Classification of Military Aircraft in Real-time Radar Systems based on Supervised Machine Learning with Labelled ADS-B Data

Abstract

Air surveillance is usually based on real-time radar tracking systems, which are able to provide object positions, kinematics and a short time history. Due to the density of the air picture, air traffic controllers normally focus on the actual object kinematics and the full identities of each object, which is received from secondary radars and ADS-B. However air surveillance systems in the military domain need additional information on objects classification and identification, since ADS-B of non-cooperative targets are not available. Hence flight characteristics and moving patterns are used as evidence for a military aircraft, which unfortunately are not often recognizable easily in real-time by an operator. This paper describes dedicated machine learning techniques that are trained with ADS-B data to predict military targets. The classifiers can be used within real-time systems.