Author

Delisio

Date of Award

6-2008

Degree Name

Master of Arts

Department

Geography

First Advisor

Dr. Charles Emerson

Second Advisor

Dr. Chansheng He

Third Advisor

Dr. Gregory Veeck

Access Setting

Masters Thesis-Open Access

Abstract

Pixel-based and object-oriented image analysis methods are two popular techniques for classifying satellite imagery. Pixel-based analysis is typically used for coarser resolution imagery while object-oriented analysis is ideal for high resolution imagery. However, the ability of object-oriented image analysis to segment images based on factors such as shape, color, texture, and spatial attributes suggest that it can perform as well or better than pixel-based analysis in classifying medium resolution imagery. A comparative analysis of the two methods was performed using Landsat 5 Thematic Mapper imagery from September 2006 to classify grassland quality and land cover in Da'erhanmaoming'an (DaMao) Banner, Inner Mongolia, China. After calculating a Normalized Difference Vegetation Index (NDVI) for each of the three images, a supervised classification was performed separately using the two methods. Land cover classes included three types of grassland (good, average, poor), built up/urban areas, agriculture, and barren. It was found that the object-oriented methodology produced higher overall accuracy and KHAT percentages in all three images. The highest achieved accuracy for the object-oriented technique was 84.93% with a KHAT of 0.8006 while the highest accuracy for the pixel based technique for the pixel based technique was 82.83% with a KHAT of 0.697.

Included in

Geography Commons

Share

COinS