WO2022027830A1 - 一种基于可穿戴设备的心电图身份认证方法 - Google Patents

一种基于可穿戴设备的心电图身份认证方法 Download PDF

Info

Publication number
WO2022027830A1
WO2022027830A1 PCT/CN2020/122365 CN2020122365W WO2022027830A1 WO 2022027830 A1 WO2022027830 A1 WO 2022027830A1 CN 2020122365 W CN2020122365 W CN 2020122365W WO 2022027830 A1 WO2022027830 A1 WO 2022027830A1
Authority
WO
WIPO (PCT)
Prior art keywords
ecg
feature
wearable device
electrocardiogram
identity authentication
Prior art date
Application number
PCT/CN2020/122365
Other languages
English (en)
French (fr)
Inventor
林峰
Original Assignee
浙江大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 浙江大学 filed Critical 浙江大学
Publication of WO2022027830A1 publication Critical patent/WO2022027830A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
    • G06F21/30—Authentication, i.e. establishing the identity or authorisation of security principals
    • G06F21/31—User authentication
    • G06F21/32—User authentication using biometric data, e.g. fingerprints, iris scans or voiceprints
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/30—Noise filtering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/15—Biometric patterns based on physiological signals, e.g. heartbeat, blood flow

Definitions

  • the invention belongs to the field of biological authentication, and in particular relates to an electrocardiogram identity authentication method based on a wearable device.
  • wearable devices such as smart watches
  • ECG measurement functions such as Apple Watch and Huami Technology's health watch.
  • the purpose of the present invention is to provide an electrocardiogram identity authentication method based on a wearable device in view of the deficiencies of the prior art.
  • a wearable device-based electrocardiogram identity authentication method comprising the following steps:
  • Step 1 The wearable device with the ECG measurement function collects the user's ECG
  • Step 2 Preprocess the ECG collected in Step 1 by applying hybrid three-layer noise reduction
  • Step 3 perform feature point detection on the ECG signal preprocessed in step 3, and locate the position of the feature point;
  • Step 4 Based on the feature points located in Step 3, extract the reference feature and the non-reference feature in the electrocardiogram, and combine the reference feature and the non-reference feature to construct a feature template to form a user database;
  • Step 5 build a class of support vector machine based on the user database obtained in step 4;
  • Step 6 Obtain the ECG feature template of the user to be authenticated according to Steps 1 to 4, and input a type of support vector machine constructed in Step 5 to perform identity authentication.
  • the wearable device includes a smart bracelet, a smart watch and a health wristband.
  • the mixed three-layer noise reduction is in sequence: the first non-local mean noise reduction, the frequency domain filtering method, and the second non-local mean noise reduction.
  • the area length L ⁇ between the target point and the domain point is set to the size of the QRS complex wave area, the size of the search area N(p) is not less than half of the length of the ECG signal, and the smoothing parameter ⁇ No more than half of the standard deviation of the signal noise;
  • L ⁇ is the size of the QRS complex area, the search area size N(p) is not less than half of the length of the ECG signal, and the smoothing parameter ⁇ is the signal noise 30%-50% of the standard deviation.
  • the frequency domain filtering method includes a wavelet denoising method and an ellipse filter.
  • step 3 is specifically: first adopt the Pan-Tumpkins algorithm to detect the position of the QRS complex wave in the electrocardiogram signal through a bandpass filter, a derivative filter, and a square sum integrator, roughly locate the position of the QRS complex wave, and then The positions of the feature points P, Q, R, S and T are precisely located by the local adaptive threshold method.
  • step 4 includes the following sub-steps:
  • step (4.3) extracting non-reference features on the average ECG segment obtained in step (4.2), and using wavelet decomposition to extract the feature information of the ECG in the wavelet domain as non-reference features;
  • step (4.4) Use Fisher algorithm to perform feature selection on the benchmark features obtained in step (4.1) to obtain the most discriminative ECG benchmark features, and use multi-dimensional scaling method to reduce the dimension of non-benchmark features; by adjusting the benchmark features in the feature template The number and the number of non-reference features are used to debug the authentication and recognition effect; the feature templates of all users constitute the user database.
  • step (4.3) is specifically: decompose the average electrocardiogram segment with the wavelet base db5, the number of decomposition layers is 4, and the approximation coefficients of the fourth layer and the detail coefficients of the third and fourth layers are used as non-reference features.
