Heartbeat signal quality evaluation method and device for smart clothing

Through real-time acquisition of smart clothing and wavelet transformation to process cardiopulmonary vibration signals, combined with the transfer learning method of AlexNet model, the problem of inaccurate heartbeat signal quality assessment is solved, and the accuracy of signal quality assessment and the accuracy of physiological parameter extraction are improved.

CN119073940APending Publication Date: 2024-12-06JIAXING UNIV +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411242920.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-05
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

When smart clothing collects heartbeat signals, the signal quality is uneven due to factors such as clothing position offset, human activities and environmental noise. It is difficult for the existing technology to accurately evaluate the quality of heartbeat signals, which affects the accuracy of subsequent physiological parameter extraction.

Method used

Intelligent clothing uses intelligent clothing to collect multi-channel human heart and lung vibration signals in real time, convert the signal segment into time-frequency diagrams through continuous wavelet transformation, and use the pre-trained AlexNet model to perform quality evaluation, and combine transfer learning method and image quality labels to output signals of each channel. Quality Assessment Score.

Benefits of technology

It improves the accuracy of heartbeat signal quality evaluation, provides richer time and frequency domain feature information, and enhances the accuracy of subsequent physiological parameter extraction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119073940A_ABST
    Figure CN119073940A_ABST
Patent Text Reader

Abstract

The invention provides a heartbeat signal quality evaluation method and device for smart clothing, and relates to the technical field of medical signal processing. The method includes the steps that target multi-channel human body cardiopulmonary vibration signals are collected in real time through the intelligent garment; extracting a target heartbeat signal from the target multichannel human body cardiopulmonary vibration signal, and segmenting the target heartbeat signal to obtain a plurality of target heartbeat signal segments; converting the plurality of target heartbeat signal segments into a plurality of target time-frequency diagrams according to continuous wavelet transform; and inputting the plurality of target time-frequency diagrams into an AlexNet model of the wavelet transform time-frequency diagram, so that the AlexNet model of the wavelet transform time-frequency diagram outputs the quality evaluation score of each channel signal, and the AlexNet model of the wavelet transform time-frequency diagram is obtained by training the AlexNet model by using a plurality of preset time-frequency diagrams and a transfer learning method. The plurality of preset time-frequency diagrams are time-frequency diagrams with image quality labels. In this way, the accuracy of heartbeat signal quality evaluation is high.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of medical signal processing, and in particular to a heartbeat signal quality assessment method and device for smart clothing. Background Art

[0002] In recent years, as people's pursuit of health has gradually increased, medical technology has continued to develop, making digital health and smart healthcare gradually integrated into daily life. With the advantages of real-time, convenient and non-intrusive, smart healthcare is widely used in daily monitoring, disease warning and other fields. In the field of smart healthcare, smart clothing has the nature of daily wear and has become one of the important carriers for monitoring human physiological information. Smart clothing is a kind of clothing embedded with electronic components and sensors. Through sensors, it can sense the vibration information of the human heart and lungs and realize real-time monitoring of vital signs such as heart rate and respiratory rate. As an important physiological signal of the human body, the heartbeat signal can reflect the health status of the heart and is one of the key monitoring objects in smart clothing.

[0003] Smart clothing meets the requirements of real-time, convenient and continuous health monitoring. However, due to various factors including clothing position deviation, human activity and environmental noise, the quality of the heartbeat signals collected by smart clothing varies, and it is necessary to evaluate the quality of the collected heartbeat signals. Clothing position deviation refers to changes in the user's body shape and posture, which causes changes in the relative position between the human body and the sensor. This change will cause changes in the quality of the signal received by sensors at different positions; human activity refers to normal human activities such as talking, coughing, shaking, etc., which cause motion artifacts in the signal; environmental noise refers to electromagnetic interference from other electronic products and power frequency interference from power supply during signal collection.

[0004] At present, the quality of heartbeat signals is mainly evaluated through time domain analysis and frequency domain analysis. However, time domain analysis, such as peak detection, is suitable for evaluating the basic characteristics of heartbeat signals, but its ability to capture subtle changes in signals and noise is limited. Frequency domain analysis evaluates the spectral components of heartbeat signals to identify periodic characteristics, but it is difficult to accurately capture the instantaneous characteristics of non-stationary heartbeat signals. If the quality of heartbeat signals cannot be effectively evaluated, the accuracy of subsequent physiological parameter extraction will be seriously affected. Summary of the invention

[0005] The purpose of the embodiments of the present invention is to provide a method and device for evaluating the quality of a heartbeat signal for smart clothing, so as to solve the problem that the quality evaluation of the heartbeat signal is of low accuracy and affects the accuracy of subsequent extraction of physiological parameters.

