Research On Dynamic Electromagnetic Simulation And Feature Extraction For Space Targets

Inverse synthetic aperture radar(ISAR)plays an important role in space situational awareness and air defense.At present,how to obtain the shape and size of the space target from the high-resolution ISAR image,achieve function evaluation and feature extraction have become the key problem in the ISAR imaging and automatic target recognition communities.Due to the limited data resources and military background,it is difficult to obtain plenty measured data of space targets.According to the available orbit information and 3-D geometric model of the space target,this thesis studies a dynamic electromagnetic simulation method,and then constructs the database of electromagnetic echoes of four typical satellites,from which the ISAR images are generated.On this basis,it proposes a robust space target recognition method based on multi-channel spatial transform network,which effectively solves the problems of non-cooperative targets and unknown complex deformation during ISAR image formation.Particularly,the deformation may be induced mainly by different moving direction,bandwidth,coherent accumulation time,range and azimuth sampling rate,etc.Furthermore,it studies the ISAR dataset expansion methods based on Generative Adversarial Network(GAN)and Deep Convolutional GAN,which attempt to solve the problems of limited ISAR image samples during network training.The related work could provide theoretical and technical supports for improving the ability of space target detection and recognition for available ground-based radar.The main work of this thesis is summarized as follows:In the first part,the basic principle of target electromagnetic calculation was introduced.Then,principles of ISAR imaging and range migration correction are studied,which could lay solid foundation for dynamic electromagnetic simulation and ISAR image database construction of typical space targets.In the second part,a dynamic electromagnetic simulation method of space targets is proposed.Firstly,the method establishes the observation coordinate system of the space target,and then calculates the observation angles according to the orbit information.On this basis,the target dynamic electromagnetic echoes are calculated.Finally,the 2-D ISAR echo matrix is obtained.The method could avoid the large amount of calculation and data interpolation caused by taking dynamic echoes from the whole ehoes of the target,which are generted from all the possible observation angles.In the third part,a robust Pol-ISAR recognition method based on multi-channel spatial transform network is proposed,which fully exploits the rich target features of Pol-ISAR.In this method,the spatial transform network module automatically adjusts the image deformation of each channel,and the concatenation layer fuses the features of the three channels.Finally,automatic target recognition based on Pol-ISAR images is realized.Compared with the available target recognition algorithms based on deep convolutional neural networks,this method achieves higher recongnition accuracy for the deformed dataset.In the fourth part,we study data expansion methods based on GAN,which could generate more ISAR image samples from the available ones for feature extraction and recognition.Experiments have shown that the methods increase the recognition accuracy for the limited dataset.