Parallel implementation of N-FINDR algorithm for hyperspectral imagery on hybrid multiple-core CPU and GPU parallel platform
Spectral unmixing in hyperspectral remote sensing image has been widely researched in the last two decades. N-FINDR algorithm is one of the most classical and commonly-used endmember extraction algorithms. Nevertheless, it is a timeconsuming task that cannot meet the time requirement of many applications. In order to make N-FINDR computationally feasible, we consider parallel implementation of N-FINDR algorithm on hybrid multiple-core CPU and GPU parallel platform. First, a multi-core CPU-based parallel N-FINDR algorithm is considered based on a modified N-FINDR with two improvements. And by using the increasing programmability and parallelism of commodity GPU, a GPU-based parallel N-FINDR is presented. Finally, by taking advantages of the capability of the aforementioned algorithms, a hybrid multiple-core CPU and GPU parallel N-FINDR is proposed by using a virtual thread technique and an adaptive algorithm in which the computational load can be adaptively adjusted according to the capability of CPU and GPU. In experiment, our proposed parallel N-FINDR algorithms improved the accuracy of the original N-FINDR algorithm, and most importantly, they greatly improved the performance of N-FINDR algorithm.
Hyperspectral remote sensing endmember extraction N-FINDR parallel computing
Wenfei Luo
School of Geographical Science, South China Normal University, Guangdong Guangzhou, China 510631
国际会议
桂林
英文
1-7
2011-11-01（万方平台首次上网日期，不代表论文的发表时间）