Research On Target Classification Method Based On Radar Low Resolution Track Information

In recent years,Unmanned Aerial Vehicles(UAVs)disturbance occurs from time to time,bird strikes are also major threats to civil aviation flights.Both UAVs and birds are low altitude,slow speed small(LSS)targets.Compared with common radar targets,LSS targets have smaller radar scattering cross section(RCS),lower flying altitude,slower speed,and less Doppler frequency shift.In addition,LSS targets are located in a complex environment with more background interference.When detecting LSS targets,vehicles,aircraft and wakes in the environment will also form tracks,which will interfere with the monitoring and processing of specific LSS targets.So,it is urgent to classify the output tracks of low-slow small target detection radar to extract specific LSS targets.However,the existing LSS targets detection radar usually has low resolution,which can not provide the high-resolution information required by traditional target recognition and classification algorithms.Therefore,this thesis attempts to preliminarily classify targets based on low resolution track information of LSS targets detection radar.This thesis mainly studies the classification of LSS targets detection radar in airport environment.In low-resolution track information,it is difficult to classify targets intuitively in high-dimensional space composed of space-Doppler-reflectivity-sum and difference information.This thesis makes full use of the high-dimensional information of the track in the low-resolution track,takes the data of the track information distribution with time as the input of the classification algorithm,and condenses the track characteristics by artificial intelligence classification algorithm,so as to complete the classification of targets.The main work is as follows:1)Collect the measured track data,make statistical analysis of the information of the track points,and get the distribution of different targets in space,Doppler,reflective intensity and sum-difference information.Based on these distribution characteristics,the data set for training classification algorithm is simulated and constructed.2)CNN and RNN classification algorithms are introduced to solve the problem of low-resolution target track classification.The simulation results verify the effectiveness of the classification algorithm,and compare the performance of different classification algorithms.3)In view of the advantages and disadvantages of CNN and RNN,the convolution layer of CNN is replaced by the convolution window constructed by bidirectional recurrent structure.The recurrent convolution neural network is constructed based on the TensorFlow deep learning framework.The performance of the classification algorithm based on recurrent convolution neural network is verified by simulation.4)According to the characteristics of track,the input is divided into two parts according to its importance.Each part is convoluted with convolution cores of different sizes.Finally,the convolution feature map is fused to enhance the feature and improve the robustness.Based on the TensorFlow framework,a multi-structure convolutional neural network is constructed,and the performance is verified.