无人机(UAV)检测和识别技术一直以来都是国内外研究的热门领域。但近年来， 随着无人机逐渐自主化和智能化发展，单无人机已无法满足日益复杂的任务需求，能 有效提高无人机生存能力和任务效率的多无人机协同作战受到以中美为首的军事强 国的高度关注和大力发展。然而，目前绝大多数的研究都是基于单无人机目标场景进 行识别的，当面临敌方多个数量的无人机攻击时，即使检测到攻击目标但由于对敌方 无人机数量的未知性从而导致无法准确判断并识别出包含的所有目标个体，最终导致 防御系统失效。因此，本文针对多无人机信号混叠场景下的目标个体识别问题，在进 行射频指纹特征有效提取的同时，构建了多通道的胶囊神经网络模型并设置相应的目 标识别算法。除此之外，本文还提出一种增强地类增量学习方法来解决在实际应用中， 当不断有新的无人机类别出现时，如何在仅利用新获取的目标类型样本的前提下更新 和完善已有模型的问题。本文的主要研究内容如下： （1）首先，不同于传统的针对单目标信号进行的希尔伯特变换、小波变换等特 征提取的方法，本文在得到电磁通信射频信号数据的基础上，针对多无人机目标混合 信号的特点，提出有效的射频指纹特征提取方法：根据不同的无人机具有不同旋转桨 的 RCS 特点，选择 CFS 特征作为区分不同无人机的有效特征；根据 FFT2 变换可以 反应空间域中复杂信号分量的特点，对原始信号的时频图做进一步的预处理，从而将 提取出的能体现多目标混合信号中各信号分量的 K 空间作为有效特征。 （2）其次，在针对多无人机场景中混叠目标数量已知的识别任务中，本文采用 胶囊网络作为模型的基础框架，克服了传统神经网络对信号空间特征关系分辨率低的 局限。同时为达到更好的识别效果，设计了多通道的胶囊网络模型，结合特征融合机 制学习信号的多视觉特征，提高模型对混合多目标的识别性能。此外，与现有技术相 比，该模型的优势在于：不需要经过复杂的信号分离预处理算法，仅需通过训练单目 标场景下无人机个体的数据训练集就可以达到同时识别多目标个体的目的。 （3）然后，为了提高模型的泛化能力，使模型即使在面对更为复杂的实际环境 中依旧具有较强的识别能力，本文继续针对多无人机场景中混叠目标数量未知的识别 任务进行研究。在原有模型的基础上做出改进的同时，设置有效的识别算法，通过引 入门限值，并利用虚警概率对该门限值进行合适的设置，从而实现在未知目标数量场 景下自动判断无人机的目标数量并对其进行正确识别分类的目的。 （4）最后，当有新类别的无人机样本到达时，传统的深度学习模型在仅使用新 样本的数据对模型进行训练时，模型会逐渐遗忘之前学习到的旧类别知识。因此，针 对实际中的这一缺陷，本文提出将增量学习应用到无人机识别领域，创造性的将胶囊网络与 LwF 增量方法相结合，并在此基础上提出两个改进的方法：其一是代替传统 的交叉熵损失，而是将边缘损失与蒸馏损失相结合；其二，采用可变的蒸馏系数，使 得模型能够根据新旧类别的比例实时自动调整。仿真结果显示改进后的方法对模型的 可扩展性方面有了较大的提升。

The detection and identification technology of UAV has always been a hot research field at home and abroad. However, in recent years, with the gradual autonomy and intelligent development of UAV, single UAV can no longer meet the increasingly complex mission requirements. Multi-UAV cooperative operations, which can effectively improve the survivability and mission efficiency of UAV, have been highly concerned and vigorously developed by military powers led by China and the United States. However, most of the current research is based on the single UAV target scene recognition. When facing the attack of multiple enemy UAVs, even if the attack target is detected, all the target individuals can not be accurately judged and identified due to the unknown number of enemy UAVs, which eventually leads to the failure of the defense system. Therefore, aiming at the target individual recognition problem in the aliasing scene of multiple UAV signals, this paper constructs a multi-channel capsule neural network model and sets the corresponding target recognition algorithm while effectively extracting RF fingerprint features. In addition, this paper also proposes an enhanced class incremental learning method to solve the problem of how to update and improve the existing model on the premise of only using the newly acquired target type samples when new categories of UAVs constantly appear in practical applications. The main research contents of this paper are as follows: First of all, different from the traditional feature extraction methods such as Hilbert transform and wavelet transform for single target signals, this paper proposes an effective RF fingerprint feature extraction method based on the RF signal data of electromagnetic communication, aiming at the characteristics of mixed signals of multiple UAV targets: According to the RCS characteristics of different UAVs with different rotary blades, CFS characteristics were selected as the effective characteristics to distinguish different UAVs. According to the FFT2 transformation can reflect the characteristics of complex signal components in the space domain, the time-frequency graph of the original signal is further preprocessed, so that the extracted K space which can reflect each signal component in the multi-target mixed signal is taken as the effective feature. Secondly, in the recognition task of the known number of aliases targets in the multi-UAV scene, the capsule network is used