Mixture processing of remotely sensed multispectral scanner data involves estimating the percent coverage of individual crops or species contained within the instantaneous field of view of the scanner. In recent years, various mixture processing algorithms have been proposed to solve the so-called "mixture problem". All of the proposed algorithms require, as inputs, the spectral signatures of the various species observed. Often it is extremely difficult to obtain the required spectral signatures of individual species.
In this paper, two methods for obtaining the required spectral signatures for a particular mixture model are considered. For the model considered, the spectral signatures become signature vectors. The first method is based upon determination of the signature vectors such that a measure of the inconsistency between the mixture model and the observed data is minimized. The second method is based upon determination of the signature vectors such that the estimated mean percent coverage of individual species match apriori or ground truth estimates. The two methods proposed are applied to actual multispectral data in order to verify the concepts presented.
Date of this Version