  • the kernel function of a type of support vector machine is set as a Gaussian kernel function, and the abnormal ratio is 0.006.
  • the present invention provides a new noise reduction method and a feature template construction method for the existing electrocardiogram noise reduction problem and feature template construction defects; solves the electrocardiogram noise reduction problem on wearable devices, Protects the state of mind feature of the ECG to improve authentication accuracy.
  • the ratio of reference features and non-reference features is comprehensively adjusted to obtain a highly representative electrocardiogram feature template.
  • the implementation method is simple and flexible, and can ensure high accuracy and low misrecognition rate of the authentication system.
  • Fig. 1 is the schematic diagram of the use scene of the present invention
  • Fig. 2 is the flow chart of the electrocardiogram identity authentication method based on wearable device
  • Fig. 3 is the flow chart of hybrid three-layer noise reduction
  • Figure 4 is a flowchart of feature extraction.
  • a wearable device such as a smart watch with an electrocardiogram measurement function can collect the user's electrocardiogram at any time for real-time authentication, and can also be used as an auxiliary authentication of the smart device to realize functions such as opening the door, unlocking the mobile phone, and paying for authentication.
  • Figure 1 a wearable device such as a smart watch with an electrocardiogram measurement function can collect the user's electrocardiogram at any time for real-time authentication, and can also be used as an auxiliary authentication of the smart device to realize functions such as opening the door, unlocking the mobile phone, and paying for authentication.
  • Figure 2 is a flowchart of an electrocardiogram authentication method based on a wearable device, including the following steps:
  • Step 1 The user wears wearable devices such as smart bracelets, smart watches and health wristbands with ECG measurement function.
  • wearable devices such as smart bracelets, smart watches and health wristbands with ECG measurement function.
  • the user can wear an existing smart watch with ECG measurement function on the wrist. raw ECG data.
  • the user touches the acquisition electrode on the wearable device, and the wearable device has a built-in ECG acquisition chip, which can capture the ECG signal through the weak potential change of the skin on the electrode, and upload the collected raw ECG signal to the terminal or server.
  • Step 2 Applying mixed three-layer noise reduction to denoise the original ECG signal to obtain an effective ECG signal.
  • This step is one of the core steps of the present invention, as shown in Figure 3, which is a flowchart of the mixed three-layer noise reduction, which is divided into the following sub-steps:
  • the collected user's ECG signal contains aliased high-energy low-frequency and high-frequency noise.
  • the mixed three-layer noise reduction is to perform three noise reductions on the original ECG signal.
  • the original ECG signal is denoised; the non-local mean denoising utilizes the non-correlation between the noise signal and the ECG signal, as well as the strong correlation of the ECG signal, to extract the effective ECG signal from the aliased signal.
  • the first non-local mean noise reduction filters out some low-frequency and high-frequency noises, which does not over-smooth the signal and preserves complete details, but still contains a lot of noise.
  • ⁇ is a smoothing parameter, which controls the degree of signal smoothing. The larger the ⁇ , the smoother the change of the Gaussian function and the higher the noise reduction level, but it is easy to cause excessive smoothing. The smaller the ⁇ , the more details will be retained, but there will be too much residue.
  • the first non-local mean noise reduction parameter should be set to moderate to prevent excessive smoothing. Therefore, it is necessary to control the ⁇ parameter not to exceed half of the signal noise standard deviation ⁇ to ensure that the first noise reduction will not lose details.
  • the area length L ⁇ between the target point and the field point is set to the size of the similar area of the electrocardiogram, that is, the size of the QRS complex area.
  • N(p) determines the size of the search area, and N(p) is set to be no less than half of the length of the ECG signal in order to find more similar morphological areas.
  • the second noise reduction adopts the frequency domain filtering method.
  • the high-frequency noise of the ECG can be filtered out by the frequency domain filtering method, and the key low-frequency information of the ECG can be protected; the detailed information of the ECG is filtered out by the wavelet denoising method, and then an elliptic low-pass filter is used to further filter out the high-frequency noise;
  • the ECG signal after domain filtering still contains low-frequency noise, but the detailed features are completely preserved.
