《《automated classification of operational sar sea ice images》.pdfVIP

2010 Canadian Conference Computer and Robot Vision Automated Classification of Operational SAR Sea Ice Images Shuhrat Ochilov and David A. Clausi Vision and Image Processing (VIP) Research Group Systems Design Engineering, University of Waterloo Waterloo, ON, Canada {sochilov, dclausi}@engmail.uwaterloo.ca Abstract tions making training samples unreliable even within scene. Hence, the classification techniques relying on the thresh- The automated classification of operational sea ice satel- olds and statistics derived from training samples [19, 22] lite imagery is important for ship navigation and environ- might not perform well for operational use. The solution mental monitoring. Annually, thousands of large synthetic might be the investigation of features and statistics that have aperture radar (SAR) scenes are manually processed by the common characteristics over all sea ice scenes or extracting Canadian Ice Service (CIS) and pixel-level interpretation is features from samples obtained from the images themselves not feasible. Trained ice analysts divide SAR images into by initial unsupervised segmentation. The last option has ”polygon” areas and then identify the number and type of been adapted in this research and the sea ice classification ice classes per polygon. Full scene unsupervised classifica- is performed in two stages: unsupervised segmentation fol- tion can be performed by first segmenting each polygon into lowed by