Spectral Angle Sensitive Forest: A solution for fast hyperspectral matching

This paper proposed a type of data structure, Spectral Angle Sensitive Forest (SASF), which was designed for indexing and matching Hyperspectral data with spectral angle metric at low computational cost. In this paper, we theoretically and experimentally proved that this new method outperformed the traditional data structure (such as Vantage Point Tree) used for high dimensional dataset, and overcome the problem of Locality Sensitive Hashing algorithm that the query data could not get matching results in certain probability. By adjusting a few parameters, SASF is convenient for users to choose between matching speed and matching accuracy in the applications.