Research On SAR Data Generation Related To Deep Learning And Radar Cross Section

Synthetic Aperture Radar(SAR)has a very wide range of applications in military reconnaissance and target recognition.Traditional SAR interpretation requires strong professional knowledge to support,which is time consuming and labor intensive.The SAR image itself will be affected by the radar bandwidth and the observation angle.It will cause great difficulty for SAR target recognition based on artificial intelligence.In order to obtain a higher recognition accuracy,it is necessary to have a large number of target SAR images under different conditions.Therefore,the research on SAR image generation is of great significance.The rapid development of artificial intelligence and deep learning provides a new idea for the research of SAR image generation.Artificial intelligence and convolutional neural networks(CNN)have a strong ability to learn the characteristics of targets and made great achievements in image processing.It is a promising research direction to design an efficient SAR image generation model based on artificial intelligence.The Generative Adversarial Networks is the hottest generation model in the field of artificial intelligence,and has shown excellent performance in the generation of ordinary optical images.The imaging mechanism of SAR images is very different from optical images.In order to improve the effect of the original Generative Adversarial Networks,the electromagnetic characteristics of the SAR image target will be used in this thesis to instruct the training of the neural network.Then combining the convolutional neural network(CNN)to design the conditional deep convolutional Generative Adversarial Networks.The research content of the this paper mainly includes the following aspects:(1)Studied the imaging mechanism and basic methods of synthetic aperture radar,analyzed the distributed model of radar target,proposed the construction method of geometric model,and gave the calculation method of Radar-Cross Section(RCS).(2)Studied the theory of deep learning and artificial intelligence,the structure of convolutional neural network;Analyzed the principle and application of convolutional layer,pooled layer and fully connected layer.Based on these study,the principle and structure of generative adversarial networks are discussed.(3)On the basis of the original Generative Adversarial Networks,the conditional generative adversarial networks and the deep convolutional generative adversarial networks are studied.Combining the advantages of these two network models,a conditional deep convolutional generative adversarial networks based on the target electromagnetic characteristics is designed.(4)Using the MSTAR dataset to train on the common DCGAN model and our improved C-DCGAN model,two sets of SAR images were generated,each of which included three targets--BMP2,TBR70 and T72.Then a convolutional neural network that can classify these three targets is designed to verify the effect of generating pictures.In this thesis,based on the requirements of SAR image recognition,the SAR image generation method based on deep learning and radar cross section is studied.It can provide a new idea for the application of artificial intelligence technology and SAR image research.