Methods for classifying remotely sensed data from multiple data sources are considered. Special interest is in general methods for multisource classification and three such approaches are considered: Dempster-Shafer theory, fuzzy set theory and statistical multisource analysis. Statistical multisource analysis is investigated further. To apply this method successfully it is necessary to characterize the "reliability" of each data source. Separability measures and classification accuracy are used to measure the reliability. These reliability measures are then associated with reliability factors included in the statistical multisource analysis. Experimental results are given for the application of statistical multisource analysis to multispectral scanner data where different segments of the electromagnetic spectrum are treated as "different" sources. Finally, a discussion is included concerning future directions for investigating reliability measures.
Date of this Version