GPU-based Analysis Of Electromagnetic Scattering Characteristics Of Targets And Feature Extracting Methods

The evaluation of electromagnetic scattering by targets,especially the calculation of radar cross section(RCS)has important significance for national defense construction and civilian application domains.The analysis on the electromagnetic scattering characteristics is of widespread use in the field of national defense and military,such as radar imaging,target classification and identification,the design of stealth and anti-stealth,et al.,and it is widely applied into antenna design,remote sensing and other civilian areas.How to calculate target RCS efficiently and accurately has been the hot research topic of the computational electromagnetics.Physical Optics(PO)algorithm and Shooting and Bounding Rays(SBR)method belong to high-frequency asymptotic methods,which have simple principle and are simple to implement in computation of electrically large complex objects.However,traditional PO and SBR algorithms demand occlusion judgment of each facet during implementation,resulting in the low operation speed.With improved octree and Z-buffer of computer graphics,this paper improves the traditional PO algorithm and achieves the algorithm in parallel efficiently by GPU.Also,another improvement made to the rapid procedure of ray tracing of SBR algorithm by use of KD-tree structure in computer graphics.By the analysis of target micro-motion parameters,and the extraction of corresponding character parameters offers the important support information for the following objectclassification and identification.Therefore,this article makes studies on the mobility model of cone target based on precession and analyzes statistical property,physics characteristic and texture features of the time-frequency pattern from micro-motion target.In addition,research on cone target classification using Support Vector Machine(SVM).The main work of this thesis is as follows:1.First,this issue implements the traditional PO and SBR algorithm,and realizes them by using C++ programming language.The RCS of simple satellite and vertebral model are calculated and the effectiveness of the PO method has been tested.According to the shortcomings of the comparably time-consuming raytracing of traditional SBR,the threaded KD-tree is used to improve the efficiency.KD-tree method is discretized according the spatial distribution and the raytracing is traversed by the tree structure,thus the algorithm reduces the time of the invalid interference-examination,which greatly improved the calculation efficiency of SBR algorithm.The RCS of dihedral angle is calculated by using the KD-tree SBR algorithm,and compared with the fast multipole algorithm of FEKO,verifying the correctness of the method.2.Owing to the high computational complexity of traditional PO method while working with electrically large complex objects,a GPU accelerated PO algorithm is designed and implemented.And the octree and Z-buffer method were employed to accelerate occlusion process.Since the GPU hardware has large amount of elements,and Z-buffer method to determine each surface element in a relatively independent way during implementation and allocate a thread for each sufrace,thus computation procedure of RCS and the occlusion indentification is accelerated.And the data update way on account of lock-free structure is implemented within the realization in the course of acceleration of GPU-based PO algorithm.Compared with lock data updating mode,the lock structure avoided deadlock due to the tread contention.At last,combination the PO algorithm of traditional CPU and GPU,the feasibility and speed enhancing effect of PO algorithm based on the GPU accelerating is verified by numerical examples of three-wave frequencies complex satellite model.3.The mobility model of cone target based on precession is studied,decribing a extraction method of precession frequency,and the procession frequency of target is extracted in twodimensional time-frequency image.The principle of radar range profiles is studied and the signature of object scale is extracted combined with range profile of the cone target.The paper investigates the statistical property,physics characteristic and texture feature on the basis of two-dimensional time-frequency image of the micro-motion cone target.Based on the extracted features,the classification experiments to three types of cone target are done by making use of SVM,and the classification performance is analyzed under different noise interference conditions.4.The graphical electromagnetic computing and imaging software is developed based on the research outcome of the thesis,which has several functions,such as electromagnetic computing,visualization of 3D model,solution of target angle,electromagnetic imaging and so on.And the software is illuminated and validated with concrete sample.