Texture-based sea ice classification on TerraSAR-X imagery
Sea ice monitoring has attracted growing attention over the last decade due to its importance in global warming. Besides the purely scientific interest, practical implications of global warming are the increased navigability of ice-infested sea passages such as the Arctic Northwestern and Northeaste...
| Main Authors: | , |
|---|---|
| Format: | Conference Object |
| Language: | English |
| Published: |
Research Publishing Services, Singapur
2014
|
| Subjects: | |
| Online Access: | https://elib.dlr.de/89495/ https://elib.dlr.de/89495/1/2014_Ressel_Lehner_ICE14-1243_final.pdf |
| Summary: | Sea ice monitoring has attracted growing attention over the last decade due to its importance in global warming. Besides the purely scientific interest, practical implications of global warming are the increased navigability of ice-infested sea passages such as the Arctic Northwestern and Northeastern passages. To assist maritime endeavors in these areas, ice type classification is pivotal. National sea ice surveillance services of several countries have provided ice charts on a continuous basis, mostly generated by human experts in a manual fashion. These classifications are based on a variety of data sources, mostly from microwave or optical spaceborne and airborne sources. In this paper we present an approach that relies on TerraSAR-X Satellite data. Such data offers images at a high resolution in a radar band so far very rarely applied for ice classification. In order to build on expert knowledge of the past, we designed an artificial neural network approach, which outputs a number of suitable ice type classes. Input neurons are fed by an automated feature extraction algorithm. These features are based on popular and wellestablished texture analysis methods, most notably graylevel co-occurrence matrices (GLCM) and local binary patterns (LBP). Images are acquired for a selected eographical area for which ground truth data can be obtained from national ice services (Baltic sea). From these datasets, training and validation samples are chosen and evaluated. Classification results are compared with official ice charts. We asses suitability for near real time services. Based on first examples and computed results, we conclude that our approach is rather promising for automatic near real time (NRT) services. |
|---|