  • the second non-local mean noise reduction removes low-frequency noise, and the details of the ECG are preserved.
  • Step 3 Locate each feature point in the electrocardiogram after noise reduction.
  • the Pan-Tumpkins algorithm and the local adaptive threshold method are combined to locate the ECG feature points.
  • the Pan-Tumpkins algorithm detects the position of the QRS complex through a band-pass filter, a derivative filter, and a sum-of-square integrator, roughly locates the position of the QRS complex, and then precisely locates the feature point position through the local adaptive threshold method.
  • Step 4 This step is one of the core steps of the present invention. Based on the feature points located in step 3, the reference feature and non-reference feature of the electrocardiogram are extracted, and the feature template corresponding to the input electrocardiogram is constructed in combination with the reference feature and the non-reference feature; as shown in the figure. 4, Figure 4 is a flowchart of feature extraction, and the feature template construction method is divided into the following sub-steps:
  • the feature points of the ECG are used as reference points to extract reference features, and the non-reference features are extracted based on the average ECG segment;
  • the feature points in each ECG segment are located, and the time interval and amplitude difference between the ECG feature points P, Q, R, S, and T points are calculated as reference features.
  • a total of 9 time interval features and 10 amplitude features are extracted (the RQ amplitude feature is eliminated), as shown in Table 1 for details.
  • Table 1 9 time interval features and 10 amplitude features of ECG
  • the average ECG segment Before extracting non-benchmark features, the average ECG segment must be calculated. From a series of ECG signals, according to the position of the feature point R, locate each ECG position, and then take the feature point R as the center to intercept a cycle of ECG segments, and calculate the mean of all the intercepted ECG segments to obtain the average ECG segment.
  • Non-reference features are extracted from the average ECG segment, and wavelet decomposition is used to extract the feature information of the ECG in the wavelet domain.
  • the average ECG segment is decomposed with the wavelet base db5, and the number of decomposition layers is 4.
  • the approximate coefficients of the fourth layer and The detail coefficients of layers 3 and 4 are used as non-reference features.
  • Feature selection is performed on the reference features
  • feature dimension reduction is performed on the non-reference features
  • the number of reference features and the number of non-reference features are adjusted to construct a feature template.
  • the Fisher algorithm is used to select the benchmark features to select the most discriminative features; the multi-dimensional scaling method is used to reduce the dimension of the high-dimensional non-benchmark features to extract the core representative non-benchmark features, reduce feature redundancy, prevent High-dimensional catastrophe.
  • the ECG data provided by the user construct a user feature template with different combinations of benchmark features and non-benchmark features.
  • the user feature template can test the recall rate and misidentification of users. The best comprehensive performance is obtained when the rate is reached, the most representative user characteristic template is obtained, and the user database is established.
  • Feature selection and dimensionality reduction are performed on benchmark features and non-benchmark features respectively, which can retain the characteristics of their respective soft and hard indicators, and can construct a complete user feature template; the user feature template constructed based on benchmark features and non-benchmark features not only retains the ECG
  • the detailed features also store the overall morphological features of the ECG.
  • Step 5 Construct a type of support vector machine for identity authentication according to the user database obtained in step 4.
  • a type of support vector machine is used to construct a hyperplane based on the user database to distinguish users from non-users. If the input features are within the hyperplane, it is a user, otherwise it is a non-user; the kernel function of the support vector machine is set is a Gaussian kernel function, and the anomaly ratio is 0.006.
  • Step 6 When performing identity authentication, obtain the ECG feature template of the user to be authenticated according to steps 1 to 4, and input a type of support vector machine constructed in step 5 to perform identity authentication; this method is simple in calculation, fast in operation, and suitable for transplantation in various platform.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Security & Cryptography (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Hardware Design (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Human Computer Interaction (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)

