A detection algorithm for road feature extraction using EO-1 hyperspectral images

Abstract

The proposed method takes advantage of spectral information content in the hyperspectral images, EO-1 hyperion, to find road candidates. The assumption of this approach assumes that each pixel in the hyperspectral images is composed of a linear mixing of the reflectance from various components of the Earth's surface. The algorithm derived from the statistical and linear model is used to highlight the targets - road features and to suppress the background features of a scene. After road detection, all potential road candidates have been presented in the detected image. The objective is to avoid predicting uncertain road points and setting complex criteria to form the road network. Finally, the detected road segments will be traced and connected to generate skeletonized results. This proposed approach could be used to detect other linear features, such as a drainage network, from the hyperspectral imagery as well. In this way, the vectorized image could be produced to provide additional thematic layers for further spatial analysis in GIS applications.