ARKTOS: a knowledge engineering software package for satellite sea ice classification

The authors describe the knowledge engineering software package of their ARKTOS project. The ARKTOS project involves acquiring knowledge from sea ice experts as visual cues for sea ice features and classification rules and ultimately building an intelligent sea ice classifier. To assist in the knowledge acquisition, evaluation, and refinement phases, the authors have designed and built three Java-based graphical user interfaces (GUIs): arktosGUI, arktosViewer, and arktosEditor. arktosGUI facilitates feature-based knowledge refinement, focusing the experts' attention on individual features, specific attributes, and rules. It allows inspection of ice features visually, numerically, and symbolically and attribute impact analysis. The objective of arktosViewer is, on the other hand, to enable quick, region-based evaluation of the classification. It displays annotated, gridded images, maintains a bookkeeping of the evaluation sessions, and allows the experts to record their observation. Finally, the arktosEditor module has a rule indexing and search mechanism to go with its complete rule and threshold editing capabilities. This tool allows the experts to better edit and organize the rule base. The software package helps the authors design and refine ARKTOS as an intelligent knowledge-based system, and also address research issues in explicit encoding of domain expertise and capture of visual cues and semantics for intelligent image analysis in remote sensing, especially for computer-aided SAR sea ice image analysis.