Share Email Print
cover

Proceedings Paper

Object-based vs. per-pixel classification of aster imagery for land cover mapping in semi-arid areas
Author(s): Mustafa M. El Abbas; E. Csaplovics
Format Member Price Non-Member Price
PDF $14.40 $18.00

Paper Abstract

Due to complexity and spectral similarity in semi-arid areas, land cover mapping with remotely sensed data encounters serious problems when applying methods based on spectral information and ignore spatial information. A research was conducted in the Blue Nile area of Sudan to evaluate the effectiveness of Object-Based analysis (OB) approaches versus pixel based approach to generate Land Use Land Cover (LULC) thematic maps based on multi-spectral imagery data. Maximum Likelihood Classifier (MLC) was applied to examine if the spectral properties of the selected classes alone can be discriminated effectively. Nine land cover classes were generated with only about 82% overall accuracy. Different segmentation strategies were applied with the OB paradigm that might be effective to separate similar spectral values into a basic unclassified image objects in groups of relatively homogeneous pixels based on shape and compactness criterion at different scales. The segmented objects assigned to different classes with methods of membership functions and Nearest Neighbor Classifiers (NNC). The membership functions (RB) provided highly overall classification accuracy (95%), while NNC achieved about 89% accuracy. This study emphasized that the OB methods applied in this study provides more accurate results than the classical per-pixel approach especially when user's expert knowledge is presented.

Paper Details

Date Published: 27 October 2011
PDF: 12 pages
Proc. SPIE 8181, Earth Resources and Environmental Remote Sensing/GIS Applications II, 81810O (27 October 2011); doi: 10.1117/12.898317
Show Author Affiliations
Mustafa M. El Abbas, Technical Univ. of Dresden (Germany)
Univ. of Khartoum (Sudan)
E. Csaplovics, Technical Univ. of Dresden (Germany)


Published in SPIE Proceedings Vol. 8181:
Earth Resources and Environmental Remote Sensing/GIS Applications II
Ulrich Michel; Daniel L. Civco, Editor(s)

© SPIE. Terms of Use
Back to Top