马尔可夫随机场的空间相关模型在非负矩阵分解线性解混中的应用

发布时间：2018-05-16 01:38

本文选题：非负矩阵分解 + 高光谱线性解混　；参考：《计算机应用》2017年12期

【摘要】：针对基于非负矩阵分解(NMF)的高光谱解混存在的初始化与"局部极小"等问题,提出一种基于马尔可夫随机场(MRF)的空间相关约束NMF线性解混算法(MRF-NMF)。首先,通过基于最小误差的高光谱信号识别(Hy Sime)法估算端元数量,同时利用顶点成分分析(VCA)和全约束最小二乘法(FCLS)初始化端元矩阵与丰度矩阵;其次,利用MRF模型建立描述地物空间分布规律的能量函数,以此描述地物分布的空间相关特征;最后,将基于MRF的空间相关约束函数与NMF标准目标函数以交替迭代的形式参与解混,得出高光谱数据的端元信息与丰度分解结果。理论分析和真实数据实验结果表明,在高光谱数据空间相关程度较低的情况下,相比最小体积约束的NMF(MVC-NMF)、分段平滑和稀疏约束的NMF(PSNMFSC)和交互投影子梯度非负矩阵分解(APS-NMF)三种参考算法,所提算法的端元分解精度仍分别提高了7.82%、12.4%和10.1%,其丰度分解精度仍分别提高了8.34%、12.6%和9.87%。MRF-NMF能够弥补NMF对于空间相关特征描述能力的不足,减小解混结果中地物的空间能量分布误差。
[Abstract]:Aiming at the problems of initialization and "local minima" of hyperspectral demultiplexing based on nonnegative matrix factorization (NMF), a spatial correlation constrained NMF linear demultiplexing algorithm based on Markov random field (MRF) is proposed. Firstly, the number of endelements is estimated by the minimum error based hyperspectral signal recognition method, and the end element matrix and abundance matrix are initialized by using the vertex component analysis (VCA) and the fully constrained least squares method (FCLS); secondly, the end element matrix and the abundance matrix are initialized by using the vertex component analysis (VCA) and the fully constrained least squares method (FCLS). The energy function describing the spatial distribution of ground objects is established by using the MRF model to describe the spatial correlation characteristics of the distribution of ground objects. Finally, the spatial correlation constraint function based on MRF and the standard objective function of NMF are decomposed in the form of alternating iteration. The end element information and abundance decomposition results of hyperspectral data are obtained. The theoretical analysis and the experimental results of real data show that the spatial correlation of hyperspectral data is low. Compared with the minimum volume constrained NMFN MVC-NMFS, piecewise smooth and sparse constrained NMFPSNMFSCs and the cross projection subgradient nonnegative matrix factorization (APS-NMFS), three reference algorithms are proposed. The resolution accuracy of the proposed algorithm is still increased by 7.82% 12.4% and 10.1%, respectively, and the accuracy of abundance decomposition is still increased by 8.34% and 12.6%, respectively, and 9.87%.MRF-NMF can make up for the deficiency of NMF's ability to describe spatially related features, and reduce the spatial energy distribution error of ground objects in the results of unmixing.
【作者单位】：南阳理工学院计算机与信息工程学院;
【分类号】：TP751

【相似文献】