Wetland mapping through semivariogram guided fuzzy segmentation of multispectral satellite imagery

To protect wetlands from loss, managers need tools to understand the status and trends of wetland resources. Remote sensing techniques provide a cost-effective way for wetland mapping and inventory establishment. However, a robust classification algorithm is the key to generate a reliable map from remotely sensed imagery. To identify wetlands from multispectral imagery, classifiers should take the natural phenomenon, i.e. spatial and spectral vagueness, into account. The Fuzzy C-Means (FCM) clustering algorithm is better suited for dealing with the imprecise data than traditional "hard" classifiers, but it completely ignores the spatial variability inherent in an image. In this thesis, the Semivariogram Guided Fuzzy C-Means (SGFCM) classifier, a modification of the FCM algorithm with spatial variances involved, has been developed for wetland mapping.; Two major tasks are included: replacing the Euclidean distance by the Mahalanobis distance and incorporating the semivariogram texture as spatial guidance in the fuzzy clustering algorithm. (Abstract shortened by UMI.)...