文献类型：期刊文献

英文题名：DCG-CNT: Few-Shot UAV RF Fingerprinting Identification via Dynamic Category Grouping and Lightweight Hybrid Encoder

作者：Tang, Jijie Tan, Weijie Cao, Xin Chen, Xiyuan Li, Chunguo

第一作者：Tang, Jijie

通信作者：Tan, WJ[1]

机构：[1]Guizhou Univ, Coll Comp Sci & Technol, Guizhou Prov Lab Big Data, Guiyang 550025, Guizhou, Peoples R China;[2]Guizhou Inst Technol, Sch Aerosp Engn, Guiyang 550025, Peoples R China;[3]Southeast Univ, Sch Instrument Sci & Engn, Nanjing 210096, Peoples R China;[4]Southeast Univ, Natl Mobile Commun Res Lab, Nanjing 210096, Peoples R China

第一机构：Guizhou Univ, Coll Comp Sci & Technol, Guizhou Prov Lab Big Data, Guiyang 550025, Guizhou, Peoples R China

通信机构：corresponding author), Guizhou Univ, Coll Comp Sci & Technol, Guizhou Prov Lab Big Data, Guiyang 550025, Guizhou, Peoples R China.

年份：2026

卷号：30

起止页码：2650-2654

外文期刊名：IEEE COMMUNICATIONS LETTERS

收录：;EI(收录号：20263021153811);Scopus(收录号：2-s2.0-105045280541);WOS:【SCI-EXPANDED(收录号:WOS:001830588000001)】；

基金：This work is supported by the General Project of Guizhou Provincial Basic Research Program (Qian-ke-he-Foundation MS [2026]159), the National Natural Science Foundation of China (No. 62361010, U25B2007), Guizhou Provincial Science and Technology Plan Project (XKBF(2025) 032), and the National Key Research and Development Program of China (No. 2024YFC3807900).

语种：英文

外文关键词：Autonomous aerial vehicles; Modeling; Few shot learning; Labeling; Training; Transformers; Fingerprint recognition; Convolutional neural networks; Radio frequency; Prototypes; Unmanned aerial vehicle identification; few-shot learning; category grouping; lightweight model; radio frequency fingerprinting

摘要：UAV radio frequency fingerprinting identification (RFFI) is crucial for flight safety and spectrum security. However, practical deployments often suffer from limited labeled samples and severe class imbalance. To address these issues, this letter proposes few-shot UAV RFFI based on dynamic category grouping and a hybrid CNN-Transformer encoder, named DCG-CNT. First, the method adaptively aggregates imbalanced original categories into semantically balanced superclasses by a learnable dynamic category grouping mechanism, which can mitigate the problems of data scarcity and class imbalance. Then, we design a lightweight hybrid encoder, TransGroupNet, which integrates CNNs to extract local temporal features and Transformers to capture global dependencies. Extensive experiments on DroneRF, UAV-M100, and LoRa-RFFI datasets demonstrate that DCG-CNT significantly outperforms existing methods in both in-domain and cross-dataset few-shot scenarios, while maintaining low computational complexity.