Spectral angle automatic cluster routine (SAALT): an unsupervised multispectral clustering algorithm

摘要

The Spectral Angle Automatic cLuster rouTine (SAALT) algorithm consists of an iterative spectral angle calculator which seeks to cluster scenes captured with multispectral and hyperspectral imaging instruments. The unique aspect of SAALT is its ability to operate with little or no a priori information about the scene. SAALT has been applied to hyperspectral data that spans the visible and near infrared (IR) and to multispectral data that spans the visible, shortwave IR, and midwave IR. Both actual and simulated scenes were used in this study. The results demonstrate the capability of SAALT to divide a scene into its natural components, such as water, clouds, grass, trees, and roads. The utility of SAALT described in this paper is demonstrated with quick and successful differentiation between cloudy and clear pixels during day, night, dawn, and sunset scenes for a hypothetical multispectral remote sensing system.