CN105286846A - Movement noise detection method suitable for heart rate signals - Google Patents

Abstract

The invention discloses a movement noise detection method suitable for heart rate signals. The method aims at improving the heart rate measurement accuracy of a wearable heart rate measurement device. In the method, the wearable heart rate measurement device collects multiple pulse oximeter signals and movement acceleration signals, during the same time period, of a user; a frequency spectrum matrix is formed by the pulse oximeter signals and the movement acceleration signals, and a combined sparse spectrum reconstitution model is established by extracting the overall sparse and line sparse structure characteristics in the frequency spectrum matrix; then, a sparse frequency spectrum matrix in the combined sparse reconstitution model is calculated through the inaccurate augmented lagrangian multiplier method, and calculated signals has the advantage that the spectrum peak positions of movement acceleration signal frequency spectrums are basically the same as the spectrum peak positions of the pulse oximeter signals. By means of the method, strong movement noise in heart rate signals can be accurately detected, and the theoretical foundation is laid for effectively removing strong movement noise in heart rate signals.

Description

Motion noise detection method suitable for heart rate signals

Technical Field

The invention relates to the field of information processing, in particular to a motion noise detection method suitable for heart rate signals.

Background

Photoplethysmography (PPG) is a non-invasive method of detecting changes in blood volume in living tissue by electro-optical means, in which a light beam of a certain wavelength is transmitted or reflected to a photoreceiver when it strikes the skin surface. In the process, the light intensity detected by the detector is weakened due to absorption attenuation of skin muscle and blood, wherein the absorption of light by skin, muscle, tissue and the like is constant in the whole blood circulation, and the blood volume in the skin is pulsated and changed under the action of the heart. So that the light intensity detected by the photoelectric receiver pulsates. The signal of the light intensity change is converted into an electric signal, namely a photoelectric volume pulse wave signal, so that the change of the volume pulse blood flow can be obtained.

Since the photoplethysmography signal is a biological signal extracted from the skin surface, it has a weak signal intensity and is easily interfered by noise. For example, in sports, there is motion noise interference due to tissue interference, venous blood volume, and optical path variation, and the frequency of motion noise is in many cases very close to that of heart rate, so it is most difficult to detect motion noise signals from heart rate.

The invention provides a motion noise detection method suitable for a heart rate signal. According to the method, a plurality of photoplethysmographic signals and motion acceleration signals form a frequency spectrum matrix, and the frequency position of motion noise in a plurality of photoplethysmographic signal frequency spectrums can be aligned with the frequency position of the acceleration signal frequency spectrum better due to the sparse row structural characteristics in the frequency spectrum matrix. The invention can accurately detect the motion noise in the heart rate signal and lays a theoretical foundation for effectively removing strong motion noise in the heart rate signal.

Disclosure of Invention

The invention aims to solve the technical problem of how to provide a method for accurately detecting motion noise in a heart rate signal under the condition of very strong motion noise, and lays a foundation for improving the heart rate measurement value precision of wearable heart rate measurement equipment.

In order to solve the technical problem, the invention provides a motion noise detection method suitable for heart rate signals, which comprises three parts of signal acquisition, joint sparse spectrum reconstruction model construction and sparse spectrum matrix solving, and is characterized in that:

the wearable heart rate measuring device collects a plurality of photoplethysmographic signals and motion acceleration signals in the same time period at the wrist of a user; and forming a frequency spectrum matrix by using the plurality of photoplethysmographic signals and the motion acceleration signals, extracting the structural characteristics of the frequency spectrum matrix, establishing the joint sparse spectrum reconstruction model, and solving the sparse frequency spectrum matrix in the joint sparse spectrum reconstruction model.

The method comprises the following steps:

the wearable heart rate measuring equipment acquires a plurality of photoplethysmographic pulse wave signals and motion acceleration signals of a user in the same time period; down-sampling the plurality of photoplethysmographic signals and the motion acceleration signals; and then performing band-pass filtering operation on the down-sampled signals.

Meanwhile, a frequency spectrum matrix is formed by the plurality of photoplethysmographic signals and the motion acceleration signals, the structural characteristics of integral sparseness and row sparseness in the frequency spectrum matrix are extracted to establish the combined sparse spectrum reconstruction model, the sparse frequency spectrum matrix in the combined sparse spectrum reconstruction model is solved through an inaccurate augmented Lagrange multiplier method, and the solved signals have the characteristic that the spectral peak positions of the motion acceleration signal frequency spectrum are basically the same as those of the plurality of photoplethysmographic signal frequency spectrums.

