Investigation Of Composite Electromagnetic Scattering From Targets On Sea Surface And Sar Imaging With Recognition Based On Sbr

In this dissertation,the shooting and bouncing ray(SBR)algorithm is optimized at first.Then the composite electromagnetic scattering model of electrically large targets and complex sea environment is constructed based on the improved SBR algorithm.In addition,the electromagnetic scattering characteristics of the composite model for sea surface with targets is analyzed.Furthermore,the SAR imaging algorithm based on electromagnetic scattering calculation is investigated,and the SAR simulation data set for several typical ship targets is established.Combined with the deep learning theory,the recognition and classification of the ship SAR image simulation data is completed.The main work of this paper is as follows:1.The basic principle of bin hidden elimination of Open Graphics Library(Open GL)as well as neighborhood search algorithm is introduced in detail.On this basis,an improved SBR algorithm,which combined Open GL with neighborhood search algorithm,is proposed.This algorithm optimizes the search problem of ray passing through the corner and edge line of tree structure as well as the judgment problem of ray passing through tree structure.Meanwhile,it can effectively reduce the code complexity and improve the computational efficiency of SBR algorithm.2.The accuracy of ray tracing is of great importance for the computational accuracy of SBR algorithm.In order to improve the accuracy of SBR algorithm effectively,the improved SBR algorithm based on two-scale subdivision technique is proposed in this dissertation.In this algorithm,the large-scale facet is utilized to fit the geometric contour of the target to reduce the number of ray occlusion judgment.While the subdivided small-scale facet is used for bidirectional tracking to improve the accuracy of ray tracing.Compared with the traditional SBR algorithm,the SBR algorithm based on two-scale subdivision technique can improve the calculation accuracy without decreasing the calculation efficiency.3.In order to further improve the computational efficiency of SBR algorithm,the GPU parallel acceleration algorithm based on CUDA architecture is investigated.Firstly,the running mode and data storage mode of CUDA Programming are depicted.Secondly,the calculation process and optimization process of parallel SBR algorithm based on CUDA are given in specifics.Finally,the electromagnetic scattering for different ship targets is computed,where the simulation results show that the parallel SBR algorithm can effectively reduce the simulation time and improve the computational efficiency.4.To meet the demands of fast prediction for the composite scattering of electrically large targets and sea surface,the proposed improved SBR algorithm is combined with the twoscale model of sea surface to construct the composite scattering model of single target and multi-target above the sea surface.Furthermore,the composite scattering echoes of sea surface and single/multi-target is simulated with different sea state parameters,different radar parameters and different target parameters.Meanwhile,the influence of each parameter on the composite scattering characteristics is analyzed.5.The basic principle of SAR imaging and the range flow of Range Doppler(RD)imaging algorithm are presented.According to the principle of SAR imaging,the frequency-domain scattering field of composite model of ship and sea surface is simulated by using the established composite electromagnetic scattering model of target and sea surface.The SAR echo data can be obtained by utlizing the frequency domain pulse coherence method.Then the ship SAR image is generated based on the RD imaging algorithm.Additionally,the conversion relationship between the echo data obtained by polar format algorithm and electromagnetic scattering calculation is clearly given.Finally,the SAR images of the composite scene with different sea sate parameters and radar parameters are simulated.6.The basic knowledge of tensorflow framework and the basic structure of deep convolution neural network are presented in detail.The SAR image data sets of six kinds of ship targets are generated by simulation method.According to the basic structure of VGG-16 network,a deep convolution neural network is built,and the network is trained by using the simulated SAR image data.The influence of different activation functions,different learning rates,different optimization methods and different dropout selection ratio on network training and recognition accuracy is researched.Meanwhile,the influence of SAR dataset with different sea conditions on network generalization is examined.