非约束环境下心电身份识别关键技术研究

The human identification based on electrocardiogram (ECG) is a novel biometric method, with outstanding characteristic of anti-spoof and liveness detection. Comparing to traditional clinical standard electrocardiogram, ECG identification under unconstrained conditions including wearable and portable mobile scenario has more important practical value. The ECG signal under unconstrained conditions is usually with lower signal-to-noise ratio and severe interference. The performance is decreased sharply affected by the individual different posture, movement, emotion and pathology etc. In the project, wearable ECG and finger ECG acquisition system are developed and massive ECG signals are acquired. Patterns of ECG under unconstrained conditions for different position and individual state are explored. Relatively stable characteristics for ECG signal under unconstrained conditions are revealed. The robust ECG denoising method and quality assessment algorithm are proposed. The optimized templates which describe the different state of ECG are created. The identification algorithm based on optimized templates has the robustness to the heart rate variability caused by various physiological and pathological conditions combined with the decision fusion strategy. Refer to the successful application of sparse representation in the computer vision field, sparse coding model of finger ECG is introduced by adopting Structural similarity; Stable templates are acquired by using Max pooling method. Finger ECG real-time identification algorithm for portable mobile scenario completely independent of the fiducials is proposed, which avoids the fiducials detection and have the robustness to the time dependency. The research lays a solid theoretical foundation and technical support for the practical application of ECG biometric method under unconstrained conditions.

心电身份识别是一种新颖的生物识别方法。与临床心电相比，可穿戴场景和便携移动场景等非约束环境下的心电身份识别更具重要的实用价值。非约束环境下心电信噪比低、干扰大，识别性能易受个体姿态、运动、情绪和病理等多种因素的影响而急剧下降。本项目自主设计可穿戴心电和手指心电采集系统，采集大量心电数据，探索非约束环境下心电的特性规律，揭示不同场景下非约束心电用于身份识别的相对稳定特征。在对心电消噪和质量评估等预处理基础上，采用聚类方法构造出表征不同状态心电的优化模板群，对优化模板群的匹配输出进行决策融合实现身份识别，解决可穿戴场景下心电身份识别对个体状态的敏感性问题。引入结构相似度构建手指心电的稀疏编码模型，采用Max pooling方法实现模板固化，无需任何特征点实现身份快速识别，解决便携移动场景手指心电身份识别的实时性和时间稳定性问题。本项目研究为非约束心电身份识别的实用化提供了理论依据和技术支撑。

随着信息化高速发展，个人身份识别技术已经在金融安全、访问控制、医疗、安防监控和资料保密等诸多领域得到了广泛的应用。以心电( Electrocardiogram, ECG)为代表的人体内源生理信号满足了生物识别的高安全性和可靠性要求，得到国内外学者的广泛关注并开展了系列研究，成为该领域的研究热点之一。. 本项目围绕非约束环境下ECG的质量评估、预处理、特征提取和匹配识别，系统研究了非约束环境下（可穿戴式和指尖采集的）心电身份识别的关键技术。在探索非约束环境下ECG和运动伪迹、波形缺失等干扰的基础上，提出了基于简单启发式融合和模糊综合评价的ECG质量评估算法，实现ECG质量的精准评估。结合指尖ECG特性，构建了基于遗传算法的小波阈值指尖心电消噪算法，保持信号特征完整的基础上最大限度地抑制噪声干扰，提高心电身份识别系统的环境适应性和鲁棒性。寻找具有可区分性且稳定的ECG特征是提高系统性能的根本途径之一。本课题首先采用基于非基准点特征提取的策略，提出了基于广义S变换和ZM交叉解析、基于KSVD+PCA下稀疏编码的指尖心电身份识别算法，综合考虑个体采集和识别时间对系统实际应用的制约，优化构造了基于改进的标签一致LC-KSVD的指尖心电身份识别算法和块稀疏分解的心电身份识别算法，缩短了受测者采集时间，获得了较高的识别精度。其次为有效抑制心率变化的影响，提取稳定性强的特征模板，构建了基于卷积神经网络的心电身份识别算法，避免了特征提取的主观性，适用于标准导联采集。综合考虑指尖ECG数据匮乏的情况，为化解模型训练对大数据样本的依赖性，提出了适于指尖采集的基于迁移学习的心电身份识别算法，实现了小样本下的稳定识别。通过对不同类型ECG的分析研究，本课题进而揭示了不同的病理状况和运动状态等对身份识别性能的影响。最后，本课题在开发指尖心电采集系统的基础上，构建了基于Android智能手机的心电身份识别系统，实现个体身份的有效辨识，拓展了心电身份别的应用范围，开发了具有身份识别功能的智能马桶盖。 . 本项目取得的系列研究成果为非约束环境下心电身份识别技术的实用化奠定了坚实的理论基础和技术支撑。