Radio Frequency Fingerprint Recognition for Uav in Low-Altitude Intelligent Networks

In low-altitude intelligent networks, reliable identification of individual uncrewed aerial vehicles (UAVs) is crucial for ensuring airspace security. Radio frequency (RF) fingerprinting offers unique advantages such as passive detection, noncooperative identification, and resilience to encryption or spoofing, making it an effective approach for distinguishing individual UAVs. However, RF fingerprint recognition remains challenging due to multipath fading, interference, and channel variations between line-of-sight (LoS) and non-line-of-sight (NLoS) conditions. This paper presents a robust UAV identification framework that integrates signal enhancement, supervised contrastive learning, and feature distillation. Specifically, a three-stage signal preprocessing pipeline is developed, which consists of energy-driven segmentation, interference-aware filtering, and spectral shifting with downsampling to extract high-quality UAV control bursts. Then, a supervised contrastive encoder learns discriminative and channel-robust representations. Furthermore, an attentionguided center aggregation and feature distillation module is introduced to align encoder features with aggregated identity centers, improving intra-class compactness and cross-scenario stability. Experiments on the DroneRFb-DIR dataset demonstrate that the proposed method achieves superior performance, with 91.5% accuracy, 0.882 F1-score, a Silhouette Score of 0.237, and an inter/intra-class ratio (IIR) of 2.69, surpassing existing baselines. These results confirm the method's effectiveness and robustness for UAV RF fingerprint recognition under varying LoS/NLoS channel conditions.