应用稀疏非负矩阵分解聚类实现高光谱影像波段的优化选择

Sparse Nonnegative Matrix Factorization for Hyperspectral Optimal Band Selection

针对高光谱影像数据高维性、高度相关性和冗余性等特点,提出应用稀疏非负矩阵分解聚类实现高光谱影像波段的优化选择。通过稀疏非负矩阵分解方法对高光谱影像进行稀疏化表示,同时顾及其可聚类的特性,在保留所选波段物理意义的基础上,得到波段选择后的高光谱影像降维数据。通过该方法对PHI-3高光谱影像进行波段选择的试验分析,应用聚类特征有效性分析波段聚类结果,并采用波段子集的信息量、相关性和可分性3类评价指标来验证方法的效果。最终,从运行效率和分类精度两方面证明了基于无监督聚类的稀疏非负矩阵分解对高光谱影像的波段选择的实用性。

Hyperspectral data provides imagery with hundreds of spectral bands, but many of them contain redundant information. Band selection is often applied to reduce the dimensionality of the data. A new technique for band selection through sparse nonnegative matrix factorization （SNMF-BS） is proposed, which decomposes the image data into the multiplication of basic matrix and coefficient matrix. Using sparseNMF, hyperspectral images can be described as sparse representation. Since sparse NMF acts as co-clustering, it can be used for band clustering without considering the distance metric between different spectral bands. Through band clustering, subbands are selected to serve the need of dimensionality reduction, while preserving the physical meanings of the selected bands. The PHI-3 real hyperspectral dataset is experimented for band selection. Clustering validity indexes, KLC, band correlation and separability are used to do the evaluation. The experimental results show that sparse NMF provides considerable insight into the unsupervised clustering-based band selection problem and also offer the proper band combinations with spectral meanings. And the results also demonstrate that the proposed method outperforms other popular methods.