Abstract
Due to the lack of data caused by limited and uncertain observations, how to classify Resident Space Object (RSO) remains to be a difficult problem. Previous RSO classifications mainly focus on the problem when the “hard data” which are obtained by physical sensors are missing. They make use of features extracted from observation data which center on RSOs themselves only and are still very limited. This paper proposes to use an RSO Ontology named OntoStar to represent hard data and soft data. This representation not only describes RSOs themselves, but also links related objects to RSOs, establishing a more comprehensive and accurate RSO description to support more accurate and robust classifications. OntoStar not only contains mined feature deducting rules to refine the RSO feature information, but also includes a variety of mined RSO recognition rules to classify RSOs based on different sets of features. Experimental results show that RSO classification based on OntoStar can effectively solve the RSO classification problem under limited or uncertain observation conditions.
Highlights
Resident Space Object (RSO) recognition plays an important role in space exploration, protection of space assets, and especially space situational awareness(1-3)
The RSO classification task refers to the classification of space objects such as satellites, debris, and rockets in the earth's orbit based on observed data(2)
In order to solve the problem of inability to derive RSO types caused by missing features due to incomplete observations, or inconsistent conclusions due to too many multi-source data, we propose 2-phase RSO recognition
Summary
Resident Space Object (RSO) recognition plays an important role in space exploration, protection of space assets, and especially space situational awareness(1-3). Ontology-based classification uses feature deducting rules to estimate missing features, and obtains various RSO classification rules through machine learning to enrich the RSO classification knowledge base(2). In space situational awareness domain, the space object ontology can integrate multi-source RSO data(15,16), represent a variety of RSO features, and enhance RSO characterization and provide robust entity tracking(15). OntoStar (Ontology of Space Target Automatic Recognition) integrates domain knowledge and knowledge obtained by machine learning such as tactics, regulations, and inductive rules estimating specific features, targeting at solving RSO classification problem when some features are missing(2). The ontologies of linked multi-source data are represented, which can be used in semantic rule mining methods to obtain hidden relational patterns, background knowledge, and interesting and useful features(17). Two-phase RSO classification is proposed to fully utilize the characteristics of hard data and soft data, aiming at classifying RSOs under circumstances of limited and uncertain observations
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
