A SUBPIXEL CLASSIFIER FOR URBAN LAND-COVER MAPPING BASED ON A MAXIMUM-LIKELIHOOD APPROACH AND EXPERT SYSTEM RULES

Abstract

This paper provides a potential solution to the problem of mixed pixels (pixels containing more than one category of land cover type) in the analysis of satellite images of urban areas. A supervised classifier was developed to estimate ground component percentages from Landsat Thematic Mapper images of urban areas. Six ground components were selected according to the Vegetation-Impervious Surface-Soil model. Implementation of this classifier involved the Bayes algorithm to calculate initial percentages and expert system rules to iteratively adjust percentages according to a linear mixture model. The end product is a six-channel image in which each channel indicates percentages of a pre-defined ground component at the subpixel level. The resulting image displays information beyond typical per-pixel classification results. Pixels are still represented by six numbers, indicating the percentages of six pre-defined ground components. A regression analysis was performed to compare the estimated percentages to the surveyed percentages, which were derived from aerial photo interpretation. Correlation coefficients were reported as indices of accuracy for each ground component. The technique was applied to a 1990 Thematic Mapper image of the Salt Lake City area. The calculated indices of accuracy show a significant relationship between the estimated and surveyed percentages. Correlation coefficients between estimated and surveyed percentages are not very high overall, due to the generalization of specific land-cover types into six broad land-cover types, which results in extremes that are not handled very well when the model is applied to real world encounters.