Data Generation Method For Radar HRRP Recognition Database Based On Generative Adversarial Networks

High resolution range profile(HRRP)is the coherent summations of the complex time returns from target scatterers by the wideband radar.HRRP contains rich structural information of targets,which is of great significance for radar target recognition.The radar HRRP recognition database is mainly used to store the HRRP dataset for training the recognition model.Besides,the number of HRRP data in the database will directly affect the final target recognition performance.However,when the targets are non-cooperative ones,it is usually difficult to obtain sufficient HRRP data.Therefore,generating HRRP data of high quality to expand the recognition database can effectively solve the above problem.In recent years,Generative adversarial networks(GAN)is one of the common methods for data generation in the field of deep learning.It is widely used in various fields because of superior data generation performance.In order to improve the performance of radar target recognition,this thesis studies the method of generating HRRP data for radar HRRP recognition database based on GAN.The contents are as follows:(1)The basic principle of HRRP data generation for HRRP is introduced.Firstly,the sensitive problems of HRRP and the corresponding data preprocessing methods are introduced.Then we introduce two existing HRRP data generation methods and then analyze their problems.Finally,four HRRP data quality evaluation criteria are introduced in detail.(2)This part focuses on HRRP data generation method based on CWGAN-GP model.In order to solve the problems of the original GAN,such as the class of the generated data is unknown and the quality of the generated data is poor,CWGAN-GP model is introduced to realize the generation of HRRP data.This model introduces label information to make the class of the generated data known,adds Wasserstein distance to guide the training process of the model and adds gradient penalty to limit the gradient range of the discriminator.Finally,it is verified through experiments that the quality of HRRP data generated by the CWGANGP model is high,and the expand dataset can improve the recognition performance of the classifier.(3)This part focuses on HRRP data generation method based on CACGAN model.In order to further improve the separability of the data generated by the CWGAN-GP model,the CACGAN model is proposed.In order to improve the quality of the generated data,this model introduces class labels to the input of the discriminator,introduces Wasserstein distance,and adds a gradient penalty term.Besides,an auxiliary classifier is added to improve the separability of the generated data.Also,a weight coefficient is introduced to adjust the proportion of the quality and separability of the generated data.Finally,the results of the experiments show that the quality of the generated HRRP data is higher,and the separability of the generated HRRP data is also guaranteed.The expanded dataset can improve the recognition performance of the classifier more effectively.