Paper No. 13
Presentation Time: 4:00 PM


ZHANG, Ling, Applied Science, university of Arkansas at little Rock, 2801 S. University Avenue, Little Rock, AR 72204, MCMILLAN, Margaret E., Department of Earth Science, University of Arkansas at Little Rock, 2801 S. University Avenue, Little Rock, AR 72204 and MILANOVA, Mariofanna, Computer Science, university of Arkansas at Little Rock, 2801 S. University Avenue, Little Rock, AR 72204,

We propose a novel, multi-feature fusion, object-based classification approach on the ESRI ArcGIS platform using very high resolution remote sensing imagery for the study area of Lake Maumelle watershed, Arkansas, USA. The Lake Maumelle watershed contains a reservoir that provides drinking water for more than 400,000 people in the Little Rock metropolitan area. The reservoir has a long history of excellent water quality. However, suburban and urban development is increasing within the watershed, prompting concerns that land use changes will impact water quality. The goal of this research is to build an easy to use tool to help water resource managers detect land use changes within the watershed. We suggest that the integration of GIS and Remote Sensing technologies on a single GIS platform could be a cost - effective approach in terms of time and money. The image classification method includes four steps. First, texture features of the image are extracted and analyzed, then composited into the image at next step. The original image is a color-infrared aerial imagery with one foot ground resolution acquired in 2009. Second, we use an ESRI ArcObjects Model based on a multiresolution statistical segmentation algorithm to generate regions. Third, training samples and test samples are generated using ESRI ArcMap and visual inspection of the image. Next, a supervised support vector machine (SVM) classifier is applied to the segmented multi-feature fusion and mulit–channels image. Fourth, classification accuracy is assessed with the test samples using a tool developed in ArcGIS platform. The experiment achieved a high overall accuracy of 94.44 with 0.91 Kappa value, which indicated an excellent probability that image pixels were correctly classified.