Preferably, a plurality of photoplethysmography sensors and a three-axis accelerometer are embedded in the wearable heart rate measurement device; the plurality of photoplethysmography sensors collect a plurality of photoplethysmography signals of a user; the three-axis accelerometer collects motion acceleration signals of a user in the same time period.

Preferably, the joint sparse spectrum reconstruction model is constructed according to a spectrum matrix formed by the plurality of photoplethysmographic signals and the motion acceleration signals, and has structural features of global sparsity and row sparsity, and an objective function of the joint sparse spectrum reconstruction model is as follows:

$\underset{X,V}{min}{\mathrm{\λ}}_1\left|\right|V|{|}_F^2+{\mathrm{\λ}}_2|\left|X\right|{|}_{1,2}+{\mathrm{\λ}}_3\left|\right|X|{|}_{1,1}$

$s.t\left\{\begin{array}{c}\left|\right|V|{|}_F=\sqrt{tr\left(V^{T}V\right)}=\sqrt{\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{R}{\mathrm{\Σ}}}{v_{i,j}}^2}\\ \left|\right|X|{|}_{1,2}=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\left(\underset{j=1}{\overset{R}{\mathrm{\Σ}}}{x_{i,j}}^2\right)}^{\frac{1}{2}}\\ \left|\right|X|{|}_{1,1}=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\underset{j=1}{\overset{R}{\mathrm{\Σ}}}\left|x_{i,j}\right|\\ Y=\mathrm{\Φ}X+V\end{array}\right.$

wherein,

||V||_Ffor constraining the error matrix V to minimize the error, | X | | | non-calculation_1，1Used for constraining the global sparsity of the spectrum matrix, | X | | non-woven phosphor_1，2For constraining the row sparsity, x, of the spectral matrix_i，jIs the ith row and jth column element, v, of the frequency spectrum matrix X_i，jIs the ith row and jth column element, λ, of the error matrix V₁、λ₂、λ₃Weight is used to weigh the importance of each item, Y is phi X + V is the constraint condition of equation, Y ∈ R^M×HIs an observation matrix, X ∈ C^N×HIs the spectrum matrix of the corresponding signal, i.e. the sparse spectrum matrix to be solved, phi ∈ C^M×N(M < N) is a redundant discrete Fourier transform basis and V is a model error or measurement error matrix.

Preferably, the inaccurate augmented lagrangian multiplier method is an iterative method which is more suitable for the joint sparse spectrum reconstruction model target function and is constructed by combining a penalty function with a traditional lagrangian function; while the inexact augmented lagrange multiplier method allows updating unknown variables in an alternating or sequential manner with a rate of quadratic convergence.

Compared with the prior art, the technical scheme provided by the invention can accurately detect the strong motion noise in the heart rate signal, and lays a theoretical foundation for effectively removing the strong motion noise in the heart rate signal, so that the heart rate measurement value precision of the wearable heart rate measurement equipment can be improved.

Drawings

Fig. 1 is a flowchart illustrating a motion noise detection method according to an embodiment of the present invention.

Detailed Description

The following describes embodiments of the present invention in detail with reference to the accompanying drawings and examples, so that how to apply technical means to solve the technical problems and achieve the corresponding technical effects can be fully understood and implemented.

According to the technical scheme, the plurality of photoplethysmographic signals and the motion acceleration signals form a frequency spectrum matrix, and the frequency position of motion noise in the frequency spectrum of the plurality of photoplethysmographic signals can be aligned with the frequency position of the frequency spectrum of the acceleration signals better due to the sparse row structural characteristics in the frequency spectrum matrix. The technical scheme accurately detects the motion noise in the heart rate signal and lays a theoretical foundation for effectively removing strong motion noise in the heart rate signal.

Embodiment I, detection method of motion noise in heart rate signal

Fig. 1 is a flowchart illustrating a motion noise detection method according to this embodiment.

The embodiment shown in fig. 1 is an overall flow of a method for detecting motion noise in a heart rate signal, and mainly includes the following steps:

step S210, the wearable heart rate measuring device acquires photoplethysmographic signals of two channels by using two photoplethysmographic sensors distributed at different positions, and acquires motion acceleration signals of three channels in the same time period by using a three-axis accelerometer.

