关键词: 关键词: 非负矩阵分解, 高光谱线性解混, 空间相关, 马尔科夫随机场, 交替迭代, 空间能量

Abstract: Aiming at the initialization and “local minima” problems of Non-negative Matrix Factorization(NMF) in hyperspectral unmixing, a spatial correlation constrained NMF linear unmixing algorithm(MRF-NMF) based on Markov Random Field(MRF) was proposed. Firstly, endmember’s quantity was estimated by hyperspectral signal identification by minimum error(HySime), the endmember matrix and abundance matrix were initialized by Vertex Component Analysis and Fully Constrained Least Squares(FCLS). Secondly, the spatial correlation distribution features of ground objects was depicted by building energy function based on Markov Random Field(MRF) model. Lastly, the adjacent pixels’ spatial correlation constraint was processed as standard NMF object function’s parallel and alternate step during each iteration of unmixing procedure. Theoretical analysis and real data experimental results showed that MRF-NMF can make up for NMF’s deficiency in depicting spatial correlation features, and reduce ground objects’ spatial energy distribution error. Comparing with the other three reference NMF unmixing algorithms, even as experimental hyperspectral data has a relatively low level of spatial correlation, MRF-NMF’s endmember decomposition precision is still increased by 7.82%, 12.4% and 10.1%, and the abundance decomposition precision is still increased by 8.34%, 12.6% and 10.1%.

Key words: Keywords: Non-negative Matrix Factorization(NMF), hyperspectral linear unmixing, spatial correlation, Markov Random Field, alternative iteration, energy function