A processing method for single-channel human physiological signals during sleep

Abstract

The present invention discloses a processing method for single-channel human physiological signals during the sleep period. First, a high-precision sensor is used to collect single-channel complex signals; a method of constructing a matrix space is used to extract the features of the single-channel complex signals, and through the mapping of different features, a high-dimensional matrix space is obtained, which contains the features of at least two or more signals; then a specific feature basis is used to perform a space transformation on the high-dimensional matrix space to separate the features related to the sleep quality of the test object from the high-dimensional matrix space; and then a gating decision filtering method is used to perform complex modeling on the separated single-channel feature signal group using multi-parameter indicators, and through pattern recognition, gating filtering separation of the single-channel complex signals is achieved. This method can achieve higher-precision filtering and separation of complex signals such as physiological signals during the sleep period, expanding the flexibility and application range of sleep monitoring devices.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing, and in particular to a method for processing single-channel human physiological signals during sleep. Background Art

[0002] Sleep is an active process of the human body, which plays an important role in restoring mental and physical strength from fatigue. Ensuring high-quality sleep can promote the maintenance and strengthening of health. With the public's further attention to high sleep quality, the market demand for long-term and comfortable personal sleep monitoring equipment is gradually expanding. In the field of sleep signal processing, researchers often face a key problem: the physiological signals during sleep are weak, and the acquisition process is easily disturbed by complex noise due to external disturbances. In response to this problem, sleep monitoring equipment concentrated in large institutions such as hospitals often uses multiple signals collected by multiple sensors to infer the sleep state. For example, PSG equipment requires multiple sets of electrodes to be worn on the human body to collect physiological signals. Although PSG equipment can improve the anti-interference ability of noise through the signal processing method of mutual calibration of multiple sensors, it will cause physiological inconvenience and psychological pressure to the user during the acquisition process, making it difficult to achieve long-term and comfortable sleep monitoring. On the other hand, devices using a single sensor need to face complex and highly random background noise, and the mixed superposition method of the target signal group is unknown. On the one hand, traditional signal processing methods use single features for filtering, which has weak anti-interference ability, and the separated signals are easily mixed with background noise or even pseudo signals. On the other hand, and more importantly, traditional signal processing methods have poor ability to extract features from single-channel signals and find it difficult to deeply mine information features. Therefore, the test results obtained are often significantly different from the standard results (the industry uses the results of PSG equipment as the standard).

[0003] In the face of the above contradictions, the research of existing technologies mainly focuses on, on the one hand, how to reduce the number of sensors used in PSG devices, and use fewer sensors to achieve the original accuracy by optimizing the placement. However, this idea still cannot solve the pain point that PSG devices cannot perform long-term and flexible sleep monitoring due to their large size, complex structure and high cost; on the other hand, products such as Xiaomi bracelets and heart rate belts focus on using fewer piezoelectric sensors (generally 2 to 3) and reducing their size to achieve the purpose of long-term and comfortable sleep monitoring. However, piezoelectric sensors have strict requirements for contact with the skin, and the actual monitoring accuracy is often far from the theoretical data in the laboratory. In addition, such a system also has the disadvantages of small amount of information collected by a single sensor and shallow depth of information mining by the signal processing method. In complex scenarios of actual applications, it cannot improve the actual accuracy. Summary of the invention

[0004] The object of the present invention is to provide a method for processing single-channel human physiological signals during sleep. This method can filter and separate signals with higher precision for complex signals such as physiological signals during sleep, expanding the flexibility and application range of sleep monitoring devices.

[0005] The object of the present invention is achieved by the following technical solutions:

[0006] A method for processing single-channel human physiological signals during sleep, the method comprising:

[0007] Step 1: First, use a high-precision sensor to collect single-channel complex signals, where the single-channel complex signals include human physiological signals such as body movements, breathing, and heartbeats related to the sleep quality of the test object;

[0008] Step 2: Use the method of constructing a matrix space to extract the features of the single-channel complex signals. Through mapping of different features, a high-dimensional matrix space is obtained, which contains the features of at least two or more signals;

[0009] Step 3: Then use a specific feature basis to perform a space transformation on the high-dimensional matrix space to separate the features related to the sleep quality of the test object from the high-dimensional matrix space; among them, the features related to the sleep quality of the test object include pulse signal and respiratory wave signal features accompanying the sleep state;

[0010] Step 4: Then use the method of gated decision filtering to perform complex modeling on the separated single-channel feature signal group using multi-parameter indicators, and achieve gated filtering separation of the single-channel complex signals through pattern recognition.

