通信学报2025，Vol.46Issue(11)：197-213,17.DOI:10.11959/j.issn.1000-436x.2025218
基于协同对抗增强生成模型的智能无人机网络异常检测方法
Anomaly detection method for intelligent UAV networks based on collaborative adversarial enhanced generative model
摘要
Abstract
To address the problem of multi-class imbalance in the anomaly detection of intelligent UAV networks,a col-laborative adversarial enhanced generative model-based anomaly detection method was proposed.For generators,dy-namic class label probability vectors were adopted to gradually increase the probability of minority anormal sample.Moreover,through weight sharing and"fine-tuning"mechanism,the stability and learning efficiency of the multiple gen-erators training were improved.For discriminators and the classifier,encoders with feature aggregation and excitation module were designed.The feature weights were recalibrated,significantly enhancing the model's ability to extract key features in the few-shot learning scenario.For training strategy,a collaborative adversarial mechanism was proposed,in which the classifier and discriminators jointly guide the generation of samples.The distribution bias between the gener-ated samples and the real ones were effectively corrected.A series of comparative experiments were conducted on four open-source datasets,against five baseline methods.Results show that the proposed method increases the F1,AUC and G-mean by 3%,5%and 10%,respectively.The results of the Friedman tests and Nemenyi post-hoc tests also demonstrate that the proposed method exhibits a significant positive difference.关键词
无人机网络/异常检测/生成对抗模型/多类别不平衡Key words
UAV network/anomaly detection/generative adversarial model/multi-class imbalance分类
信息技术与安全科学引用本文复制引用
隋翯,马春燕,龙岭春,顾兆军,刘佳佳,丁磊..基于协同对抗增强生成模型的智能无人机网络异常检测方法[J].通信学报,2025,46(11):197-213,17.基金项目
国家自然科学基金资助项目(No.U2333201)The National Natural Science Foundation of China(No.U2333201) （No.U2333201）
