Open-Set Radio Frequency Fingerprinting Identification Based on Feature Space Optimization and Transfer Learning

Current research on radio frequency fingerprinting identification (RFFI) mostly focuses on closed-set scenarios, with limited consideration of unknown samples in test sets. Also, collecting high-quality labeled RF fingerprint signals requires precise receivers. To address these issues, we propose a novel openset RFFI method combining feature space optimization and transfer learning. During training, the model is trained using cross-entropy loss and a combined loss function to optimize feature space distribution and enhance model performance. Additionally, transfer learning strategies are employed to handle small sample sizes and accelerate model convergence. Experiments show that the proposed method outperforms original OpenMax and centerloss-optimized OpenMax approaches.