Research On Classification And Recognition Method Of Flying Target Based On Time-frequency Analysis And Deep Learning

Feature extraction and analysis of micro-Doppler in radar echo signals is an important method to identify flying targets.This paper aims at two kinds of flying targets,such as rotor drones and birds,by extracting the micro-motion features in the radar echo signal and combining with artificial intelligence technology to realize the flight target recognition.In this paper,the method of modeling and simulation is adopted to study and analyze,and the mathematical model of the micro motion of the rotor blade and the wing of the bird is established respectively,The time-frequency spectrum of micro Doppler produced by fretting is obtained and analyzed by three time-frequency analysis methods: STFT,WVD and stfrft,The selected time-frequency spectrum is taken as the sample of machine learning,and the difference feature of time-frequency spectrum is learned by the representation learning algorithm and then classified by convolution neural network.Through the combination of time-frequency analysis and deep learning method,the system research of flight target intelligent recognition and classification is realized.The conclusions are as follows:1)The mathematical model of the micro-motion of the UAV and the bird is esTab.lished and the micro-motion characteristics are analyzed to provide a basis for the subsequent acquisition of the signal set.First replace the flying targets with dots,and through the two processes of translation and rotation,the Doppler signal has the micro-motion component signal.Then,the two flying targets are modeled by the motion of the propeller blades and bird wings.Fretting echo signal.2)Selection of time-frequency spectrogram.Through the time-frequency analysis of the fretting echo signal through the three time-frequency methods of STFT,Wigner-ville and STFRFT,the comparison shows that the STFRFFT method is better,followed by the STFTmethod,and the Wigner-ville method is not applicable.3)Research on image representation extraction and convolutional neural network algorithm.First,preprocess the obtained time-spectrum map,select the sparse auto-encoder algorithm to extract the time-spectrum map features,and use convolutional neural network to realize classification recognition.The simulation result is: the recognition rate of the two-blade propeller UAV reaches 97.32%.The recognition rate of the three-blade propeller drone reached 97.01%,and the recognition rate of birds reached 95.23%.