The Air Force makes extensive use of hyperspectral imaging and creates hyperspectral cubes shown at right with data recorded in three dimensions. The added variable of time increases the complexity.
However, the complexity may be decreased by noting that each pixel in the data cube has a set number of spectral components. By limiting the analysis to single pixels, the data complexity may be reduced to three (one pixel, color, and time). If the analysis is further limited to cases where there is only a single constituent in each pixel, it is not necessary to confront the problem of a mixture of items in each pixel.
The Air Force has further simplified the analysis of rich hyperspectral data. In this system, spectral and temporal data is received and formulated into a vector/matrix. At that point feature extraction is performed by any of the common methods including independent component analysis, non-linear prime component analysis, support vector decomposition, or combinations of these.
Feature extraction yields at least two of the largest principal components from which a cluster diagram is produced and subsequently, a distance is calculated. Defined clusters of significant distance indicate that the data came from two different sources. If these had been a hyperspectral measurement of crops in two different fields, then the clustering method would be useful for identification of the individual species.
- Takes complex, high dimensional data sets and assists in reducing them to manageable sizes by eliminating data redundancy
- Efficient analysis of electro-optical, hyperspectral, radar and other data wherein a large number of independent dimensions also include a time varying component
- US patent 8,861,855 available for license