In step S220, the initial sampling frequency of the original signal is 125Hz, and in order to reduce the amount of calculation, the original signal needs to be down-sampled to a sampling frequency of 25 Hz.

In step S230, the down-sampled signals need to be filtered by a second-order butterworth filter with a passband of 0.4Hz to 4Hz to eliminate the interference of motion noise and other noise outside a certain frequency range.

In step S240, a spectrum matrix is formed by the two photoplethysmographic signals and the three motion acceleration signals.

And step S250, extracting the structural characteristics of global sparsity and row sparsity in the frequency spectrum matrix, and then constructing a combined sparse spectrum reconstruction model by using the structural characteristics.

In this step, typically, formula (1) is an objective function of the joint sparse spectrum reconstruction model:

$\underset{X,V}{min}{\mathrm{\&lambda;}}_1\left|\right|V|{|}_F^2+{\mathrm{\&lambda;}}_2|\left|X\right|{|}_{1,2}+{\mathrm{\&lambda;}}_3\left|\right|X|{|}_{1,1}$

$s.t\left\{\begin{array}{c}\left|\right|V|{|}_F=\sqrt{tr\left(V^{T}V\right)}=\sqrt{\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{R}{\mathrm{\&Sigma;}}}{v_{i,j}}^2}\\ \left|\right|X|{|}_{1,2}=\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{\left(\underset{j=1}{\overset{R}{\mathrm{\&Sigma;}}}{x_{i,j}}^2\right)}^{\frac{1}{2}}\\ \left|\right|X|{|}_{1,1}=\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{R}{\mathrm{\&Sigma;}}}\left|x_{i,j}\right|\\ Y=\mathrm{\&Phi;}X+V\end{array}\right.---\left(2\right)$

wherein | V | Y luminance_FFor constraining the error matrix V to minimize the error, | X | | | non-calculation_1，1Used for constraining the global sparsity of the spectrum matrix, | X | | non-woven phosphor_1，2For constraining the row sparsity, x, of the spectral matrix_i，jIs the ith row and jth column element, v, of the frequency spectrum matrix X_i，jIs an error matrixV ith row and jth column element, λ₁、λ₂、λ₃Weight is used to weigh the importance of each item, Y is phi X + V is the constraint condition of equation, Y ∈ R^M×HIs an observation matrix, X ∈ C^N×HIs the spectrum matrix of the corresponding signal, i.e. the sparse spectrum matrix to be solved, phi ∈ C^M×N(M < N) is a redundant discrete Fourier transform basis and V is a model error or measurement error matrix.

And step S260, solving the target function in the joint sparse spectrum reconstruction model by an inaccurate and augmented Lagrange multiplier method.

In this step, typically, to solve the complex objective function in equation (1), two relaxation variables X are introduced in equation (2)₁、X₂Two equality constraints are added simultaneously:

$\underset{X_1,X_2,V}{min}{\mathrm{\&lambda;}}_1\left|\right|V|{|}_F^2+{\mathrm{\&lambda;}}_2|\left|X_1\right|{|}_{1,2}+{\mathrm{\&lambda;}}_3\left|\right|X_2|{|}_{1,1}$

$s.t\left\{\begin{array}{c}Y=\mathrm{\&Phi;}X+V\\ X=X_1\\ X=X_2\end{array}\right.---\left(2\right)$

then, combining the equation constraint term and the objective function in the formula (2) by using an augmented Lagrange multiplier method, and obtaining a minimized expression through a series of simple closed alternative updating operations, as shown in the formula (3):

$\begin{array}{c}L(X,X_1,X_2,V,Q_{1,...,3},\mathrm{\&mu;})={\mathrm{\&lambda;}}_1\left|\right|V|{|}_F^2+{\mathrm{\&lambda;}}_2|\left|X_1\right|{|}_{1,2}+{\mathrm{\&lambda;}}_3\left|\right|X_2|{|}_{1,1}\\ +<Q_1,Y-\mathrm{\&Phi;}X-V>+<Q_2,X-X_1>+<Q_3,X-X_2>\\ +\frac{\mathrm{\&mu;}}{2}\left|\right|Y-\mathrm{\&Phi;}X-V|{|}_F^2+\frac{\mathrm{\&mu;}}{2}||X-X_1|{|}_F^2+\frac{\mathrm{\&mu;}}{2}\left|\right|X-X_2|{|}_F^2\\ \&DoubleRightArrow;\underset{X,X_1,X_2,V,Q_{1,...,3},\mathrm{\&mu;}}{\min }L(X,X_1,X_2,V,Q_{1,...,3},\mathrm{\&mu;})\end{array}---\left(3\right)$

wherein Q is_1，...，3Is Lagrange multiplier, mu > 0 is penalty parameter;

the inexact augmented lagrange multiplier method is an iterative method, and can independently update each variable in formula (3) in parallel, wherein formulas (4), (5), (6), (7) and (8) are updating formulas of each variable:

