The use of structural information for improving land-cover classification accuracies at the rural-urban fringe

Abstract

A methodology for incorporating structural information into conventional classification procedures is described. The technique is based on the use of an edge-density image which is generated using a Laplacian operator. This image is included in a Mahalanobis classifier as an additional band of data. The method is particularly designed for higher spatial resolution data in which plenty of spatial information is available. It has been tested using SPOT HRV multispectral data obtained over part of the rural-urban fringe of Metropolitan Toronto, Canada. Twelve land-cover types have been used to evaluate the approach and the classification results have been compared with those obtained by conventional maximum-likelihood classification. An overall accuracy of 86.1 percent was achieved by incorporating structural information into the classification compared with an accuracy of only 76.6 percent obtained without the structural information.