Abstract

一种基于可穿戴设备的心电图身份认证方法，该方法对用户输入的心电图进行预处理，特征点检测与特征提取，被提取的心电图特征与用户数据库中的特征模板相匹配比较。针对可穿戴设备上的心电图降噪问题，该方法设计了混合三层降噪可以有效去除噪声信号，混合三层降噪结合了形态学滤波和频域滤波的优势，其实现方法简便，滤波后获得心电图的信号质量能得到显著保证。该方法构造的用户特征模板保留了心电图的基准特征与非基准特征各自的硬指标与软指标特性，保证了***的认证准确率。

Description

一种基于可穿戴设备的心电图身份认证方法 技术领域
本发明属于生物认证领域，尤其涉及一种基于可穿戴设备的心电图身份认证方法。
背景技术
近年来，可穿戴设备(比如智能手表)广泛应用在移动支付，移动通信，健康信息监测，智能家居交互等领域中。目前已存在具有心电图测量功能的智能手表，比如Apple Watch和华米科技的健康手表。
以往，心电图研究中的研究数据均由抗干扰性高，精度高的心电图机所采集，这类心电图数据信噪比高，没有高能量的低频噪声与高频噪声。因此，之前的心电图识别与认证研究仅用一次带通滤波即可去除心电图中的噪声。然而，这种频域滤波法不适于去除由可穿戴设备所采集的心电图中的噪声。可穿戴设备所采集的心电图信噪比较低，高能量的噪声频段覆盖了心电图关键的低频区域，频域滤波无法有效地从信号与噪声混叠的频段内提取出代表心电图信号的频段。另一方面，在构造用户特征模板时，以往研究中常忽略心电图特征的特性而直接构造特征模板的做法，很少有研究将这两类特征区分处理，更没有调整两者在用户特征模板中的比重，这样构造的特征模板不能充分代表用户心电图信息。
发明内容
本发明的目的在于针对现有技术的不足，提供一种基于可穿戴设备的心电图身份认证方法。
本发明的目的是通过以下技术方案来实现的：一种基于可穿戴设备的心电图身份认证方法，包括以下步骤：
步骤一：具有心电图测量功能的可穿戴设备采集用户的心电图；
步骤二：应用混合三层降噪对步骤一采集的心电图进行预处理；
步骤三：对步骤三预处理后的心电图信号进行特征点检测，定位特征点位置；
步骤四：基于步骤三定位的特征点，提取心电图中的基准特征与非基准特征，结合基准特征与非基准特征构造特征模板，形成用户数据库；
步骤五：基于步骤四得到的用户数据库构建一类支持向量机；
步骤六：根据步骤一至四得到待身份认证用户的心电图特征模板，输入步骤五构建的一类支持向量机，进行身份认证。
进一步地，所述步骤一中，可穿戴设备包括智能手环、智能手表和健康腕带。
进一步地，所述步骤二中，混合三层降噪依次为第一次非局部均值降噪、频域滤波法、第二次非局部均值降噪。
进一步地，第一次非局部均值降噪中，目标点与领域点的区域长度L Δ设置为QRS复合波区域的大小，搜索区域大小N(p)不小于心电图信号长度的一半，平滑参数λ不超过信号噪声标准差的一半；第二次非局部均值降噪中，L Δ为QRS复合波区域的大小，搜索区域大小N(p)不小于心电图信号长度的一半，平滑参数λ为信号噪声标准差的30％-50％。
进一步地，所述频域滤波法包括小波去噪法和椭圆滤波器。
进一步地，所述步骤三具体为：首先采用Pan-Tompkins算法通过带通滤波器、导数滤波器、平方和积分器来检测心电图信号中QRS复合波的位置，粗略定位QRS复合波的位置，然后通过局部自适应阈值法精确定位特征点P，Q、R、S和T点的位置。
进一步地，所述步骤四包括以下子步骤：
(4.1)计算心电图特征点P，Q、R、S和T点之间的时间间隔和幅度差作为基准特征；
(4.2)从心电图信号中，以特征点R为中心截取一个周期的心电图片段，计算所有被截取的心电图片段的均值得到平均心电图片段；
