基于Fan模型非负矩阵分解的光谱解混并行计算

摘要

The generalized non-negative matrix factorization algorithm based on Fan model is an effective nonlinear hyperspectral unmixing algorithm without the pure pixel assumption.First,the parallel optimization algorithm based on CUDA programming model and memory model was designed for the fast implementation of the generalized nonnegative matrix factorization algorithm based on FAN model.Then,task assignment and thread mapping were performed for the serial and parallel parts of the optimized algorithm and reasonable kernel functions were designed for the implementation of the key steps.The spectral unmixing experiment on real hyperspectral data shows that the CUDA parallel optimization algorithm can achieve higher speedup than the serial algorithm,which verifies the validity of the proposed algorithm.%基于FAN模型的广义非负矩阵分解是一种非纯像元假设下有效的高光谱图像非线性光谱解混算法.针对基于FAN模型的广义非负矩阵分解算法的快速实现问题,基于CUDA编程模型与存储器模型设计并行优化,对优化后算法的串行与并行部分进行任务分配与线程映射,设计合理的核函数实现各关键步骤.通过真实高光谱数据的光谱解混实验,结果表明CUDA并行优化后的算法相比串行算法,能达到较高的加速比,验证了其有效性.