as the basic framework of the model in this paper to overcome the limitation of the low resolution of the spatial feature relation of signals of the traditional neural network. At the same time, in order to achieve better recognition effect, a multi-channel capsule network model is designed, combining with the feature fusion mechanism to learn the multi-visual features of the signal, to improve the recognition performance of the model for mixed multi-targets. In addition, compared with the existing technology, the advantage of this model is that it does not need to go through the complex signal separation pretreatment algorithm, but can achieve the purpose of simultaneously identifying multiple target individuals by training the data training set of the UAV individual in the single target scenario. Then, in order to improve the generalization ability of the model and make the model still have strong recognition ability even in the face of a more complex actual environment, this paper continues to study the recognition task of the unknown number of aliasing targets in the multi-UAV scene. In addition to making improvements on the basis of the original model, an effective recognition algorithm is set. By introducing a threshold value, the threshold value is appropriately set with false alarm probability, so as to realize the purpose of automatically judging the number of UAV targets and correctly identifying and classifying them under the unknown number of targets. Finally, when a UAV sample of a new category arrives, the traditional deep learning model will gradually forget the knowledge of the old category learned before training the model only with the data of the new sample. Therefore, in view of this defect in practice, this paper proposes to apply incremental learning to the field of UAV recognition, creatively combines capsule network with LwF method, and proposes two improved methods on this basis: one is to replace the traditional cross entropy loss, but to combine the margin loss and distillation loss; Second, a variable distillation coefficient is used, so that the model can be adjusted automatically in real time according to the ratio of old and new categories. The simulation results show that the improved method greatly improves the scalability of the model.

[1] Alsalam B H Y, Morton K, Campbell D, et al. Autonomous UAV with vision based on-board decision making for remote sensing and precision agriculture[C]//2017 IEEE Aerospace Conference. IEEE, 2017: 1-12.
[2] 张凌寒, 杜婧. 民法典背景下无人机侵害个人隐私的法律规制[J]. 西北工业大学学报(社会科学版), 2020, 145(3): 85-92.
[3] 安恬澹. 增量学习在雷达HRRP目标识别中的应用[D]. 西安电子科技大学, 2020.
[4] Tao Y, Tu Y, Shyu M L. Efficient incremental training for deep convolutional neural networks[C]//2019 IEEE Conference on Multimedia Information Processing and Retrieval (MIPR). IEEE, 2019: 286-291.
[5] McCloskey M, Cohen N J. Catastrophic interference in connectionist networks: The sequential learning problem[M]. Psychology of learning and motivation. Academic Press, 1989, 24: 109-165.
[6] French R M. Catastrophic forgetting in connectionist networks[J]. Trends in cognitive sciences, 1999, 3(4): 128-135.
[7] Bernardini A, Mangiatordi F, Pallotti E, et al. Drone detection by acoustic signature identification[J]. Electronic Imaging, 2017, 19(10): 60-64.
[8] Nguyen P, Truong H, Ravindranathan M, et al. Cost-effective and passive rf-based drone presence detection and characterization[J]. GetMobile: Mobile Computing and Communications, 2018, 21(4): 30-34.
[9] Ivanov L I, Obukhova N A, Baranov P S. Review of modern UAV detection algorithms using methods of computer vision[C]//2020 IEEE Conference of Russian Young Researchers in Electrical and Electronic Engineering (EIConRus). IEEE, 2020: 322-325.
