数据与计算发展前沿 ›› 2025, Vol. 7 ›› Issue (6): 13-22.

CSTR: 32002.14.jfdc.CN10-1649/TP.2025.06.002

doi: 10.11871/jfdc.issn.2096-742X.2025.06.002

• 专刊：第40次全国计算机安全学术交流会征文 • 上一篇    下一篇

RVFNet：一种用于击键身份验证的网络模型

王畅1,3(),李沛谕1,林予松1,3,*()   

  1. 1.郑州大学，网络空间安全学院，河南 郑州 450002
    2.郑州大学，计算机与人工智能学院，河南 郑州 450001
    3.郑州大学，河南省教育科研计算机网网络中心，河南 郑州 450052
  • 收稿日期:2025-08-01 出版日期:2025-12-20 发布日期:2025-12-17
  • 通讯作者: 林予松
  • 作者简介:王畅，郑州大学，硕士研究生，主要研究方向为网络与信息安全。
    本文主要负责前期调研与模型设计、算法实现及论文撰写。
    WANG Chang, master’s student at Zhengzhou University, focuses on network and information security.
    In this paper, he is primarily responsible for the preliminary research, model design, algorithm implementation, and manuscript preparation.
    E-mail: 2392547946@qq.com|林予松，郑州大学，教授，博士生导师，主要研究方向为网络与信息安全，医学影像与人工智能。
    本文主要负责研究工作指导。
    LIN Yusong is a professor and Ph.D. supervisor at Zhengzhou University.His research interests include network and information security, medical imaging, and artificial intelligence.
    In this paper, he is primarily responsible for guiding research work.
    E-mail: yslin@ha.edu.cn
  • 基金资助:
    河南省教育厅重点科研基金(23A520019);郑州市协同创新重大专项(20XTZX05015)

RVFNet: A Network Model for Keystroke Authentication

WANG Chang1,3(),LI Peiyu1,LIN Yusong1,3,*()   

  1. 1. School of Cyber Science and Engineering, Zhengzhou University, Zhengzhou, Henan 450002, China
    2. School of Computer Science and Artificial Intelligence, Zhengzhou University, Zhengzhou, Henan 450001, China
    3. Henan Provincial Education and Research Network Center, Zhengzhou University, Zhengzhou, Henan 450052, China
  • Received:2025-08-01 Online:2025-12-20 Published:2025-12-17
  • Contact: LIN Yusong

RichHTML

5

PDF

0

摘要：

【目的】解决了传统击键身份验证方法因忽略全局行为模式而导致的认证准确性不足问题。【方法】首先，通过三通道图像编码将键盘敲击序列转换为视觉表征，充分利用卷积神经网络在图像识别领域的特征提取优势，构建精准的用户身份模型；其次，创新性地设计RVFNet网络架构，并集成自主研发的特征融合增强模块（FEBlock），实现了击键行为的时空特性深度融合。【结果】在Aalto桌面数据集上，当击键序列长度仅设置为50时，RVFNet模型便实现了98.62%的ROC曲线下面积（AUC）以及低至1.46%的等错误率（EER）。【局限】模型参数量较大，可能不适用于资源受限的设备。【结论】实验结果证明了该方法于检测精度方面的有效性。

关键词: 信息安全, 身份验证, 生物特征, 击键动态, 神经网络

Abstract:

[Objective] This paper aims to address the accuracy deficit of traditional keystroke-based authentication caused by overlooking global behavioral patterns. [Methods] First, we convert keystroke sequences into visual representations via three-channel image encoding, leveraging CNNs’ proven image-recognition power to build precise user-identity models. Second, we introduce the novel RVFNet architecture, integrating our proprietary Feature Enhancement Block (FEBlock) to deeply fuse the spatiotemporal dynamics of keystroke behavior. [Results] On the Aalto desktop dataset, RVFNet attains an AUC of 98.62% and an EER as low as 1.46% with only 50 keystrokes. [Limitations] The model’s large parameter count limits deployment on resource-constrained devices. [Conclusions] Experimental results confirm the method’s effectiveness in detection accuracy.

Key words: information security, authentication, biometrics, keystroke dynamics, neural network