The evaluation of fuzzy membership of land cover classes in the suburban zone

Abstract The reflectance values of pixels, recorded by remote sensors, are often generated by more than one ground phenomenon. This, the so-called mixed pixel problem, has always been a property of scanner-type imaging, but its effect on the image classification process is arguably still a major problem to deriving accurate land cover maps, in spite of the increasing spatial resolution of sensors. This paper explores use of a fuzzy classifier to determine the constituent land cover components of pixels in a suburban environment. The classifier derives a measure of the fuzzy membership of a pixel belonging to each land cover class. For some land cover types including water, wetland, and woodland, a high correlation is shown between the fuzzy membership values for a pixel and the portion of the area of that pixel which belongs to a particular land cover type. The correlation for other land cover types is statistically significant but qualitatively poorer, and may indicate a lack of signature purity. The results show that the fuzzy classifier may enable the extraction of information about individual pixels and about subpixel phenomena not addressed by other classifiers.