(4.3)在步骤(4.2)得到的平均心电图片段上提取非基准特征，采用小波分解以提取心电图在小波域的特征信息作为非基准特征；
(4.4)采用Fisher算法对步骤(4.1)得到的基准特征进行特征选择，获得最具区分度的心电图基准特征，采用多维标度法对非基准特征进行降维；通过调整特征模板中的基准特征数目与非基准特征数目调试身份认证识别效果；所有用户的特征模板构成用户数据库。
进一步地，所述步骤(4.3)具体为：用小波基db5分解平均心电图片段，分解层数为4层，将第4层的近似系数和第3、4层的细节系数作为非基准特征。
进一步地，所述步骤五中，一类支持向量机的核函数设为高斯核函数，异常比例为0.006。
本发明的有益效果是：本发明针对现有存在的心电图降噪问题和特征模板构造缺陷，提供一种新的降噪方式以及特征模板构造方法；解决了可穿戴设备上的心电图降噪问题，保护心电图的心态特征以提高认证准确率。综合调整基准特征与非基准特征的比例以获得代表性强的心电图特征模板，其实现方法简单，手段灵活，能够保证认证***的高准确率与低误识率。
附图说明
图1是本发明使用情景示意图；
图2是基于可穿戴设备的心电图身份认证方法流程图；
图3是混合三层降噪的流程图；
图4是特征提取的流程图。
具体实施方式
为了更进一步阐述本发明所采取的技术手段及取得的效果，下面根据附图详细说明本发明。
本发明提出用具有心电图测量功能的智能手表等可穿戴设备可随时采集用户心电图以实时认证，也可同时作为智能设备的辅助认证以实现开启门禁，手机解锁，支付认证等功能，具体情境图如图1所示。
如图2所示，图2为基于可穿戴设备的心电图身份认证方法的流程图，包括以下步骤：
步骤一：用户佩戴具有心电图测量功能的智能手环，智能手表和健康腕带等可穿戴设备，比如用户可以在手腕处佩戴市面上已有的具有心电图测量功能的智能手表，可穿戴设备采集用户的原始心电图数据。
在本步骤中，用户接触可穿戴设备上的采集电极，可穿戴设备内置心电图采集芯片，可通过电极上皮肤出微弱的电势变化捕捉到心电图信号，将采集得到的原始心电图信号上传终端或者服务器。
步骤二：应用混合三层降噪对原始的心电图信号进行降噪，得到有效的心电图信号。
本步骤是本发明的核心步骤之一，如图3所示，图3为混合三层降噪的流程图，分为以下子步骤：
1)采集到的用户的心电图信号含有混叠的高能量的低频与高频噪声，混合三层降噪即对原始心电图信号进行三次降噪，第一次降噪先用非局部均值降噪对原始心电图信号进行降噪；非局部均值降噪利用噪声信号与心电图信号的非相关性，以及心电图信号的强相关性，将有效的心电图信号从混叠的信号中提取出来。
第一次非局部均值降噪滤除部分的低频与高频噪声，既不过度平滑信号又能保存完整的细节特征，但依旧含有较多的噪声。第一次非局部均值降噪的各个参数分别设置为：λ＝0.5*0.08，L Δ＝10和N(p)＝3000。λ是平滑参数，控制信号平滑后的程度，λ越大则高斯函数变化越平缓，降噪水平越高，但也易导致过度平滑，λ越小则会保留更多细节，但会残留过多的噪声数据，平滑效果不够明显；λ的取值与信号噪声有关，平滑参数λ与噪声标准差σ成正相关，即λ＝kσ；当信号噪声标准差过大，需要一个较大的λ去平滑噪声，如果信号噪声标准差过小，则需要较小的λ去平滑噪声。第一次非局部均值降噪参数应该设置的缓和以防过度平滑，因此需要控制λ参数不超过信号噪声标准差σ的一半，保证第一次降噪不会丢失细节特征。目标点与领域点的区域长度L Δ设置为心电图相似区域大小即QRS复合波区域的大小。N(p)决定了搜索区域大小，N(p)设置为不小于心电图信号长度的一半，以便寻 找更多的相似形态区域。
2)第二次降噪采用频域滤波法。
通过频域滤波法可滤除心电图的高频噪声，保护心电图的关键低频信息；先通过小波去噪法滤除心电图的细节信息，再用一个椭圆低通滤波器进一步滤除高频噪声；频域滤波后的心电图信号依旧含有低频噪声，但是细节特征保存完整。
