Sea ice concentration from single-polarized SAR data using second-order grey level statistics and learning vector quantization |
|
| Abstract | A classification system for obtaining the sea ice concentration from satellite based single-polarized SAR data is presented. The method is based on a supervised neural network classification of second-order grey level statistics features. Learning Vector Quantization is used for describing the boundaries between the surface classes. Five RADARSAT ScanSAR Wide images of the sea ice off of the East and West coast of Greenland were classified with resulting classification accuracies from 80 to 98 percent. The algorithm is robust to the selection of training data which was measured by classifying the same dataset twice using multiple experts for selecting training data. Thus, the precision by which sea ice and open water was assessed to be approximately two percent. The classification accuracy is dependent on the sea state, especially the degree to which the water is roughened by wind. The current classification system is not able to distinguish between high backscatter sea ice and high backscatter ocean areas and therefore areas of wind roughened sea water are masked out after classification. |
| Type | Technical report |
| Year | 2005 No. 05-04 |
| Publisher | Danish Meteorological Institute |
| Series | Scientific Report |
| Publication link | http://www.dmi.dk/dmi/sr05-04.pdf |
| BibTeX data | [bibtex] |
| IMM Group(s) | Image Analysis & Computer Graphics, Geoinformatics |