Development of a supervised classifier for subpixel remote sensing

A new supervised classifier is developed in this research. It is designed and implemented according to the V-I-S model. The V-I-S model is a conceptual model to represent urban environments as a linear combination of vegetation, impervious surface, and soil, three basic land cover types in urban areas. It is proposed to study urban morphology from ground component composition. As a step further, six ground components are selected and defined in this research as basic urban ground components (two for vegetation, three for impervious surface, and one for soil). The end product is a six-channel image, with each channel indicating percentages of one ground component. The accuracy assessment is based on the comparisons between the predicted component percentages and the surveyed component percentages from visual interpretation of aerial photographs. Correlation coefficients are reported as index of accuracy for each component. This supervised classifier is tested by a TM image of partial Salt Lake City area. The result is considered acceptable. According to the correlation coefficients, there are strong relationship between the predicted and the surveyed percentages for two of six components, moderate relationship between predicted and surveyed percentages for three of six components, weak relationship between predicted and surveyed percentages for one of six components.