[0006] To solve the above technical problems, the embodiments of the present invention provide the following technical solutions:

[0007] A first aspect of the present invention provides a heartbeat signal quality assessment method for smart clothing, the heartbeat signal quality assessment method for smart clothing comprising:

[0008] Using smart clothing, real-time collection of target multi-channel human cardiopulmonary vibration signals;

[0009] Extracting a target heartbeat signal from a target multi-channel human heart and lung vibration signal, and segmenting the target heartbeat signal to obtain a plurality of target heartbeat signal segments;

[0010] According to continuous wavelet transform, multiple target heartbeat signal segments are converted into corresponding multiple target time-frequency graphs;

[0011] Multiple target time-frequency graphs are input into the AlexNet model of wavelet transform time-frequency graphs so that the AlexNet model of wavelet transform time-frequency graphs outputs quality assessment scores for signals of each channel. The AlexNet model of wavelet transform time-frequency graphs is obtained by training the AlexNet model using multiple preset time-frequency graphs and a transfer learning method. The multiple preset time-frequency graphs are time-frequency graphs with image quality labels. The multiple preset time-frequency graphs are time-frequency graphs obtained by sequentially performing preset heartbeat signal extraction, segmentation, image conversion, and image quality label acquisition on preset multi-channel human cardiopulmonary vibration signals.

[0012] A second aspect of the present invention provides a heartbeat signal quality assessment device for smart clothing, the heartbeat signal quality assessment device for smart clothing comprising:

[0013] A real-time acquisition module is used to collect target multi-channel human cardiopulmonary vibration signals in real time using smart clothing;

[0014] An extraction and segmentation module is used to extract a target heartbeat signal from a target multi-channel human cardiopulmonary vibration signal, and segment the target heartbeat signal to obtain a plurality of target heartbeat signal segments;

[0015] A conversion module, used for converting a plurality of target heartbeat signal segments into a plurality of corresponding target time-frequency graphs according to continuous wavelet transform;

[0016] The quality assessment module is used to input multiple target time-frequency graphs into the AlexNet model of the wavelet transform time-frequency graph so that the AlexNet model of the wavelet transform time-frequency graph outputs the quality assessment score of each channel signal. The AlexNet model of the wavelet transform time-frequency graph is obtained by training the AlexNet model using multiple preset time-frequency graphs and a transfer learning method. The multiple preset time-frequency graphs are time-frequency graphs with image quality labels. The multiple preset time-frequency graphs are time-frequency graphs obtained by sequentially performing preset heartbeat signal extraction, segmentation, image conversion, and image quality label acquisition on preset multi-channel human cardiopulmonary vibration signals.

[0017] A third aspect of the present invention provides a computer-readable storage medium, the storage medium including a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute the heartbeat signal quality assessment method for smart clothing according to the above-mentioned first aspect or any optional embodiment of the first aspect.

[0018] Compared with the prior art, the heartbeat signal quality assessment method and device for smart clothing provided by the present invention utilize smart clothing to collect target multi-channel human cardiopulmonary vibration signals in real time; extract the target heartbeat signal from the target multi-channel human cardiopulmonary vibration signal, and segment the target heartbeat signal to obtain multiple target heartbeat signal segments; convert the multiple target heartbeat signal segments into corresponding multiple target time-frequency graphs according to continuous wavelet transform; input the multiple target time-frequency graphs into the AlexNet model of the wavelet transform time-frequency graph, so that the AlexNet model of the wavelet transform time-frequency graph outputs the quality assessment score of each channel signal, the AlexNet model of the wavelet transform time-frequency graph is obtained by training the AlexNet model using multiple preset time-frequency graphs and a transfer learning method, the multiple preset time-frequency graphs are time-frequency graphs with image quality labels, and the multiple preset time-frequency graphs are time-frequency graphs obtained by sequentially performing preset heartbeat signal extraction, segmentation, image conversion, and image quality label acquisition on the preset multi-channel human cardiopulmonary vibration signal. In this way, according to the continuous wavelet transform, multiple target heartbeat signal segments are converted into corresponding multiple target time-frequency diagrams, which can provide rich time domain and frequency domain feature information of the heartbeat signal, making the heartbeat signal quality information richer; through richer heartbeat signal quality information and multiple preset time-frequency diagrams with image quality labels, the AlexNet model can learn the characteristics of high-quality signals, so that the AlexNet model of the wavelet transform time-frequency diagram obtained by training the AlexNet model using multiple preset time-frequency diagrams and the transfer learning method has more accurate quality assessment scores for each channel signal output, thereby making the quality assessment of the heartbeat signal more accurate and improving the accuracy of subsequent physiological parameter extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] By reading the detailed description below with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0020] Figure 1 The process of the heartbeat signal quality assessment method for smart clothing is schematically shown. Figure 1 ;

[0021] Figure 2 The process of the heartbeat signal quality assessment method for smart clothing is schematically shown. Figure 2 ;

[0022] Figure 3 The structure diagram of a heartbeat signal quality assessment device for smart clothing is schematically shown. DETAILED DESCRIPTION

[0023] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0024] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in the present invention should have the common meanings understood by those skilled in the art to which the present invention belongs.

[0025] The method in the embodiment of the present invention is described in detail below.

