)The Dirichlet mixed model is selected to classify the signals in the unknown complex radio environment without prior knowledge such as the number of signal types:firstly,the radio signals in the complex electromagnetic environment are preprocessed to facilitate the subsequent extraction of feature vectors to form a radio signal feature vector database;Using Dirichlet mixed model to process the data of feature vector library,complete the classification of radio signals and form a classification database;According to the stratified random sampling method,the samples are sampled from the classification database by class,and the signals obtained by sampling draw the signal waterfall diagram and the signal bispectrum analysis diagram to generate the sample database to be processed.(2)Develop UAV signal classification recognition program and process the classified sample data: develop UAV recognition program by using residual neural network model commonly used in image classification combined with self-calibration convolution and attention mechanism;Collecting radio signals of standard unmanned aerial vehicle,drawing waterfall diagram and bispectrum analysis diagram of radio signals,establishing a sample library as a training set and a verification set,and training the identification program;After training,the program is used for UAV radio signal recognition,and it is continuously trained and learned during the recognition process.