•  
  •  
 

Abstract

Sensors for remote sensing have improved enormously over the past few years and now deliver high resolution multispectral data on an operational basis. Most Land-use/Land-cover (LULC) classifications of high spatial resolution imagery, however, still rely on basic image processing concepts (i.e., image classification using single pixel-based classifiers) developed in the 1970s. This study developed the methodology using an object-based classifier to characterize the LULC for the Buffalo River sub-basin and surrounding areas with a 0.81- hectare (2-acre) minimum mapping unit (MMU). Base imagery for the 11-county classification was orthorectified color-infrared aerial photographs taken from 2000 to 2002 with a one-meter spatial resolution. The object-based classification was conducted using Feature Analyst® , Imagine® , and ArcGIS® software. Feature Analyst® employs hierarchical machine learning techniques to extract the feature class information from the imagery using both spectral and inherent spatial relationships of objects. The methodology developed for the 7-class classification involved both automated and manual interpretation of objects. The overall accuracy of this LULC classification method, which identified more than 146,000 features, was 87.8% for the Buffalo River sub basin and surrounding areas.

Share

COinS