[0011] It can be seen from the above technical solutions provided by the present invention that the above method can filter and separate signals with higher precision for complex signals such as physiological signals during sleep, expanding the flexibility and application range of sleep monitoring devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0013] Figure 1 It is a schematic flowchart of the method for processing single-channel human physiological signals during sleep provided by the embodiments of the present invention;

[0014] Figure 2 It is a schematic diagram of the respiratory wave signal after complex signal separation in the embodiments of the present invention;

[0015] Figure 3 Schematic diagram of the heart rate signal after complex signal separation according to the embodiment of the present invention;

[0016] Figure 4 Schematic diagram of using the two separated heart rate signals as test inputs according to the embodiment of the present invention;

[0017] Figure 5 Schematic diagram of the PRS parameter indicators of the two corresponding output heart rate signals. Detailed implementation manners

[0018] The following combines the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments, which does not constitute a limitation to the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0019] As Figure 1 shown is a schematic diagram of the processing method flow for single-channel human physiological signals during sleep provided by the embodiment of the present invention, and the method includes:

[0020] Step 1: First, use a high-precision sensor to collect single-channel complex signals, and the single-channel complex signals include human physiological signals such as body movement, breathing, and heart beating related to the sleep quality of the test object;

[0021] Step 2: Use the method of constructing a matrix space to extract the features of the single-channel complex signals, and obtain a high-dimensional matrix space through mapping of different features. The high-dimensional matrix space contains the features of at least two or more signals;

[0022] In this step, first, based on the single-channel complex signals collected by the high-precision sensor, sampling is performed at a time interval of t = 0.05 sec, and 1000 data points are selected. Then the value of the nth data point is X(k n ), n = 1, 2, 3... N (N = 1000);

[0023] Let the one-dimensional matrix Xoc = [X(k1), X(k2), X(k3)... X(k n )], and at the same time, let a segmentation point of the matrix Xoc be From this segmentation point, the matrix can be divided into two sub-matrices, namely:

[0024] Xoc 1-M = [X(k1), X(k2), X(k3)... X(k M )]

[0025] Xoc M-N = [X(k M ),X(k M-1 )...X(k N )]

[0026] Take the sub - matrix Xoc 1-M and the value of Xoc M-n as the row and column respectively, and form a new matrix Xob through Hankel transform;

[0027] Let α1, α2,..., α m and n1, n2,..., n m be the eigenvalues and eigenvectors of the matrix C = Xob(Xob) T , where the eigenvalues are arranged in descending order, then:

[0028] Xob j = n j n j T Xob, j = 1, 2,..., M

[0029] Let G = {r1, r2,..., r d} be the subscripts corresponding to d eigenvalues, then Xob G corresponding to the G - th group can be expressed as:

[0030]

[0031] Divide j = {1, 2,..., M} into c subsets, then:

[0032]

[0033] According to the magnitudes of the d eigenvalues, for the collected single - channel complex signal, first divide it into two groups. Assume the splitting point is P, that is, the above - mentioned matrix Xob can be divided into Xob1 and Xob2. Construct two M×(N - M + 1) zero matrices Xo1 and Xo2. Perform addition operations on Xob1, Xob2 with Xo1, Xo2 respectively, and obtain matrices Ro1 and Ro2 which are both M×(N - M + 1) matrices. At the same time, calculate the average values of the anti - diagonal elements of matrices R1 and R2, thus forming two sequences:

[0034] Ra1 = {r a1 , ra2,..., r an} T

[0035] Rb1 = {r b1 , rb2,..., r bn} T

[0036] Finally, a 2×N matrix Xoa = [Ra1, Rb1] is obtained.

