Research On Image Inversion Parallel Algorithm For Synthetic Aperture Microwave Radiometer

Developed from real aperture to synthesis aperture,the key paramemters of microwave radiometer,such as resolution,detection distance,obtain a certain elevation.The enhancement is based on the signal processing complexity greatly increased.In G-matrix inversion method,G-matrix size can reach 20 GB.The general computer can't calculate the brightness temperature image in effective time.In this paper,we use parallel computing to accelerate this process,which may lays the foundation for real-time imaging.And,we choose MPI and CUDA,namely the message passing parallel model and the GPU model,to parallelize the image inversion algorithm.The former runs on the Tianhe 2 supercomputing platform,through the high-speed Internet computer cluster to speed up the calculation;the latter run on NVIDIA's GPU,make full use of computing core to accelerate the calculation.There are many kinds of image inversion algorithms for synthetic aperture microwave radiometer,which are mainly divided into Fourier transform and G-matrix inversion.We select three typical image inversion methods,namely array factor synthesis method,conjugate gradient method,Tikhonov regularization method,representing Fourier inversion method,G-Matrix numerical iteration method,G-matrix inversion method.Firstly,we introduce the calculating principle of the algorithms,then analyze the parallelism,propose the design idea,implementing the parallel program with the two framework.Finally,the parallel program of the implementation of different models run on the emulational data,verifying the correctness of the computation result,and computing the acceleration ratio and extensibility.Comparing with the serial C++ and MATLAB code running time,the target of accelerating has achieced.However,there is still a lot of progress in the parallel algorithm,such as optimizing the storage structure and enhancing the portability of the program.