基于多尺度卷积和LSTM的多模态隐式认证  

Implicit authentication of multi-modal based on multi-scale convolution and LSTM

作　　者：金瑜瑶 张晓梅[1] JIN Yuyao;ZHANG Xiaomei(School of Electric and Electronic Engineering,Shanghai University of Engineering Science,Shanghai 201620,China)

机构地区：[1]上海工程技术大学电子电气工程学院,上海201620

出　　处：《智能计算机与应用》2023年第12期87-92,共6页Intelligent Computer and Applications

基　　金：国家自然科学基金(61802252)。

摘　　要：针对以往身份认证大多基于单一模态信号、准确率不够高等问题,提出了一种基于多尺度卷积和长短期记忆网络融合的多模态隐式认证方案(MMC-LSTM)。结合智能移动设备在多传感器下的运动特征和触摸特征作为多模态特征进行输入,根据并行的多尺度卷积层捕获多维度的行为特征,并使用长短期记忆网络弥补对短序列识别的不足,从而实现更准确的认证。为减少用户姿态转变带来的影响,构建了先识别不同姿态再进行认证的总体框架。实验结果表明,所提方案在公开数据集上的认证准确率可达到98.2%,比单模态特征认证准确率提高了1.3%,能有效提升身份认证的准确性。A multi-modal implicit authentication scheme(MMC-LSTM)based on the fusion of multi-scale convolution and long and short-term memory neural network is proposed to address the problem that most previous authentication is based on single-modal signals and the accuracy rate is not high enough.Combining the motion features and touch features of smart mobile devices under multiple sensors as multimodal feature inputs,extracting behavior characteristics of different dimensions in parallel based on multi-scale convolution with different-sized convolution kernels,and using long and short-term memory networks to compensate for the lack of recognition of short sequences,a more accurate authentication can be achieved.To reduce the impact of user posture shifts,a general framework of recognizing different postures before authentication is constructed.Experimental results show that the proposed scheme can achieve an authentication accuracy of 98.2%on public datasets,which is 1.3%higher than the accuracy of single-modal feature authentication and can effectively improve the accuracy of identity authentication.

关 键 词：多尺度卷积 LSTM 多模态特征 隐式认证 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]