[10] Zheng Y, Chen Z, Lv D, et al. Air-to-air visual detection of micro-uavs: An experimental evaluation of deep learning[J]. IEEE Robotics and automation letters, 2021, 6(2): 1020-1027.
[11] Liu H, Wei Z, Chen Y, et al. Drone detection based on an audio-assisted camera array[C]//2017 IEEE Third International Conference on Multimedia Big Data (BigMM). IEEE, 2017: 402-406.
[12] Aledhari M, Razzak R, Parizi R M, et al. Sensor fusion for drone detection[C]//2021 IEEE 93rd Vehicular Technology Conference (VTC2021-Spring). IEEE, 2021: 1-7.
[13] 章鹏飞, 李刚, 霍超颖,等. 基于双雷达微动特征融合的无人机分类识别[J]. 雷达学报, 2018, 7(5):1-8.
[14] Shi X, Yang C, Xie W, et al. Anti-drone system with multiple surveillance technologies: Architecture, implementation, and challenges[J]. IEEE Communications Magazine, 2018, 56(4):68-74.
[15] Basak S, Rajendran S, Pollin S, et al. Combined RF-based drone detection and classification[J]. IEEE Transactions on Cognitive Communications and Networking, 2021, 8(1): 111-120.
[16] Zong L, Xu C, Yuan H L. A RF Fingerprint Recognition Method Based on Deeply Convolutional Neural Network[C]//2020 IEEE 5th Information Technology and Mechatronics Engineering Conference(ITOEC). IEEE, 2020: 1778-1781.
[17] Peng L, Zhang J, Liu M, et al. Deep learning based RF fingerprint identification using differential constellation trace figure[J]. IEEE Transactions on Vehicular Technology, 2019, 69(1): 1091-1095.
[18] Gong J, Xu X, Lei Y. Unsupervised specific emitter identification method using radio-frequency fingerprint embedded InfoGAN[J]. IEEE Transactions on Information Forensics and Security, 2020, 15: 2898-2913.
[19] Wang B, Liu X, Yu B, et al. An Improved Wi Fi Positioning Method Based on Fingerprint Clustering and Signal Weighted Euclidean Distance[J]. Sensors, 2019, 19(10): 2294-2300.
[20] C. Zhao, M. Shi, Z. Cai, et al. Detection of unmanned aerial vehicle signal based on Gaussian mixture model[C]. International Conference on Computer Science and Education (ICCSE), 2017: 289-293.
[21] Zhang Z, Wang C, Gan C, et al. Automatic modulation classification using convolutional neural network with features fusion of SPWVD and BJD[J]. IEEE Transactions on Signal and Information Processing over Networks, 2019, 5(3): 469-478.
[22] Liu G, Xi Z, Liu R. A Novel Wireless Interference Identification and Scheduling Method based on Convolutional Neural Network[C]//2022 IEEE International Conference on Communications Workshops (ICC Workshops). IEEE, 2022: 1-6.
[23] Youssef K, Bouchard L, Haigh K, et al. Machine learning approach to RF transmitter identification[J]. IEEE Journal of Radio Frequency Identification, 2018, 2(4): 197-205.
[24] Grimaldi S, Mahmood A, Gidlund M. Real-time interference identification via supervised learning: Embedding coexistence awareness in IoT devices[J]. IEEE Access, 2018, 7: 835-850.
[25] Guibene W, Slock D. Signal separation and classification algorithm for cognitive radio networks[C]//2012 International Symposium on Wireless Communication Systems (ISWCS). IEEE, 2012: 301-305.
[26] 曾政智, 周嘉伟, 罗正华. 同频段混合信号中的无人机信号盲检测识别[J]. 电讯技术, 2020, 60(6): 689-694.
[27] Zhang M L, Zhou Z H. A review on multi-label learning algorithms[J]. IEEE transactions on knowledge and data engineering, 2013, 26(8): 1819-1837.
[28] Zhang W, Li G, Baker C. Dictionary Learning for Radar Classification of Multiple Micro-Drones[C]//2019 International Radar Conference (RADAR). IEEE, 2019: 1-4.
[29] Chen H, Kim S J. Robust RF Mixture Signal Recognition Using Discriminative Dictionary Learning[J]. IEEE Access, 2021, 9: 141107-141120.
