Research On HRRP Generation Method Based On Generative Adversarial Networks

At this stage,the deep learning algorithm can be applied to radar automatic target recognition technology based on High Resolution Response Profile(HRRP),which can achieve better results than traditional methods,but most of these research methods or results are based on a large number of complete HRRP data.However,in practical application scenarios,radar data and its annotation information are difficult to obtain,especially for the non-cooperating target.Even if part of real data could be obtained,there might be a problem of poor data quality due to factors such as signal interference These problems have greatly limited the application of data-driven deep learning algorithms in HRRP target recognition.In view of the insufficient sample size and poor sample quality in radar application scenarios,this paper starts from the perspective of data itself and studies the generation method of radar HRRP.In order to expand the existing real dataset into a sample set sufficient to support various training methods,this paper constructs a two-stage generation model by introducing image translation theory and supervised information,and proposes a HRRP generation method based on generative adversarial networks.Rate distance image generation algorithm.The main research contents and innovations of this thesis include:Firstly,for the target recognition task of insufficient sample radar data,the research of radar HRRP generation algorithm based on generative adversarial networks is firstly carried out,which makes the generation model more controlled and the new data can be ideally generated.Secondly,based on the theory of image-to-image translation,the research on radar HRRP generation algorithm based on image-to-image translation is further developed,so that the generated samples have more style changes while retaining the original structure information,which enhances the sample diversity.Thirdly,based on the simulated plus unsupervised learning model,the radar HRRP generation algorithm based on the secondary refinement model is developed,which makes the generated data more realistic and more recognizable.Finally,the experiment is carried out under different real datasets and different sample conditions,and the quality of the generated data obtained by each model is evaluated by different evaluation indicators.The experimental results show that the two-stage generation model proposed in this paper can effectively expand.The experimental results show that the two-stage generation model proposed in this paper can effectively expand the existing real dataset,and to some extent alleviate a series of problems caused by insufficient samples in artificial intelligence radar research.