Research On Optimization Of Airborne Radar Data Processing Method Based On Deep Learning Network

Maneuvering target tracking and multi-target data association are always hot research directions in airborne radar data processing field.However,the complexity of battlefield environment and target not only leads to the uncertainty of movement state of the tracked target,but also the uncertainty of target measurement,which makes the traditional tracking filtering algorithm and multi-target data association algorithm difficult to effectively deal with the target tracking task in the complex battlefield environment.Deep learning networks are data-driven and do not need to assume prior information such as clutter density,noise covariance and target motion model in advance,which is expected to provide a new solution for maneuvering target tracking and multi-target data association in complex battlefield environments.Therefore,the optimization of airborne radar data processing methods based on deep learning network is carried out in this thesis to improve tracking filtering performance under target maneuvering state and multi-target data association performance under clutter environment.The main research work is summarized as follows:1.Aiming at the problem of large error in tracking maneuvering target by traditional tracking filtering algorithm,a target state estimation(SE)algorithm based on long shortterm memory(LSTM)network(SE-LSTM)is proposed.Meanwhile,a second-order total variation loss function is designed to make the tracking track smoother.The experimental results show that the proposed algorithm can estimate the target state effectively,and has smaller tracking error and better tracking robustness compared with the classical tracking filtering algorithm and the mnemonic Kalman filtering algorithm based on deep learning network.2.Traditional data association algorithms need to estimate prior information in advance,such as clutter density and target motion model,etc.However,it is difficult to accurately estimate these prior information in actual scenes.To solve this problem,a data association algorithm based on deep learning network is proposed.The concept of virtual measurement is introduced to reconstruct the data association model for the possibility of false alarm and missing detection of radar detected targets.On this basis,a data association(DA)network(Transformer-DA)based on Transformer is proposed to solve the association problem between multiple targets and multiple measurements,and a loss function combining masked cross entropy loss and dice loss(MCD)is designed to train the proposed network.The experimental results of simulation and real measurement data show that the proposed algorithm can effectively solve the multi-target data association problem under different detection probabilities,and has smaller tracking error than the classical data association algorithm and the algorithm based on bidirectional long shortterm memory network.3.A track initiation algorithm based on bidirectional long short-term memory(BiLSTM)network is designed.On this basis,an optimization method of airborne radar data processing based on deep learning network is proposed,including three modules of track management,data association and tracking filtering.The simulation results show that the proposed method can effectively solve the multi-target tracking problem with unknown number of targets in complex environment,and has smaller tracking error compared with the classical multi-target tracking algorithm.