$X_1^{*}=W_{\frac{{\mathrm{\&lambda;}}_2}{\mathrm{\&mu;}}}(X+\frac{Q_2}{\mathrm{\&mu;}})---\left(4\right)$

$X_2^{*}=S_{\frac{{\mathrm{\&lambda;}}_3}{\mathrm{\&mu;}}}(X+\frac{Q_3}{\mathrm{\&mu;}})---\left(5\right)$

$V^{*}=\frac{Q_1+\mathrm{\&mu;}(Y-\mathrm{\&Phi;}X)}{2{\mathrm{\&lambda;}}_1+\mathrm{\&mu;}}---\left(6\right)$

$X^{*}={({\mathrm{\&Phi;}}^T\mathrm{\&Phi;}+2I)}^{-1}\&lsqb;\frac{1}{\mathrm{\&mu;}}({\mathrm{\&Phi;}}^T{Q}_1-Q_2-Q_3)+{\mathrm{\&Phi;}}^T(Y-V)+X_1+X_2\&rsqb;---\left(7\right)$

$\left\{\begin{array}{c}{Q}_1=Q_1+\mathrm{\&mu;}(Y-\mathrm{\&Phi;}X-V)\\ Q_2=Q_2+\mathrm{\&mu;}(X-X_1)\\ Q_3=Q_3+\mathrm{\&mu;}(X-X_2)\\ \mathrm{\&mu;}=\min (\mathrm{\&rho;}\mathrm{\&mu;},\mathrm{\&mu;}\max )\end{array}\right.---\left(8\right)$

step S260, a sparse frequency spectrum matrix is obtained through the imprecise augmented Lagrange multiplier method, and signals in the sparse frequency spectrum matrix have the characteristic that the positions of the spectral peaks of the motion acceleration signal frequency spectrum are basically the same as the positions of the spectral peaks of the multiple photoplethysmography signal frequency spectra, so that the purpose of accurately detecting motion noise in the heart rate signal is achieved.

In this embodiment, two photoplethysmography sensors and a three-axis accelerometer are embedded in the wearable heart rate measurement device, two photoplethysmography pulse wave signals and a motion acceleration signal in the same time period are collected at a wrist of a user, a spectrum matrix is formed by the two photoplethysmography pulse wave signals and the motion acceleration signal, global sparse and row sparse structural features in the spectrum matrix are extracted to establish a combined sparse spectrum reconstruction model, a sparse spectrum matrix in the combined sparse spectrum reconstruction model is solved by an inaccurate augmented lagrange multiplier method, and finally, a signal having the characteristic that a peak position of a motion acceleration signal spectrum and peak positions of two photoplethysmography pulse wave signal spectra are basically the same is obtained. The spectrum peak position of the motion noise in the two photoplethysmography signal frequency spectrums can be aligned with the spectrum peak position of the motion acceleration signal frequency spectrum accurately by the aid of the sparse structural characteristics of the rows in the frequency spectrum matrix. The method can accurately detect the strong motion noise in the heart rate signal, and lays a theoretical foundation for effectively removing the strong motion noise in the heart rate signal, so that the heart rate measurement value precision of the wearable heart rate measurement equipment can be improved.

Although the embodiments of the present invention have been described above, the above descriptions are only for the convenience of understanding the present invention, and are not intended to limit the present invention. However, it should be understood that the present invention is not limited to the above-described embodiments, but may be embodied in many different forms and details.

