Share Email Print

Proceedings Paper

Development of an algorithm for the knowledge-based classification of multitemporal SAR images
Author(s): Martin Habermeyer; Christiane C. Schmullius
Format Member Price Non-Member Price
PDF $17.00 $21.00

Paper Abstract

Radar images show a characteristic interference, the speckle effect, as a result of the coherent nature of the radar signal, complicating classifications based on the normal distribution. An algorithm is introduced to overcome the disadvantages of a pixel-based classification by use of an object-based approach through the use of digital field boundaries. Besides estimating the characteristic parameters on a field- rather than on a pixel-basis, a method is developed to incorporate prior knowledge into the classification process. An object-based texture approach is compared to a pixel-based approach. Both methods are based on the multinomial distribution. The texture approach incorporates the full information stored in the co-occurrence- matrix in the classification process. The pixel-based approach models the entire histogram over the multinomial distribution. The prior knowledge is formulated in the form of transition matrices. The results of the image classification and the prior knowledge are combined using Dempster's Rule of Shafer's Theory of Evidence because of its ability to combine probability values of various probability distributions. The algorithm was tested with rotation schemes of an agricultural area and, depending on the classification method, showed classification accuracies up to 94 percent.

Paper Details

Date Published: 24 September 1997
PDF: 11 pages
Proc. SPIE 3161, Radar Processing, Technology, and Applications II, (24 September 1997); doi: 10.1117/12.283950
Show Author Affiliations
Martin Habermeyer, DLR (Germany)
Christiane C. Schmullius, DLR (Germany)

Published in SPIE Proceedings Vol. 3161:
Radar Processing, Technology, and Applications II
William J. Miceli, Editor(s)

© SPIE. Terms of Use
Back to Top