[0037] Step 3: Then, use a specific eigenbasis to perform a spatial transformation on the high-dimensional matrix space to separate the features related to the sleep quality of the test object from the high-dimensional matrix space;

[0038] Among them, the features related to the sleep quality of the test object include the pulse signal and the respiratory wave signal features accompanying the sleep state;

[0039] In this step, assume an independent source S. Then, it is considered that the observed mixed matrix Z is obtained by weighting the independent source S by A. The goal of signal separation is to find a separation matrix D through the mixed matrix Z such that the signal Y obtained by applying the separation matrix D to the mixed matrix Z is the optimal approximation of the independent source S. This relationship can be expressed by the following formula:

[0040] Y = DZ = DAS

[0041] Assume that there are n independent sources s. Therefore, each independent source s has a corresponding mixed matrix z, which is z = As. Assume that each independent source S has a probability density, which can be expressed as p F (S i ). Then, the joint distribution of the original signal at a given moment is:

[0042]

[0043] From the above formula, the following can be obtained:

[0044]

[0045] Assume that the cumulative distribution function of s conforms to:

[0046]

[0047] Given m as the number of training samples x, the following can be obtained:

[0048]

[0049] Derive D, and through iteration, find D to obtain the matrix Soa = DXoa composed of single-channel signals;

[0050] As Figure 2 shown in the schematic diagram of the respiratory wave signal after complex signal separation described in the embodiment of the present invention, as Figure 3 shown in the schematic diagram of the heart rate signal after complex signal separation described in the embodiment of the present invention, which respectively correspond to the signals in the first column and the second column of the matrix Soa.

[0051] Step 4: Then, use the gated decision filtering method to perform complex modeling on the separated single-channel feature signal groups using multi-parameter indicators, and achieve gated filtering separation of the single-channel complex signals through pattern recognition.

[0052] In this step, the multi-parameter indicators include PRS parameters and parameters, where:

[0053] For the PRS parameter, set a time period Then the PRS parameter can be expressed as:

[0054]

[0055] From FFT, we can get:

[0056] F T (ω)=F[Soa T (t)]

[0057] When T→∞, Xoa T (t)→Xoa(t), then:

[0058]

[0059] Thus, the expression of PRS(ω) can be obtained as

[0060]

[0061] As Figure 4 shown is a schematic diagram of using the two separated heart rate signals as test inputs in the embodiment of the present invention. As Figure 5 shown is a schematic diagram of the PRS parameter indicators of the corresponding output two-channel heart rate signals;

[0062] For parameters, according to the research of the American Academy of Sleep Medicine AASM on individual differences in sleep stages, there are differences in the average occurrence rates of different ages and genders in sleep stages. Therefore, the age range is divided into groups of young (≤44 years old), middle-aged (45 - 59 years old), young old (60 - 74 years old), and old (75 - 89 years old);

[0063] Let the occurrence rates of the average sleep stages in the population for the five defined stages of wakefulness, REM sleep, Non-REM sleep stages 1, 2, and 3 be P1, P2, P3, P4, and P5 respectively; let the average sleep stage occurrence rates in the young group, middle-aged group, young-old group, and old group be p1, p2, p3, and p4 respectively; let the average sleep stage occurrence rates in the male population and female population be g1, g2, g3, g4, and g5 respectively; let the average sleep stage occurrence rates of an individual be St1, st2, St3, St4, and St5, and wg be the gender weight and wa be the age weight. Then, an indicator parameter is ζg, and its expression is as follows:

[0064]

[0065] With the prior knowledge of obtaining the gender and age information of the tested individual, The expression of the parameter is as follows:

[0066]

[0067] In a specific implementation, the process of using the above multi-parameter index to perform complex modeling on the separated single-channel feature signal group is specifically as follows:

[0068] Using the principle of the kernel method to perform complex modeling on the separated single-channel feature signal group, indicating as the Gaussian kernel function, assuming a set of data:

[0069] U = {(te1, re1), (te2, re2), …, (te k , re k )}

[0070] where te i is the PRS parameter calculated for the corresponding k groups of single-channel feature signals obtained above or parameter;

[0071] re i ∈ {+1, -1},

[0072] Select an appropriate B > 0 and construct the problem:

[0073]

[0074] Suppose

[0075]

[0076] and 0 ≤ α i ≤ B, and thus obtain the optimal solution:

[0077] α * = (α1* , α2 * , ..., α k * ) T

[0078] Select a component α j * , and calculate:

[0079]

[0080] Finally, obtain:

[0081]

[0082] By performing complex modeling on the separated single-channel characteristic signal groups, gating filtering separation of the single-channel complex signals is achieved.

[0083] Thus, through complex modeling of the high-dimensional signals of individuals, the embodiment of the present invention constructs a gating decision module to achieve pattern recognition, thereby obtaining a cleaner single-channel signal group, filtering and separating the signals with higher accuracy, and expanding the flexibility and application range of the sleep monitoring device.

