网络安全与数据治理2025，Vol.44Issue(4)：10-18,9.DOI:10.19358/j.issn.2097-1788.2025.04.002
两阶段物联网资产识别模型的研究
Research on two-stage IoT asset identification model
权晓文 1汪连连 2韩卫东 2任高峰 2杨家海3
作者信息
- 1. 清华大学 网络科学与网络空间研究院,北京 100084||远江盛邦(北京)网络安全科技股份有限公司,北京 100085
- 2. 远江盛邦(北京)网络安全科技股份有限公司,北京 100085
- 3. 清华大学 网络科学与网络空间研究院,北京 100084
- 折叠
摘要
Abstract
This article proposes a two-stage IoT asset identification model.Firstly,this article analyzes heterogeneous protocols to address the difficulty of extracting traffic characteristics from diverse protocols.Secondly,this article utilizes the lightweight model SqueezeNet to filter non IoT devices,improving recognition efficiency and reducing computational burden.To further address the issue of data imbalance,a Generative Adversarial Network(GAN)is introduced to generate synthetic sample data and balance the data distribution.Finally,this article adopts a model combining XLNet and attention mechanism to identify the types of IoT de-vices,effectively improving the recognition accuracy and efficiency of the model in large-scale IoT network environments.The ex-perimental results show that the accuracy of the model on public datasets reaches 99.48％,the recall rate increases by 2.02％,the Fl score increases by 1.85％,and maintains an accuracy of 99.01％in real environments.This model provides an effective solution for IoT asset management and security management.关键词
物联网/特征选择/设备过滤/资产识别/样本均衡/深度学习Key words
IoT/feature selection/equipment filtering/asset recognition/sample balancing/deep learning分类
信息技术与安全科学引用本文复制引用
权晓文,汪连连,韩卫东,任高峰,杨家海..两阶段物联网资产识别模型的研究[J].网络安全与数据治理,2025,44(4):10-18,9.
