Authors
Department of Surveying Engineering Techniques, Technical Engineering College of Kirkuk, Northern Technical University, Kirkuk, Iraq
Abstract
This paper explores different classification algorithms for the extraction of water bodies within TerraSAR-X polarimetric imagery with a case study situated in the Atchafalaya delta region of Louisiana. Various classification techniques, such as Maximum Likelihood Classification (MLH), Mahalanobis Distance (MHD), Minimum Distance (MND), Neural Networks (NN) and Support Vector Machines (SVM) were evaluated. Each approach was tested on its capability to delimit water in synthetic aperture radar (SAR) imagery despite environmental influences, such as wind speed effects and the emergence of vegetation.
The outcomes indicated that the conventional technique including MLH, MHD and MND had difficulty for accurate results where MLH achieved an overall accuracy of the 85.73% and Kappa of 0.7594, while MHD as well as MND also showed the results along the same line with the accuracy as 98.5% and Kappa of 0.97. The comparison shows that the NN and SVM methods are highly competitive compared with other methods, the NN method reached 99.57% accuracy and 0.9913 Kappa value, whereas SVM presented better classification performance than the other methods (99.69% for the overall accuracy and 0.9938 for the Kappa value).
These are essential inputs for refining the accuracy when great attention is required for deriving this information from SAR imagery, to support better environmental monitoring and management of resources.