[30] Karita S, Watanabe S, Iwata T, et al. Semi-supervised end-to-end speech recognition using text-to-speech and autoencoders[C]//ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2019: 6166-6170.
[31] Sazdić-Jotić B, Pokrajac I, Bajčetić J, et al. Single and multiple drones detection and identification using RF based deep learning algorithm[J]. Expert Systems with Applications, 2022, 187: 115928.
[32] Girosi F, Jones M, Poggio T. Regularization theory and neural networks architectures[J]. Neural computation, 1995, 7(2): 219-269.
[33] KIRKPATRICK J, PASCANU R, RABINOWITZ N, et al. Overcoming catastrophic forgetting in neural networks[J]. Proceedings of the National Academy of Sciences, 2017, 114(13): 3521-3526.
[34] KANG M, PARK J, HAN B. Class-incremental learning by knowledge distillation with adaptive feature consolidation[C]// Proceedings of the IEEE/ CVF Conference on Computer Vision and Pattern Recognition, 2022: 16071-16080.
[35] Liu Y, Su Y, Liu A A, et al. Mnemonics training: Multi-class incremental learning without forgetting[C]//Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition. 2020: 12245-12254.
[36] REBUFFI S A, KOLESNIKOV A, SPERL G, et al. iCaRL: incremental classifier and representation learning[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 2001-2010.
[37] Zhang J, Zhang J, Ghosh S, et al. Class-incremental learning via deep model consolidation[C]//Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. 2020: 1131-1140.
[38] Hu N, Ge Z, Chen X, et al. Research on Online Reinforcement Learning Method Based on Experience-Replay[C]//2018 IEEE International Conference on Information and Automation (ICIA). IEEE, 2018: 1338-1343.
[39] Mallya A, Lazebnik S. Packnet: Adding multiple tasks to a single network by iterative pruning[C]//Proceedings of the IEEE conference on Computer Vision and Pattern Recognition. 2018: 7765-7773.
[40] KANAKIS M, BRUGGEMANN D, SAHA S, et al. Reparameterizing convolutions for incremental multi-task learning without task interference[C]// European Conference on Computer Vision. Springer, Cham, 2020: 689-707.
[41] 严丹莉. 基于通信信号指纹的无人机个体识别[D].电子科技大学, 2020.
[42] 任东方, 张涛, 韩洁. 结合ITD与非线性分析的通信辐射源个体识别方法[J]. 信号处理, 2018, 34(3): 331-339.
[43] Yan G, Ma G, Zhu L, et al. Combining Multiple Support Vector Machines using Fuzzy Integral for Classification[C]//2006 International Conference on Machine Learning and Cybernetics. IEEE, 2006: 3438-3441.
[44] 王文聪. 基于神经网络的增量学习模型研究与实现[D]. 吉林大学, 2021.
[45] Abraham W C, Robins A. Memory retention–the synaptic stability versus plasticity dilemma[J]. Trends in neurosciences, 2005, 28(2): 73-78.
[46] Liu L, Kuang Z, Chen Y, et al. IncDet: In Defense of Elastic Weight Consolidation for Incremental Object Detection[J]. IEEE Transactions on Neural Networks and Learning Systems, 2021, 32(6): 2306-2319.
[47] 贺文亮, 朱敏玲. 胶囊神经网络研究现状与未来的浅析[J]. 计算机工程与应用, 2021, 57(3): 1-11.
[48] Mazzia V, Salvetti F, Chiaberge M. Efficient-capsnet: Capsule network with self-attention routing[J]. Scientific reports, 2021, 11(1): 14634.
[49] 潘迪祺. 基于胶囊网络的多标签图像分类和语义分割[D]. 兰州大学, 2020.
[50] Kim B K, Kang H S, Park S O. Drone Classification Using Convolutional Neural Networks With Merged Doppler Images[J]. IEEE Geoscience & Remote Sensing Letters, 2016, 14(1):38-42.
[51] Nguyen P, Truong H, Ravindranathan M, et al. Matthan: Drone presence detection by identifying physical signatures in the drone's rf communication[C]//Proceedings of the 15th Annual International Conference on Mobile Systems, Applications, and Services. ACM, 2017: 211-224.
[52] Zhang W, Li G. Detection of Multiple Micro-Drones via Cadence Velocity Diagram Analysis[J]. Electronics Letters, 2018, 54(7):441-443.