Claims (4)

1. A motion noise detection method suitable for heart rate signals comprises three parts of signal acquisition, joint sparse spectrum reconstruction model construction and sparse spectrum matrix solving, and is characterized in that:

the wearable heart rate measuring device collects a plurality of photoplethysmographic signals and motion acceleration signals in the same time period at the wrist of a user; forming a frequency spectrum matrix by using the plurality of photoplethysmographic signals and the motion acceleration signals, extracting the structural characteristics of the frequency spectrum matrix, establishing the joint sparse spectrum reconstruction model, and solving a sparse frequency spectrum matrix in the joint sparse spectrum reconstruction model;

the method comprises the following steps:

the wearable heart rate measuring equipment acquires a plurality of photoplethysmographic pulse wave signals and motion acceleration signals of a user in the same time period; down-sampling the plurality of photoplethysmographic signals and the motion acceleration signals; then, performing band-pass filtering operation on the downsampled signal;

meanwhile, a frequency spectrum matrix is formed by the plurality of photoplethysmographic signals and the motion acceleration signals, the structural characteristics of integral sparseness and row sparseness in the frequency spectrum matrix are extracted to establish the combined sparse spectrum reconstruction model, the sparse frequency spectrum matrix in the combined sparse spectrum reconstruction model is solved through an inaccurate augmented Lagrange multiplier method, and the solved signals have the characteristic that the spectral peak positions of the motion acceleration signal frequency spectrum are basically the same as those of the plurality of photoplethysmographic signal frequency spectrums.

2. A method of motion noise detection adapted for use with a heart rate signal as defined in claim 1, wherein:

a plurality of photoplethysmography sensors and a three-axis accelerometer are embedded in the wearable heart rate measuring equipment; the plurality of photoplethysmography sensors collect a plurality of photoplethysmography signals of a user; the three-axis accelerometer collects motion acceleration signals of a user in the same time period.

3. A method of motion noise detection adapted for use with a heart rate signal as defined in claim 1, wherein:

the combined sparse spectrum reconstruction model is constructed according to the fact that a spectrum matrix formed by the plurality of photoplethysmographic signals and the motion acceleration signals has structural characteristics of global sparsity and row sparsity, and an objective function of the combined sparse spectrum reconstruction model is as follows:

$\underset{X,V}{min}{\mathrm{\&lambda;}}_1\left|\right|V|{|}_F^2+{\mathrm{\&lambda;}}_2|\left|X\right|{|}_{1,2}+{\mathrm{\&lambda;}}_3\left|\right|X|{|}_{1,1}$

$\begin{array}{cc}s.t& \left\{\begin{array}{c}\left|\right|V|{|}_F=\sqrt{tr\left(V^{T}V\right)}=\sqrt{\underset{i=1}{\overset{M}{\&Sigma;}}\underset{j=1}{\overset{R}{\&Sigma;}}{v_{i,j}}^2}\\ \left|\right|X|{|}_{1,2}=\underset{i=1}{\overset{N}{\&Sigma;}}{\left(\underset{j=1}{\overset{R}{\&Sigma;}}{x_{i,j}}^2\right)}^{\frac{1}{2}}\\ \left|\right|X|{|}_{1,1}=\underset{i=1}{\overset{N}{\&Sigma;}}\underset{j=1}{\overset{R}{\&Sigma;}}|x_{i,j}|\\ Y=\mathrm{\&Phi;}X+V\end{array}\right.\end{array}$

wherein,

||V||_Ffor constraining the error matrix V to minimize the error, | X | | | non-calculation_1，1Used for constraining the global sparsity of the spectrum matrix, | X | | non-woven phosphor_1，2For constraining the row sparsity, x, of the spectral matrix_i，jIs the ith row and jth column element, v, of the frequency spectrum matrix X_i，jIs the ith row and jth column element, λ, of the error matrix V₁、λ₂、λ₃Weight is used to weigh the importance of each item, Y is phi X + V is the constraint condition of equation, Y ∈ R^M×HIs an observation matrix, X ∈ C^N×HIs the spectrum matrix of the corresponding signal, i.e. the sparse spectrum matrix to be solved, phi ∈ C^M×N(M < N) is a redundant discrete Fourier transform basis and V is a model error or measurement error matrix.

4. A method of motion noise detection adapted for use with a heart rate signal as defined in claim 1, wherein:

the inaccurate augmentation Lagrange multiplier method is an iterative method which is more suitable for the combined sparse spectrum reconstruction model target function and is constructed by combining a penalty function with a traditional Lagrange function; while the inexact augmented lagrange multiplier method allows updating unknown variables in an alternating or sequential manner with a rate of quadratic convergence.