3)第三次降噪继续再一次采用非局部均值降噪，获得干净有效的心电图信号。
第二次非局部均值降噪去除低频噪声，心电图的细节特征保存完整，第二次非局部均值降噪的各个参数分别设置为：λ＝0.4*0.03，L Δ＝10和N(p)＝3000，L Δ依旧设置为心电图相似区域大小即QRS复合波区域的大小，N(p)仍然设置为不小于心电图信号长度的一半，以便寻找更多的相似形态区域。因为经过两次去噪后，心电图所含的噪声较少，因此λ参数也不需要过大，λ参数为信号噪声标准差σ的30％-50％。
这样做可以既去除混叠的心电图噪声，还能有效保存心电图的形态特征。经过三次降噪后的心电图信号不仅噪声信号被很好地滤除，而且心电图细节特征均被完整保存。
步骤三：定位降噪后的心电图中的各个特征点。
在本步骤中，结合使用Pan-Tompkins算法与局部自适应阈值法以定位心电图特征点。首先Pan-Tompkins算法通过带通滤波器、导数滤波器、平方和积分器来检测QRS复合波的位置，粗略定位QRS复合波的位置，然后通过局部自适应阈值法精确定位特征点位置。
步骤四：本步骤是本发明的核心步骤之一，基于步骤三中定位的特征点，提取心电图的基准特征与非基准特征，结合基准特征与非基准特征构造输入心电图对应的特征模板；如图4所示，图4为特征提取的流程图，特征模板构造方法分为以下子步骤：
1)将心电图的特征点作为基准点提取基准特征，基于平均心电图片段提取非基准特征；
经过上一步特征点检测后，各个心电图片段中的特征点均被定位，计算心电图特征点P，Q、R、S和T点之间的时间间隔和幅度差作为基准特征。本发明中一共提取了9个时间间隔特征和10种幅度特征(RQ幅度特征被剔除)，具体见表1。
表1：心电图的9个时间间隔特征和10种幅度特征
Figure PCTCN2020122365-appb-000001
提取非基准特征前，须计算平均心电图片段。从一系列心电图信号中，根据特征点R的 位置，定位出一个个心电图位置，再以特征点R为中心截取一个周期的心电图片段，计算所有被截取的心电图的均值即可得到平均心电图片段。
在平均心电图片段上提取非基准特征，采用小波分解以提取心电图在小波域的特征信息，具体为：用小波基db5分解平均心电图片段，分解层数为4层，将第4层的近似系数和第3，4层的细节系数作为非基准特征。
2)对基准特征进行特征选择，对非基准特征进行特征降维，调整基准特征数目与非基准特征数目以构造特征模板。
采用Fisher算法对基准特征进行特征选择以挑选最具区分度的特征；采用多维标度法对高维的非基准特征进行降维以提取核心代表性的非基准特征，降低特征冗余度，防止高维灾难。根据用户提供的心电图数据，构造含有不同基准特征与非基准特征组合搭配的用户特征模板，通过调整特征模板中的基准特征数目与非基准特征数目以使用户特征模板在测试用户召回率和误识率时获得综合最佳的性能，得到最具代表性的用户特征模板，建立用户数据库。分别对基准特征和非基准特征进行特征选择与降维，可以保留各自的软，硬指标的特性，可以构造完整的用户特征模板；基于基准特征和非基准特征构造的用户特征模板，既保留心电图的细节特征，又存储了心电图的整体形态特征。
步骤五：根据步骤四得到的用户数据库构造身份认证的一类支持向量机。
在本步骤中，采用一类支持向量机基于用户数据库构造一个超平面来区分用户与非用户，如果输入的特征在超平面以内，说明是用户，否则就是非用户；支持向量机的核函数设为高斯核函数，异常比例为0.006。
步骤六：进行身份认证时，根据步骤一至四得到待身份认证用户的心电图特征模板，输入步骤五构建的一类支持向量机，进行身份认证；该方法计算简单，运行快速，适于移植在各个平台。

