体积约束的稀疏NMF高光谱解混  

Hyperspectral unmixing of sparse non-negative matrix factorization based on volume constraints

作　　者：王伞[1] 韩月 王立国[1] WANG San;HAN Yue;WANG Liguo(College of Information and Communication,Harbin Engineering University,Harbin 150001,China)

机构地区：[1]哈尔滨工程大学信息与通信工程学院

出　　处：《哈尔滨工程大学学报》2019年第12期2077-2082,共6页Journal of Harbin Engineering University

基　　金：国家自然科学基金项目(61675051)

摘　　要：为了解决单纯非负矩阵分解计算繁复,收敛速度慢的问题,本文提出了一种基于自然梯度下降的体积最小及丰度稀疏约束的非负矩阵分解方法。该方法在目标函数中加入体积最小和丰度稀疏约束,可以对混合图像进行较好地分解;采用自然梯度下降的方法进行迭代,加快了算法收敛速度。实验结果表明:该方法能有效克服最小体积约束非负矩阵分解法速度慢且不稀疏的缺陷,相对于解混效果(SAD)相近的方法提速100倍,相对于解混时间相近的算法,此方法的解混精度提高0.02°;此方法尤其适用于像元较多的高光谱图像。To solve the problem of the complex computation and slow convergence of simple non-negative matrix factorization(NMF),in this paper,we propose an NMF method based on natural gradient descent for minimum volume and abundance sparsity constraints(the natural-gradient-based minimum volume and sparseness constraint NMF,NG-MVSC-NMF).We add the minimum volume and sparseness constraints to the objective function,which can decompose the mixed image well.This NG-MVSC-NMF method is adopted to speed up the convergence of the algorithm.The experimental results show that the proposed method effectively overcomes the shortcomings of the minimum-volume-constraint NMF method,which is slow and not sparse.Compared with a method with a similar unmixing performance(SAD),the speed is increased by two orders of magnitude.Compared with an algorithm with a similar unmixing time,the mixing accuracy of this method realizes an improvement of 0.02°.This method is especially suitable for hyperspectral images with many pixels.

关 键 词：高光谱图像 线性光谱混合模型 非负矩阵分解 体积最小 丰度稀疏 自然梯度 端元提取 光谱解混 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]