Date of this Version
We present an intelligent system for satellite sea ice image analysis named ARKTOS (Advanced Reasoning using Knowledge for Typing Of Sea ice). The underlying methodology of ARKTOS is to perform fully automated analysis of sea ice images by mimicking the reasoning process of sea ice experts and photo-interpreters. Hence, our approach is feature-based, rule-based classification supported by multisource data fusion and knowledge bases. A feature can be an ice floe, for example. ARKTOS computes a host of descriptors for that feature and then applies expert rules to classify the floe into one of several ice classes. ARKTOS also incorporates information derived from other sources, fusing different data towards more accurate classification. This modular, flexible, and extensible approach allows ARKTOS be refined and evaluated by expert users. As a software package, ARKTOS comprises components in image processing, rule-based classification, multisource data fusion, and GUI-based knowledge engineering and modification. As a research project over the past 10 years, ARKTOS has undergone phases such as knowledge acquisition, prototyping, refinement, evaluation and deployment, and finally operationalization at the National Ice Center (NIC). In this paper, we will focus on the methodology of ARKTOS.