Detection of icebergs using ERS SAR images

• Main Authors: Wesche, Christine, Dierking, Wolfgang
• Format: Conference Object
• Language: unknown
• Published: 2009
• Subjects: Weddell; Weddell Sea; Antarc*; Antarctica
• Online Access: https://epic.awi.de/id/eprint/21413/; https://hdl.handle.net/10013/epic.33763

Within the framework of the DFG project BergCAT we develop a method for detecting small icebergs in SAR images from the Weddell Sea, Antarctica. Dependent on the season, icebergs appear as bright or dark targets in SAR images. Statistical analyses of 624 iceberg regions of interest (ROIs) show that the threshold between bright and dark appearing icebergs is -7.5 dB. For a first classification, SAR data were divided into images with mainly bright and mainly dark appearing icebergs, independent from the background signature. The iceberg and background pixels were investigated separately for the two image groups to reveal possible correlations between thesignatures. We calculated the mean and the standard deviation and use the 1-sigma interval. The 1-sigma intervals show overlapping ranges of the backscattering coefficients of icebergs and background in the image groups and with the aid of these information we formulated detection conditions. Then we used a combination of thresholds in the non-overlapping ranges and a constant false alarm rate (CFAR) in the overlapping ranges. This detection algorithm detected 72 % of former extracted iceberg pixels as icebergs and 84 % of the background pixels as background.