高光谱遥感影像降维最大噪声分数变换(maximum noise fraction rotation,简称MNF rotation)方法运算量大,耗时长.基于多核CPU与众核MIC(many integrated cores)平台,研究MNF算法的并行方案和性能优化.通过热点分析,针对滤波、协方差矩阵运算和MNF变换等热点,提出相应并行方案和多种优化策略,量化分析优化效果,设计MKL(math kernel library)库函数实现方案并测评其性能;设计并实现基于多核CPU的C-MNF和基于CPU/MIC的M-MNF并行算法.实验结果显示,C-MNF算法在多核CPU取得的加速比为58.9~106.4,而基于CPU/MIC异构系统的M-MNF算法性能最好,加速比最高可达137倍.

Maximum noise fraction (MNF) rotation is a classical method of hyperspectral image dimensionality reduction, and it needs a large amount of calculation and thus is time-consuming. This paper investigates the code transplantation and performance optimization for the maximum noise fraction algorithm on multi-core CPU and many integrated core (MIC) architecture. By analyzing hotspots of the MNF algorithm, parallel schemes are first designed for filtering, covariance matrix calculating and MNF transforming. Then, a series of optimization methods are presented and validated for various parallel schemes of different hotspots, including using math kernel library (MKL) functions. Finally, a C-MNF algorithm on multi-cores CPUs and an M-MNF algorithm on the CPU/MIC heterogeneous system are constructed. Experiments show that the C-MNF algorithm achieves impressive speedups (ranging from 58.9 to 106.4), and the M-MNF parallel algorithm runs the fastest, reaching a maximum speed-up of 137X.

国家自然科学基金(61272146, 41375113)