Research On Separation And Recognition Method Of Aircraft Target Micro-Motion Signal Based On Deep Learning

Airborne aircraft targets can be divided into helicopters,propellers,and jets according to their working principles and uses.The three airborne target radar echo signals mainly include the body component and the Micro-motion component of the target’s rotating component.Separating the Micro-motion components in the aircraft target radar echo signals and extracting and identifying the micro-motion features are widely concerned by academic and industrial circles at home and abroad,and have important application value in both military and civilian fields.For aircraft target recognition,the traditional feature extraction method requires more manual operations,which is time-consuming and easy to cause errors due to the randomness of manual operation.As a machine learning method,deep learning can realize the automatic extraction of target features,greatly improve the accuracy of target feature extraction,and achieve more accurate recognition of targets.For the three types of aircraft,this paper mainly focuses on the mathematical modeling analysis of the micro-motion signals of the rotating components,the separation and extraction of the micro-motion signals and the target recognition of according to the micro-motion features based on the deep learning method.The main work of the thesis includes the following aspects:1.First,the principle of the aircraft target Micro-motion is studied,the aircraft target micro-motion signal model and the corresponding mathematical expression are established.Based on the signal modeling,the micro-motion characteristics such as the maximum Doppler frequency,the number of single-sided spectrum,the period and the angular frequency of rotation of the aircraft target under different motion states are analyzed in detail.Through simulation experiments,the echo signals and micro-motion characteristics of three types of aircraft target rotating parts are analyzed from time domain,frequency domain and time-frequency domain,and the correctness of the model is verified.2.Second,The method of aircraft fretting signal separation based on deep learning network is studied.It mainly includes Deeplabv3 model based on deep learning semantic segmentation.It can learn the attribute labels corresponding to each pixel of the three types of aircraft target time-frequency map from a large number of aircraft target time-frequency signals,and accordingly,the time-frequency mask of the micro-motion signal of the rotating component can be generated.According to the time-frequency diagram after the mask,the inverse short-time Fourier transform can be used to recover the micro-motion signal of the target rotating component of the aircraft,and the micro-motion signal of the rotating component is extracted.The effectiveness of the aircraft fretting signal separation method based on deep learning network is verified by the separation experiments of three types of aircraft signals.3.The aircraft target recognition method based on deep learning network and micro-Doppler characteristics is studied.Based on the existing deep convolutional neural network,an aircraft target recognition method based on deep network and time-frequency map of micro-motion is proposed.According to the time-frequency characteristics and network performance of the aircraft target,the corresponding network structure is designed.The training data set is built based on the established aircraft signal model.The recognition performance under different observation conditions is analyzed through simulation experiments.