Research On Radar Clutter Simulation Technology Based On Deep Learning

Environmental clutter seriously affects radar performance.Modeling and simulation of radar clutter is an important part of radar development.Fidelity,flexibility,and versatility are the three important performance indicators of clutter simulation technology.The traditional method of determining the statistical model through real data and then performing clutter simulation not only loses fidelity,but also has low versatility.Therefore,It is necessary to propose new high-performance simulation techniques to cover the insufficients of traditional methods.Aiming at the problem of radar clutter generation simulaton,this thesis adopts a deep learning method based on generative adversarial network and the problem of insufficient fidelity of traditional radar clutter simulation technology is solved.Simulation verification shows that the clutter generated by the deep learning method is closer to the real clutter data than the traditional simulation method.the method is more versatile.The research content of the full thesis is summarized as follows:(1)Analyze the generation mechanism of surface clutter,explain the common clutter power spectrum model and amplitude distribution model,and give the waveform diagram of each statistical model.The method of deep generative model is studied,and the pros and cons of each method are analyzed to clarify the reasons for choosing GAN.(2)In-depth study of clutter simulation methods based on statistical models,this thesis describes the basic frameworks of the two clutter simulation methods,The simulation principle of the four kinds of clutter distribution is analyzed by the ZMNL method,and the simulation is implemented.The estimated parameters of the distribution model are calculated from real data,and then the most suitable distribution model is judged by fitting verification,and finally the clutter is simulated by the clutter simulation method of the distribution model.Such a simulation method solves the problems caused by empiricism and will make the simulated clutter closer to the clutter in the real environment.(3)A clutter simulation method based on WaveGAN is proposed.This thesis designs the processing flow of training data to provide input for clutter model training;optimizes the WaveGAN model and adds batch normalization operations to its generator;Compared with the clutter simulation method based on statistical model,the clutter simulation method based on WaveGAN performs better on the MMD evaluation index,which shows that the clutter data generated by this method is more realistic.At the same time,the application of this method is not limited to a specific scene environment,making its versatility stronger.(4)A conditional control clutter simulation method based on Info-WaveGAN is proposed.Based on the WaveGAN model,this thesis designs a conditional control method with info GAN added to solve the conditional control problem of the clutter generation model;Using one-shot vector coding classification to represent different sea state clutter,conditional training generates 2?4 level sea state clutter.Using MMD indicators to evaluate the generated clutter of different sea conditions,the conditional control clutter simulation method based on Info-WaveGAN is better than the traditional method for the simulation of different sea state clutter.