A Fast Spectral Matching Algorithm for Larger-Scale Hyperspectral Data:Spectral Angle Sensitive Forest

We propose a spectral matching algorithm,spectral angle sensitive forest(SASF),which improves the spectral matching efficiency in high dimensional large-scale hyperspectral dataset.The locality sensitive hashing(LSH) is expanded to the metric space of spectral angle.Moreover,we introduce a new scheme to index the data bucket,which remove the flaw of the original LSH method that a part of the query points won't get any neighbors.We provide systematical analysis of the parameters,and theoretical and experimental evaluation of the algorithm.The computational efficiency of SASF is proved outperforming the former algorithms.And SASF also provides a tradeoff between efficiency and precision of spectral matching by which make the user has more choices in the applications.