Computer Engineering and Applications ›› 2024, Vol. 60 ›› Issue (3): 280-291.DOI: 10.3778/j.issn.1002-8331.2209-0204

• Network, Communication and Security • Previous Articles     Next Articles

Multi-Scene Continuous Authentication Based on Attention Module for Mobile Devices

JIN Yuyao, ZHANG Xiaomei, WANG Yajie   

  1. School of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai 201620, China
  • Online:2024-02-01 Published:2024-02-01

基于注意力模块的移动设备多场景持续身份认证

金瑜瑶,张晓梅,王亚杰   

  1. 上海工程技术大学 电子电气工程学院,上海 201620

Abstract: In view of the fact that the user may change the scene when interacting with the mobile device, the existing works have limitations on the specific single scene when collecting features and low authentication accuracies, and cannot achieve multi-scene conversion authentication. To overcome these issues,a movement patterns based multi-scene continuous authentication scheme, which combines the attention module with the convolutional neural network (CNN-SACA) is proposed. Under unrestricted usage scenarios and operations, the movement patterns (MP) features are extracted when the user interacts with the mobile device and then hand micro-motion can be captured in dynamic and static scenes, by which the multi-scene authentication is realized. A convolutional neural network including 5 convolutional layers is designed. After the convolution of the first layer, the improved spatial and channel attention sub modules are sequentially passed, and then the weights are inversely distributed after the convolution of the multiple layers to enhance the key feature representation. MP features characterized by the convolution are assigned double attention weights from two dimensions. A public data set is used to verify the effectiveness and feasibility of the proposed method in multi-scene authentication. The experimental results show that the proposed deep learning model based on movement patterns can get over the limitations caused by the single authentication scenario, and achieve accuracy of 99.6%. Meanwhile, comparing with the CNN model alone, the accuracy of the proposed CNN-SACA model is improved by 1.5?percentage points, which effectively improves the authentication capability of mobile devices in multiple scenarios.

Key words: convolutional neural network, attention module, multi-scene, continuous authentication, mobile device

摘要: 针对用户与移动设备交互时会产生场景变化,现有工作中只能采集特定的单一场景特征,无法实现多场景转换认证,并且身份认证准确率较低的问题,提出了一种基于移动模式的、注意力模块和卷积神经网络融合(CNN-SACA)的多场景持续认证方案。在不限使用场景和操作的情况下,提取用户与移动设备交互时的移动模式(movement patterns,MP)特征,捕捉在动态和静态场景下产生的手部微运动,从而实现多场景的身份认证。设计并使用了一个包括五层卷积层结构的卷积神经网络,在第一层卷积后按序通过改进的空间和通道注意力子模块,再在多层卷积后进行反序分配权重,从两个维度来对通过卷积后所表征的MP特征分配双重注意力权重,增强关键特征表达。利用公开数据集验证所提方案在多场景身份认证方面的有效性和可行性。实验结果表明,所提出的基于移动模式的深度学习模型可以较好地解决身份认证场景单一的局限性,多场景的身份认证的准确率达到99.6%;同时,所提出的CNN-SACA模型与单独的CNN模型相比准确率提高了1.5个百分点,有效改善多场景下的移动设备身份认证能力。

关键词: 卷积神经网络, 注意力模块, 多场景, 持续身份认证, 移动设备