Texture evaluation of RADARSAT imagery for wetland mapping

Abstract

Wetland studies have shown that single-date radar satellite images produce poor land cover classification results, whereas the results obtained from multi-date radar images are comparable with those obtained from optical satellite images. In some cases however, only single-date data are available. Therefore, improving the classification accuracy of single-date radar images is valuable because they provide a reliable alternative to optical images, whose accessibility is limited due to the presence of cloud. This paper evaluates the use of texture analysis on the single-date radar satellite imagery for wetland classification. A single-date RADARSAT image is used to classify the wetland and other land cover types in Walpole Island, Ontario, Canada. The grey-level co-occurrence matrix (GLCM) is employed for texture measurement, and a maximum likelihood supervised classifier is used for classification. Considerations are given to the combination of texture features and GLCM input parameters, such as window size, inter-pixel distance, and inter-pixel angle. The results of this study indicate that texture analysis of a single-date RADARSAT image greatly improved the wetland classification accuracy. Selecting the suitable combination of texture features and using the appropriate input parameters to generate these texture features are crucial to obtaining the best results.