LT-SEI: Long-Tailed Specific Emitter Identification Based on Decoupled Representation Learning in Low-Resource Scenarios

Abstract

In the case of COVID-19, which requires stable and reliable tracking of personnel movement, aircraft identification by specific emitter identification (SEI) is a hot-button issue. It refers to the process of identifying individual aircraft by comparing features extracted from the Radio Frequency (RF) signal of a given aircraft. Deep learning (DL) has been widely used in SEI research due to its excellent feature extraction capability, but in the actual low-resource reception scenario, the aircraft signal data acquired for training are long-tailed in distribution, and the imbalance of the signal data increases the challenge of training the network. In this paper, we propose a novel long-tailed specific emitter identification (LT-SEI) method using decoupled representation (DR) learning. Specifically, we separate the learning process into two stages: representation learning and classification, which includes unbalanced training and balanced classifier learning. The proposed DR-based LT-SEI approach is assessed using aircraft Automatic Dependent Surveillance Broadcast (ADS-B) data collected in the real world and compared to state-of-the-art methods. Experiment results show that the method has better long-tail recognition performance than the existing methods. When the data imbalance factor is 0.01, the F1 score of the model for the recognition result can reach 71.2%, which is 9% higher than that of the baseline model.