Statistical evaluation of decision-level fusion methods for non-cooperative target identification by radar signatures

A promising method of non-cooperative identification is the classification of a target by high-resolution radar signatures. By simultaneous tracking and classification one obtains a set of successive radar range profiles which contain information on the target from different aspect angles. At this point data fusion of the declaration series can help to stabilize the identification against misclassifications by putting it on a broader basis. In this paper we use an experimental data base filled with radar range profiles of eight different aircraft types of similar size for a statistical analysis and comparison of fused and unfused declaration series. As fusion methods we consider the simple voting algorithm as well as more advanced techniques as the Bayes and the Dempster-Shafer approach.