Performance Analysis of Time-series Network for Pattern Classification of Dynamic Radar Cross Section
A-Ran Kim, Ha-Seon Kim, Chang-Ho Kang, Sunyoung Kim · Journal of Institute of Control Robotics and Systems · 2023
In this paper, we propose a method to conduct dynamic radar cross section (DRCS) measurements. This method uses an electromagnetic simulator for the target classification of moving objects and a time-series network with excellent classification performance. The DRCS method yields measurements based on theoretical calculations after the generation of the static RCS. To propose a reliable, time-series network with excellent classification performance, we consider two cases. The first uses only a ballistic trajectory, and the second uses a maneuver and ballistic trajectory. In addition, to compensate for the disadvantage of the RCS, which is vulnerable to noise, we simulated a noisy situation. We confirmed that the classification performance was lower in the first compared with the second case. However, a bidirectional long short-term memory (BiLSTM) yielded the best classification performances in both cases, even in noisy conditions. Furthermore, we verified that the performance of LSTM was significantly improved compared with that of the gated recurrent unit when the bidirectional theory was applied. Hence, we suggest the use of an optimized BiLSTM as a DRCS classification network.