[0084] It should be noted that the content not described in detail in the embodiments of the present invention belongs to the prior art well-known to those skilled in the art. For example, the signal processing method in this embodiment is not only applicable to the classification of physiological signals during sleep, but can also be adapted to a system for extracting single-channel signals with different characteristics from other types of single-channel complex signals by appropriately modifying parameters.

[0085] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims. The information disclosed in the background art part of this article is only intended to deepen the understanding of the overall background art of the present invention, and should not be regarded as an admission or any form of implication that this information constitutes the prior art well-known to those skilled in the art.

Claims

1. A processing method for single-channel human physiological signals during sleep, characterized in that, The method includes the following steps: Step 1: First, use a high-precision sensor to collect single-channel complex signals, where the single-channel complex signals include human physiological signals related to body movement, breathing, and heartbeats of the test object related to sleep quality. Step 2: Use the method of constructing a matrix space to extract the features of the single-channel complex signals. Through mapping of different features, a high-dimensional matrix space is obtained, and this high-dimensional matrix space contains the features of at least two or more signals. Step 3: Then, use a specific eigenbasis to perform a space transformation on the high-dimensional matrix space to separate the features related to the sleep quality of the test object from the high-dimensional matrix space. Among them, the features related to the sleep quality of the test object include pulse signal and respiratory wave signal features accompanying the sleep state. Step 4: Then, use the method of gated decision filtering to perform complex modeling on the separated single-channel feature signal group using multi-parameter indicators, and achieve gated filtering separation of the single-channel complex signals through pattern recognition. In Step 4, the specific process of performing complex modeling on the separated single-channel feature signal group is as follows: Using the principle of kernel method to perform complex modeling on the separated single-channel feature signal groups, indicating is a Gaussian kernel function. Assume a set of data: U = {(te1, re1), (te2, re2),..., (te k , re k )} Among them, te i is the PRS parameter calculated for k groups of single-channel characteristic signals or parameters; re i ∈ {+1, -1}, i = 1, 2,... k Select an appropriate B>0 and construct the problem: Assume and 0 ≤ α i ≤ B, from which the optimal solution is obtained: α * ＝(α1 * ，α2 * ，...，α k * ) T Select a component α j * , and calculate: Finally, obtain: Through performing complex modeling on the separated single-channel feature signal group, gated filtering separation of the single-channel complex signals is achieved.

2. The processing method for single-channel human physiological signals during sleep according to claim 1, wherein In Step 2, Based on the single-channel complex signal collected by the high-precision sensor, sampling is carried out at a time interval of t = 0.05 sec, and 1000 data points are selected. Then the value of the nth data point is X(k n ), where n = 1, 2, 3... N (N = 1000); Let the one-dimensional matrix Xoc = [X(k1), X(k2), X(k3)... X(k n )], and at the same time, let a splitting point of the matrix Xoc be From this splitting point, the matrix can be divided into two sub-matrices, namely: Xoc 1-M = [X(k1), X(k2), X(k3)...X(k M )] Xoc M-N = [X(k M ),X(k M-1 )...X(k N )] Take the sub-matrix Xoc 1-M and Xoc M-n as the row and column respectively, and form a new matrix Xob through Hankel transform; Indicate α1, α2,..., α m and n1, n2,..., n m are the eigenvalues and eigenvectors of the matrix C = Xob(Xob) T , where the eigenvalues are arranged in descending order, then: Xob j = n j n j T Xob, j = 1, 2,..., M Let \(G = \{r_1, r_2, \ldots, r\) d \}\) be the subscripts corresponding to \(d\) eigenvalues, then \(X_{ob}\) G corresponding to the \(G\)-th group can be expressed as: If j={1, 2,..., M} is divided into c subsets, then: According to the magnitudes of d eigenvalues, for the collected single-channel complex signals, first divide them into two groups. Assume the splitting point is P, that is, the above matrix Xob is divided into Xob1 and Xob2. Construct two M×(N - M + 1) zero matrices Xo1 and Xo2. Perform addition operations on Xob1, Xob2 with Xo1, Xo2 respectively to obtain matrices Ro1 and Ro2, both of which are M×(N - M + 1) matrices. At the same time, calculate the average values of the anti-diagonal elements of matrices R1 and R2, thereby forming two sequences: Ra1 = {r a1 , ra2,..., r an} T Rb1 = {r b1 , rb2,..., r bn} T Finally, a 2×N matrix Xoa = [Ra1, Rb1] is obtained.

Images