Research On Parallel Processing Methods Of Polarimetric Radar Data With Many Cores

With the development of Polarimetric Synthetic Aperture Radar(PolSAR)technology,the application fields of polarimetric radar data are more extensive,and major breakthroughs have been made in the processing of polarized data.Applications such as parameter extraction achieve better results.However,high-precision polarimetric data is also accompanied by a large amount of data,and the amount of polarimetric radar data reaches at least the GB level.The increase in the amount of data leads to the complexity and redundancy of the calculation process,which directly affects the computation efficiency of polarized radar data,Hence,the efficiency of a large amount of polarized data processing has also become a major problem in polarized radar data processing.To deal with the problem,this dissertation proposes a parallel processing of many cores for polarized radar data processing based on graphics processing unit(GPU).This method makes full use of GPU high-performance many-core parallel computing function to optimize the processing of polarized radar data,so as to achieve the purpose of efficiently processing polarized radar data.This dissertation studies the entire processing chain of polarized radar data,including both qualitative and quantitative aspects,and mainly analyzes several classic qualitative and quantitative algorithms.In qualitative processing,a pipelined algorithm is used for research.From the processing steps of filtering to classification,each one classic algorithm is selected for parallel analysis.In quantitative processing,for the same polarization radar data,four classic forward scattering models and their corresponding inversion algorithms are analyzed.Different algorithms have different requirements on polarized radar data,application and computing performance.Specifically,this dissertation mainly improves the computational efficiency of several processing procedures of polarized radar data through GPU high concurrent threads.According to the differences of algorithm processes,various optimization strategies are used to improve the processing efficiency of polarized radar data,such as parallel task allocation,instruction level,data storage,GPU internal data transmission,and data transmission between CPU and GPU.The experimental results also fully confirm the superiority of GPU in computationally intensive data processing,in which the computational efficiency of different polarization radar data processing algorithms is one to two orders of magnitude higher than the single-core computing efficiency of CPU.The improvement of the whole-process calculation performance of polarized radar data greatly increases the real-time capability of the data,making the study of polarized radar data of broader significance.