Semantic Sea-Ice Classification for Belgica Bank in Greenland

Each Sentinel-1 image is tiled into patches of 256x256 pixels. The size of the images is different and we reduced to the smallest one. In total for each image are 6,400 patches [1-2, 4]. See the excel file for the 24 Sentinel-1 ids.

The semantic classes are:

1. Black border
2. Old ice
3. First-Year ice
4. Glaciers
5. Icebergs
6. Mountains
7. Young ice
8. Water group

The last class combines the Floating ice, Water body, Water ice current and melted snow defined in [3] because they have very similar physical properties.

References:

1. C.O. Dumitru, G. Schwarz, C. Karmakar, and M. Datcu, “Machine Learning-Based Paradigm for Boosting the Semantic Annotation of EO Images”, IGARSS, Belgium, July 2021, pp. 1-4.

2. C. Karmakar, C.O. Dumitru, and M. Datcu, “Explainable AI for SAR Image Time Series: Knowledge Extraction for Polar Areas”, MDPI Remote Sensing Journal, 2021, pp. 1-21 (under review).

3. C.O. Dumitru, V. Andrei, G. Schwarz, and M. Datcu, “Machine Learning for Sea Ice Monitoring from Satellites”, Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-2/W16, pp. 83-89, 2019.

4. C. Karmakar, C.O. Dumitru, G. Schwarz, and M. Datcu, “Feature-Free Explainable Data Mining in SAR Images Using Latent Dirichlet Allocation”, IEEE JSTARS, vol. 14, pp. 676-689, 2021.