摘要
针对后向投影算法运算量大和实时性差的问题,该文在图形处理器的并行处理架构下,结合时域快速后向投影算法,提出了面向像素的并行快速后向投影算法。该方法距离向根据分辨率要求进行划分并分配并行线程,方位向聚焦则通过多级子孔径合并与图像分裂技术来降低运算量。通过比较该文算法、全局后向投影算法和距离多普勒算法的理论计算量、成像时间和加速比可以看出,该文算法的运算速度比后向投影算法大大提高。三种算法针对实测数据成像时间与成像效果的对比验证了该文算法的有效性和工程实用价值。
In view of that the back projection(BP) algorithm's huge computatxonal cost ana poor real time performance hinder its applications, the fast back projection algorithm is combined with the parallel architectural of graphics processing unit (GPU) to present the pixel-oriented paralleled fast back projections (PFBP)algorithm. The range direction is divided and paralleled threads are distributed depending on range resolution. In azimuth direction, the technique of sub-aperture merging and image splitting is used to deduce the computational cost. The comparison of theoretical calculation and acceleration ratio among the pixel-oriented PFBP, the range Doppler (RD) and the global back projection (GBP)algorithm show that the acceleration ratio is gained greatly. Experimental results of three algorithms of the data collected by SAR system illustrate the validity of pixel-oriented PFBP algorithm and its value in engineering.
出处
《南京理工大学学报》
EI
CAS
CSCD
北大核心
2014年第5期651-657,共7页
Journal of Nanjing University of Science and Technology
基金
中国航天科技集团航天科技创新基金(9140A07010713BQ02025)
国家自然科学基金(61302188
61471198)
关键词
合成孔径雷达
图形处理器
面向像素
并行快速后向投影算法
synthetic aperture radar
graphics processing unit
pixel-oriented
paralleled fast backprojection algorithm