基于多核CPU和GPU的高光谱数据并行几何校正  

Hyperspectral data parallel geometry correction based on multicore CPU and GPU

作　　者：李娜[1] 白勇[1] 赵慧洁[1] 徐秋[1] 

机构地区：[1]北京航空航天大学仪器科学与光电工程学院精密光机电一体化技术教育部重点实验室,北京100191

出　　处：《现代电子技术》2013年第2期110-112,共3页Modern Electronics Technique

基　　金：国家“863”计划(2008AA12121102;2009AA12Z119);国家自然基金项目(61008047;61177008);中国地质调查局项目(1212011120227);长江学者和创新团队发展计划(IRT0705)

摘　　要：针对高光谱几何校正计算复杂,大数据量频繁传输降低处理效率,无法满足实时需求等问题,提出基于多核CPU和GPU的并行计算模型。实现基于GPU的并行几何校正,并引入流水线并行思想提出基于多线程的数据读写优化方法,实现重采样部分的数据I/O优化。应用航空推扫成像仪所得高光谱数据进行实验,验证该方法能够有效地隐藏部分硬盘与内存间的数据I/O时间,几何校正加速比达到4.03,在基于GPU的并行计算基础上提高了1.74倍。Aiming at the quick processing of geometry correction,a parallel computation model based on multicore CPU and GPU is proposed in consideration of the complex computation and frequent IO operation during geometry correction.GPU based parallel geometry correction was implemented.By introducing the technique of stream parallel,an optimized method of data IO is proposed to realize the data IO optimization of re-sampling section.The experiment of airborne hyperspectral data collected by push-broom hyperspectral imager was executed.The result shows that the geometry correction processing after optimization can effectively hide the data IO time of partial hardware and internal memory.The speedup ratio of geometry correction reaches 4.03,incresed by 1.74 times of GPU based parallel computation.

关 键 词：高光谱数据 几何校正 并行计算 多核CPU 

分 类 号：TN911.34[电子电信—通信与信息系统] TP79-34[电子电信—信息与通信工程]