SAR image classification of ice in the Gulf of Bothnia

This paper investigates the potential of ERS-l SAR imagery for sea ice classification. Two classification approaches, a pixel-based and a region-based, are developed and compared using the BEPERS-88 airborne SAR imagery together with the aerial photography for reference. The pixel-based approach is based on adaptive filtering, while the region-based approach is based on segmentation using edge detection with a region growing algorithm. The latter gives the better result as found by inspecting the confusion matrices and the classified images. The region-based method is tested on a simulated ERS-l Fast Delivery imagery based on BEPERS-data. It is shown that for the studied case a classification accuracy of 80 per cent is obtained when classifying the ERS-l SAR image into four classes (open water, level ice, brash ice and rubble fields). A classification accuracy of 90 per cent is obtained for three classes when brash ice and rubble fields are combined.