Claims (9)

  1. 一种基于可穿戴设备的心电图身份认证方法，其特征在于，包括以下步骤：
    步骤一：具有心电图测量功能的可穿戴设备采集用户的心电图；
    步骤二：应用混合三层降噪对步骤一采集的心电图进行预处理；
    步骤三：对步骤三预处理后的心电图信号进行特征点检测，定位特征点位置；
    步骤四：基于步骤三定位的特征点，提取心电图中的基准特征与非基准特征，结合基准特征与非基准特征构造特征模板，形成用户数据库；
    步骤五：基于步骤四得到的用户数据库构建一类支持向量机；
    步骤六：根据步骤一至四得到待身份认证用户的心电图特征模板，输入步骤五构建的一类支持向量机，进行身份认证。
  2. 根据权利要求1所述基于可穿戴设备的心电图身份认证方法，其特征在于，所述步骤一中，可穿戴设备包括智能手环、智能手表和健康腕带。
  3. 根据权利要求1所述基于可穿戴设备的心电图身份认证方法，其特征在于，所述步骤二中，混合三层降噪依次为第一次非局部均值降噪、频域滤波法、第二次非局部均值降噪。
  4. 根据权利要求3所述基于可穿戴设备的心电图身份认证方法，其特征在于，第一次非局部均值降噪中，目标点与领域点的区域长度L Δ设置为QRS复合波区域的大小，搜索区域大小N(p)不小于心电图信号长度的一半，平滑参数λ不超过信号噪声标准差的一半；第二次非局部均值降噪中，L Δ为QRS复合波区域的大小，搜索区域大小N(p)不小于心电图信号长度的一半，平滑参数λ为信号噪声标准差的30％-50％。
  5. 根据权利要求3所述基于可穿戴设备的心电图身份认证方法，其特征在于，所述频域滤波法包括小波去噪法和椭圆滤波器。
  6. 根据权利要求1所述基于可穿戴设备的心电图身份认证方法，其特征在于，所述步骤三具体为：首先采用Pan-Tompkins算法通过带通滤波器、导数滤波器、平方和积分器来检测心电图信号中QRS复合波的位置，粗略定位QRS复合波的位置，然后通过局部自适应阈值法精确定位特征点P，Q、R、S和T点的位置。
  7. 根据权利要求1所述基于可穿戴设备的心电图身份认证方法，其特征在于，所述步骤四包括以下子步骤：
    (4.1)计算心电图特征点P，Q、R、S和T点之间的时间间隔和幅度差作为基准特征；
    (4.2)从心电图信号中，以特征点R为中心截取一个周期的心电图片段，计算所有被截取的心电图片段的均值得到平均心电图片段；
    (4.3)在步骤(4.2)得到的平均心电图片段上提取非基准特征，采用小波分解以提取心电图在小波域的特征信息作为非基准特征；
    (4.4)采用Fisher算法对步骤(4.1)得到的基准特征进行特征选择，获得最具区分度的心电图基准特征，采用多维标度法对非基准特征进行降维；通过调整特征模板中的基准特征数目与非基准特征数目调试身份认证识别效果；所有用户的特征模板构成用户数据库。
  8. 根据权利要求7所述基于可穿戴设备的心电图身份认证方法，其特征在于，所述步骤(4.3)具体为：用小波基db5分解平均心电图片段，分解层数为4层，将第4层的近似系数和第3、4层的细节系数作为非基准特征。
  9. 根据权利要求1所述基于可穿戴设备的心电图身份认证方法，其特征在于，所述步骤五中，一类支持向量机的核函数设为高斯核函数，异常比例为0.006。
PCT/CN2020/122365 2020-08-07 2020-10-21 一种基于可穿戴设备的心电图身份认证方法 WO2022027830A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010788412.1A CN112016405B (zh) 2020-08-07 2020-08-07 一种基于可穿戴设备的心电图身份认证方法
CN202010788412.1 2020-08-07

Publications (1)

Publication Number Publication Date
WO2022027830A1 true WO2022027830A1 (zh) 2022-02-10

Family

ID=73500174

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/122365 WO2022027830A1 (zh) 2020-08-07 2020-10-21 一种基于可穿戴设备的心电图身份认证方法

Country Status (2)

