基于双频配对分段与多表示学习的无人机射频识别方法
RF-Based UAV Recognition with Dual-Band Paired Segments and Multi-Representation Learning
摘要: DroneRF以长序列形式存储射频数据，先生成短窗口再随机划分可能使同一低频段/高频段(L/H)配对分段的相关窗口进入不同数据子集。本文将454个单频段CSV文件整理为227个L/H配对分段，在窗口生成前划分为160、34和33个训练、验证与测试配对分段；每个配对分段生成32个尺寸为2 × 8192的双频索引对齐窗口，共计7264个窗口，并固定数据清单、测试清单和指标规则。在该协议下，采用Raw-FFT-STFT三分支网络比较CE、MildAug、Source-aware SupCon及联合配置。三个随机种子下，联合配置按窗口统计的平衡准确率为0.9926 ± 0.0051，较CE的0.9901 ± 0.0037提高0.25个百分点，但该变化未在全部随机种子中保持一致；四种配置汇总33个测试配对分段后均正确分类。去除Raw和STFT分支后，平衡准确率分别下降0.0059和0.0022；去除FFT分支后的变化为+0.0006，处于实验波动范围内。固定随机种子下，非零对比损失权重均未超过CE；Source-aware SupCon提高了跨配对分段的同类相似度，但伴随类别间隔收缩。所建协议为固定数据与统一规则下的方法比较提供了可复核基础，结论限于DroneRF现有文件信息与Background/UAV二分类任务。
Abstract: DroneRF stores radio-frequency measurements as long sequences. Randomly splitting short windows may place correlated windows from the same low-/high-band (L/H) paired segment in different subsets and reduce evaluation reliability. This study therefore organizes 454 single-band CSV files into 227 L/H paired segments according to BUI activity codes and segment indices, and assigns 160, 34, and 33 paired segments to the training, validation, and test sets before window generation. Each paired segment yields 32 index-aligned dual-band windows of size 2 × 8192, producing 7264 windows. Data manifests, the test list, and metric definitions are fixed, and identifier exclusivity is audited. Under this protocol, a Raw-FFT-STFT three-branch network is evaluated with cross-entropy (CE), mild data augmentation (MildAug), Source-aware SupCon, and their joint configuration. Across three random seeds, the joint configuration achieves a per-window balanced accuracy of 0.9926 ± 0.0051 versus 0.9901 ± 0.0037 for CE, an increase of 0.25 percentage points that is not retained for every seed. Probability aggregation correctly classifies all 33 test paired segments for all four configurations. Removing the Raw and STFT branches reduces balanced accuracy by 0.0059 and 0.0022, respectively, whereas removing FFT changes it by +0.0006, within experimental variation. With a fixed seed, no nonzero contrastive-loss weight exceeds CE. Source-aware SupCon increases within-class similarity across paired segments but reduces class separation. The protocol provides a reproducible basis for comparison under fixed data and evaluation rules; the conclusions are limited to the available DroneRF metadata and the Background/UAV binary task.
文章引用：林成琨, 郝东来, 何天洋, 王加瑞. 基于双频配对分段与多表示学习的无人机射频识别方法[J]. 计算机科学与应用, 2026, 16(9): 140-153. https://doi.org/10.12677/csa.2026.169296

参考文献

[1] Shi, X., Yang, C., Xie, W., Liang, C., Shi, Z. and Chen, J. (2018) Anti-Drone System with Multiple Surveillance Technologies: Architecture, Implementation, and Challenges. IEEE Communications Magazine, 56, 68-74.
https://doi.org/10.1109/mcom.2018.1700430
[2] O’Shea, T.J. and Hoydis, J. (2017) An Introduction to Deep Learning for the Physical Layer. IEEE Transactions on Cognitive Communications and Networking, 3, 563-575.
https://doi.org/10.1109/tccn.2017.2758370
[3] Al-Sa’d, M.F., Al-Ali, A., Mohamed, A., Khattab, T. and Erbad, A. (2019) RF-Based Drone Detection and Identification Using Deep Learning Approaches: An Initiative towards a Large Open Source Drone Database. Future Generation Computer Systems, 100, 86-97.
https://doi.org/10.1016/j.future.2019.05.007
[4] Allahham, M.S., Al-Sa’d, M.F., Al-Ali, A., Mohamed, A., Khattab, T. and Erbad, A. (2019) DroneRF Dataset: A Dataset of Drones for RF-Based Detection, Classification and Identification. Data in Brief, 26, Article ID: 104313.
https://doi.org/10.1016/j.dib.2019.104313
[5] Nemer, I., Sheltami, T., Ahmad, I., Yasar, A.U. and Abdeen, M.A.R. (2021) RF-Based UAV Detection and Identification Using Hierarchical Learning Approach. Sensors, 21, Article No. 1947.
https://doi.org/10.3390/s21061947
[6] Shulman, D. (2026) How Much Do RF Drone Benchmarks Overstate? A Controlled Study and Theory of Data Leakage in UAV Signal Identification.
https://doi.org/10.48550/arXiv.2607.01025
[7] Kapoor, S. and Narayanan, A. (2023) Leakage and the Reproducibility Crisis in Machine-Learning-Based Science. Patterns, 4, Article ID: 100804.
https://doi.org/10.1016/j.patter.2023.100804
[8] Khosla, P., Teterwak, P., Wang, C., Sarna, A., Tian, Y., Isola, P., Maschinot, A., Liu, C. and Krishnan, D. (2020) Supervised Contrastive Learning. Advances in Neural Information Processing Systems 33: Annual Conference on Neural Information Processing Systems 2020, NeurIPS 2020, 6-12 December 2020, 18661-18673.
[9] O’Shea, T.J., Corgan, J. and Clancy, T.C. (2016) Convolutional Radio Modulation Recognition Networks. In: Jayne, C. and Iliadis, L., Eds., Engineering Applications of Neural Networks, Springer International Publishing, 213-226.
https://doi.org/10.1007/978-3-319-44188-7_16
[10] Mo, Y., Huang, J. and Qian, G. (2022) Deep Learning Approach to UAV Detection and Classification by Using Compressively Sensed RF Signal. Sensors, 22, Article No. 3072.
https://doi.org/10.3390/s22083072
[11] Cohen, L. (1995) Time-Frequency Analysis. Prentice Hall PTR.
[12] Ganin, Y., Ustinova, E., Ajakan, H., Germain, P., Larochelle, H., Laviolette, F., Marchand, M. and Lempitsky, V. (2016) Domain-Adversarial Training of Neural Networks. Journal of Machine Learning Research, 17, 1-35.
[13] Sagawa, S., Koh, P.W., Hashimoto, T.B. and Liang, P. (2020) Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalization. 8th International Conference on Learning Representations, ICLR 2020, Addis Ababa, 26-30 April 2020, 1-19.
https://openreview.net/forum?id=ryxGuJrFvS