RFSD-REI: A Real and Fake Samples-Driven Radar Emitter Identification Method for Imbalanced Data.

To address the issue of radar emitter identification (REI) with imbalanced data, we propose a real and fake samples-driven REI method for imbalanced data. First, we use the synthetic minority oversampling technique (SMOTE) to generate fake samples and balance the imbalanced data. Next, we use the edited nearest neighbors (ENN) to purify the balanced dataset composed of real samples and generated fake samples. We then train a convolutional neural network (CNN) using the purified, balanced dataset containing both real and fake samples. Subsequently, we create a one-step optimization dataset using only the real samples from the original imbalanced data to further enhance the performance of the trained CNN. Finally, we use the enhanced CNN to identify different radar emitters. Our experimental results demonstrate the effectiveness, robustness, and generalization of the proposed method. This work provides a new solution to the problem of REI with imbalanced data and has the potential to improve the measurement and perception capabilities of electronic warfare forces.