Country Link
CN (1) CN112016405B (zh)
WO (1) WO2022027830A1 (zh)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102940485A (zh) * 2012-10-10 2013-02-27 广州三瑞医疗器械有限公司 一种胎心率基线自动识别算法
CN104573458A (zh) * 2014-12-30 2015-04-29 深圳先进技术研究院 一种基于心电信号的身份识别方法、装置及***
CN105748066A (zh) * 2016-03-03 2016-07-13 深圳竹信科技有限公司 一种心电信号波形特征点的提取方法及装置
CN107980151A (zh) * 2017-02-22 2018-05-01 清华大学深圳研究生院 一种基于心电认证的门禁***及其认证方法

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101773394B (zh) * 2010-01-06 2011-09-07 中国航天员科研训练中心 身份识别方法及应用该方法的身份识别***
US9364160B2 (en) * 2012-03-28 2016-06-14 Qualcomm Incorporated Systems and methods for ECG monitoring
CN104102915B (zh) * 2014-07-01 2019-02-22 清华大学深圳研究生院 一种心电异常状态下基于ecg多模板匹配的身份识别方法
CN104706350B (zh) * 2014-11-13 2017-04-26 华中科技大学 一种心电信号降噪方法
US10885361B2 (en) * 2016-04-29 2021-01-05 Inesc Tec—Instituto De Engenharia De Sistemas E Computadores, Technologia E Ciena Biometric method and device for identifying a person through an electrocardiogram (ECG) waveform

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102940485A (zh) * 2012-10-10 2013-02-27 广州三瑞医疗器械有限公司 一种胎心率基线自动识别算法
CN104573458A (zh) * 2014-12-30 2015-04-29 深圳先进技术研究院 一种基于心电信号的身份识别方法、装置及***
CN105748066A (zh) * 2016-03-03 2016-07-13 深圳竹信科技有限公司 一种心电信号波形特征点的提取方法及装置
CN107980151A (zh) * 2017-02-22 2018-05-01 清华大学深圳研究生院 一种基于心电认证的门禁***及其认证方法

Also Published As

Publication number Publication date
CN112016405A (zh) 2020-12-01
CN112016405B (zh) 2022-07-15

Similar Documents

Publication Publication Date Title
Sabherwal et al. Automatic detection of the R peaks in single-lead ECG signal
WO2016025961A1 (en) Herma - heartbeat microwave authentication
CN103345600A (zh) 一种心电信号数据处理方法
CN104305992B (zh) 一种交互式胎儿心电快速自动提取方法
CN105942974A (zh) 一种基于低频脑电的睡眠分析方法及***
Chawla et al. RETRACTED: A new statistical PCA–ICA algorithm for location of R-peaks in ECG
CN107184187B (zh) 基于DTCWT-Spline的脉搏波信号去噪处理方法
Zou et al. An ultra-low power QRS complex detection algorithm based on down-sampling wavelet transform
CN106485208A (zh) 单通道脑电信号中眼电干扰的自动去除方法
CN109506938A (zh) 基于改进小波包阈值去噪与局部均值分解的故障提取方法
Sun Classification of imagery motor EEG data with wavelet denoising and features selection
Verma et al. An improved algorithm for noise suppression and baseline correction of ECG signals
Yang et al. Automated intra-patient and inter-patient coronary artery disease and congestive heart failure detection using EFAP-Net
Chen et al. Finger ECG based two-phase authentication using 1D convolutional neural networks
CN108717535B (zh) 一种基于混合特征和长短时记忆网络的麻醉深度估计方法
WO2022027830A1 (zh) 一种基于可穿戴设备的心电图身份认证方法
NL2025147B1 (en) Method for Correcting Baseline Drift of Dynamic ECG Signal with Sudden Jump Noise
CN113476074A (zh) 基于智能穿戴***的肺音实时监测方法
Hu et al. An approach to QRS complex detection based on multiscale mathematical morphology
CN106097264B (zh) 基于双树复小波和形态学的卫星遥测数据滤波方法
CN114757236A (zh) 基于tqwt与svmd的脑电信号去噪优化方法及***
Zhong et al. A tree-search method for single-channel fetal QRS complexes detection in fetal heart rate monitoring
CN113159205A (zh) 一种基于最佳通道的稀疏时频块共空间模式特征提取方法
CN113160788A (zh) 基于波形分析的反相降噪平台
Pin et al. Circuit architectures reviews for portable ECG signal analyzer

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20948455

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20948455

Country of ref document: EP

Kind code of ref document: A1