Research On Aircraft Target Classification Method Based On Physical Drive And Data Driven Features

The micro-Doppler modulation effect of the target is unique to other targets,including the details of the target micro-motion component geometry and micro-motion,which provides a new way for radar automatic target recognition.The structure of the micro-moving components of the three types of aircraft targets(helicopter,propeller,jet),such as rotor length,speed,and number,are significantly different,resulting in different micro-Doppler modulation characteristics in narrow-band radar echoes.According to this,relevant features can be extracted from the echoes to realize the classification of the three types of aircraft targets.Common classification features include physical mechanism features that reflect the physical structural parameters of the target micro-motion component and data-driven features that do not have a clear physical meaning.In this paper,we study the two kinds of features and propose corresponding feature extraction methods.The content is summarized as follows: 1.In the actual working scene of the far-field of narrow-band radar,the coordinate system of the parameterized micro-motion component of the aircraft is established,and the echo model of the aircraft micro-motion component is derived.The validity of the model is verified by electromagnetic simulation software.The difference in micro-Doppler modulation characteristics of helicopters,propellers,and jets.2.Analyze the adaptation scenarios of the existing aircraft target rotor blade physical parameter extraction method,and point out that the high repetition frequency conditions required by it are often difficult to meet in practice.Aiming at this problem,a method for extracting the characteristics of rotor physical parameters under low-frequency conditions is proposed.The method firstly obtains the non-uniform sampling echo by means of repetitive frequency jitter;and then uses the short-term sparsity of the signal in the sliding window in the time-frequency analysis,and recovers the time spectrum under the unambiguous condition by means of the sparse reconstruction algorithm,and finally based on the time spectrum extracts the instantaneous Doppler curve and inverts the physical parameters of the rotor.The simulation results show that the proposed method can extract the physical characteristics of the aircraft rotor under the condition of low average repetition frequency.3.The traditional data-driven features are mostly manual and manual extraction.On the one hand,the versatility between radar data in different bands is poor.On the other hand,the feature extraction personnel also put forward higher requirements.Aiming at this problem,this paper studies the data-driven feature extraction method based on deep network.The deep network has powerful nonlinear feature extraction ability,and has achieved better performance than traditional manual features in the fields of image and voice.In this paper,the feature extraction methods based on convolutional neural network and deep confidence network are proposed respectively,and their performances with classical data driven features under various experimental conditions are compared.The results show that the features extracted based on deep networks are short.Resident and low signal to noise ratio conditions have better recognition capabilities.