A New Paradigm for Knowledge-Data Driven Electromagnetic Target Representation

Abstract

Electromagnetic target representation is a common fundamental problem in electromagnetic space situational awareness. Early target representation was based on expert empirical knowledge, which required designers to have strong professional background and prior knowledge, and is performed poorly in complex signal environments. Deep learning, which has been developed in recent years, provides a new way for signal representation in complex electromagnetic environments. It simulates the deep structure of the human brain to build a machine learning model to automatically represent and process target data in an end-to-end manner, and shows good performance in perception tasks such as electromagnetic target detection, classification, identification, parameter estimation, and behavioral cognition. However, deep learning relies heavily on massive amounts of high-quality labelled data, and has certain limitations in the real electromagnetic environment. Incorporating know-ledge into intelligent systems has always been the research direction of artificial intelligence. Combining know-ledge and data for electromagnetic target representation will hopefully improve target perception accuracy and generalization ability, and is becoming a new direction in electromagnetic target representation. This paper reviews the development process of electromagnetic target representation techniques, and provide an outlook on the new paradigm of electromagnetic target perception driven by joint knowledge-data.

Keywords

• |target representation|expert knowledge|deep learning|joint knowledge-data-driven|know-ledge graph