Incorporating intra-pixel reflectance variability in the multispectral classification process of high-resolution satellite imagery of urbanised areas

Publisher Website
Google Scholar
Abstract
Monitoring and planning of highly urbanised areas require the analysis of continuous streams of spatial data in a cost-effective way. Although successfully applied to rural areas, conventional multispectral classification is very prone to error when applied to urbanised areas, mainly because of the high object density of these scenes. The incorporation of local spatial structure may decrease the error-prone nature of multispectral classification. The forthcoming availability of very-high resolution satellites is a promising data source for this approach. In particular, the high resolution of the panchromatic bands, which are an order of magnitude of five higher than that of multispectral bands, enables the incorporation of local spatial structure in the form of intra-pixel reflectance variability. The aim of the present paper is to develop a method to incorporate the intra-pixel reflectance variability in the multispectral classification process and to investigate its suitability. Tests demonstrate the approach is worthwhile to be investigated more closely on real, high-resolution satellite imagery once they become available.

This publication has 1 reference indexed in Scilit: