Home

Technology Semi-Automatic Terrain Segmentation (TERSEG) The ability to reliably decompose a hyperspectral image into endmember classes is a critical component in many hyperspectral data processing scenarios. To address this problem Data Fusion Corporation (DFC) has developed a semi-automatic, physics-based, method to segment material types in hyperspectral imagery that is based on a recursive principle component analysis coupled to a matched filter. The method is advantageous because: It can be used for anomaly detection, TERSEG, and classifying shade pixels. It does not require any ground truth It does not require any a priori information in the form of spectral endmembers as does Spectral Mixture Analysis (SMA) One does not have to specify the multitude of parameters (number of classes, change thresholds, class standard deviation, class distance, number of merged pairs, etc.) that are needed to run ISODATA and K-means algorithms It allows for mixed pixel classification It has a physical interpretation The principal upon which DFC’s proposed method is based is that the highest energy spectra in a hyperspectral data cube can be identified by the first eigen-image obtained from a principal component analysis. We assume that this pixel forms an endmember for a given spectral class which can be used as a target in a spectral detection problem. We also assume that by recursively removing pixels which belong to a given spectral class, the lower energy modes will begin to dominate the spectral variability and will themselves be members of their own spectral class which can then be used to further decompose an image. See Composite Images Back to Hyperspectral Data Fusion Corporation 10190 Bannock St., Suite 246 Northglenn, C) - 80260 Phone: 720.872.2145 x 114