Use of Ancillary Data in Object Based Classification of High Resolution Satellite Data

Abstract

The purpose of this paper is to establish the benefit in utilizing object based image classification in combination with ancillary data such as a digital line graph to classify land cover types associated with urban land use versus object based image classification that relies solely on the native image. Specifically, this paper will focus on the classification of roads from multi-spectral IKONOS satellite imagery that has a spatial resolution of 4 meters using the object base classification software eCognition Professional 4.0 by Definiens Imaging Co. and a rasterized Digital Line Graph (DLG) from the United States Geological Survey (USGS). The hypothesis presented in this paper is that the USGS' DLGs can be used to improve object-based classification of high-resolution satellite imagery by improving the creation of image objects in eCognition. A mask of pixels that were known to be representing physical roads in the raw image was used to assess accuracy. The two classified images were then compared with the mask and the accuracies were calculated using the TTA Mask Error method. The Overall Accuracy of the classification of roads from the IKONOS multi-spectral image rose from 0.7989 in the unclassified image to 0.9578. This is an increase of 0.1589 or nearly 15.1%, and is a substantial increase of accuracy that would lead us to the conclusion that the inclusion of ancillary data in the process of object based image classification would increase the classification accuracy of roads.