[0026] Figure 1 The process of the heartbeat signal quality assessment method for smart clothing in an embodiment of the present invention is schematically shown. Figure 1 , see Figure 1 As shown, the method may include:

[0027] S101. Using smart clothing to collect target multi-channel human cardiopulmonary vibration signals in real time.

[0028] Sensors are embedded in multiple locations inside the smart clothing, which have the characteristics of small size, high sensitivity and stable performance. The user wears the smart clothing normally, and uses the smart clothing to collect the target multi-channel human cardiopulmonary vibration signal in real time, with a sampling frequency of fs Hz.

[0029] The number of channels N1 of the target multi-channel human cardiopulmonary vibration signal is usually 2 to 9.

[0030] S102: extracting a target heartbeat signal from the target multi-channel human heart and lung vibration signal, and segmenting the target heartbeat signal to obtain a plurality of target heartbeat signal segments.

[0031] Multiple target heartbeat signal segments can be represented as S = {s1, s2, ..., s N}, where S is a number of target heartbeat signal segments, s k is the kth target heartbeat signal segment, k∈[1,N], and N is the number of target heartbeat signal segments. The length of each target heartbeat signal segment is t seconds. Each target channel obtains a target heartbeat signal segment, and the number of multiple target heartbeat signal segments N is the same as the number of channels N1 of the target multi-channel human cardiopulmonary vibration signal.

[0032] S103. Convert multiple target heartbeat signal segments into corresponding multiple target time-frequency graphs according to continuous wavelet transform.

[0033] According to continuous wavelet transform, multiple target heartbeat signal segments S = {s1, s2, ..., s N} is converted into corresponding multiple target time-frequency graphs C = {c1, c2, ..., c N}, where C is a time-frequency graph of multiple targets, c k is the target time-frequency graph corresponding to the kth target heartbeat signal segment, k∈[1,N].

[0034] Continuous Wavelet Transform (CWT) is suitable for time-frequency analysis of non-stationary signals and can provide localized information of heartbeat signals in the time domain and frequency domain. CWT analyzes the frequency components of heartbeat signals by scaling and translating a basic wavelet function (mother wavelet). According to the continuous wavelet transform, multiple target heartbeat signal segments are converted into corresponding multiple target time-frequency graphs, and a comprehensive representation of the time domain and frequency domain characteristics of multiple target heartbeat signal segments can be obtained, providing richer heartbeat signal quality information.

[0035] S104, inputting the multiple target time-frequency graphs into the AlexNet model of the wavelet transformed time-frequency graph, so that the AlexNet model of the wavelet transformed time-frequency graph outputs the quality assessment score of each channel signal.

[0036] Among them, the AlexNet model of the wavelet transform time-frequency graph is obtained by training the AlexNet model using multiple preset time-frequency graphs and the transfer learning method. The multiple preset time-frequency graphs are time-frequency graphs with image quality labels. The multiple preset time-frequency graphs are time-frequency graphs obtained by performing preset heartbeat signal extraction, segmentation, image conversion and image quality label acquisition on the preset multi-channel human cardiopulmonary vibration signal in sequence.

[0037] Multiple target time-frequency graphs C = {c1, c2, ..., c N} is input to the AlexNet (AlexNet-Continuous Wavelet Transform, AlexNet-CWT) model of the wavelet transform time-frequency graph, so that the confidence Conf of the AlexNet-CWT model output time-frequency graph is a high-quality signal = {p1, p2, ..., p N}, confidence is the quality assessment score of each channel signal, Conf is the quality assessment score of each channel signal, p k is the signal quality evaluation score of the channel corresponding to the kth target heartbeat signal segment.

[0038] The AlexNet model is a pre-trained neural network model. The AlexNet model has been trained on approximately 1.2 million images in the ImageNet dataset. The original neural network model has 23 layers and can classify images into 1,000 categories.

[0039] Based on the above Figure 1 It can be seen from the implementation method that the embodiment of the present invention uses smart clothing to collect target multi-channel human cardiopulmonary vibration signals in real time; extracts target heartbeat signals from the target multi-channel human cardiopulmonary vibration signals, and segments the target heartbeat signals to obtain multiple target heartbeat signal segments; according to continuous wavelet transform, converts the multiple target heartbeat signal segments into corresponding multiple target time-frequency graphs; inputs the multiple target time-frequency graphs into the AlexNet model of the wavelet transform time-frequency graph, so that the AlexNet model of the wavelet transform time-frequency graph outputs the quality evaluation score of each channel signal, the AlexNet model of the wavelet transform time-frequency graph is obtained by training the AlexNet model using multiple preset time-frequency graphs and transfer learning method, the multiple preset time-frequency graphs are time-frequency graphs with image quality labels, and the multiple preset time-frequency graphs are time-frequency graphs obtained by extracting, segmenting, converting images and obtaining image quality labels for preset multi-channel human cardiopulmonary vibration signals in sequence. In this way, according to the continuous wavelet transform, multiple target heartbeat signal segments are converted into corresponding multiple target time-frequency diagrams, which can provide rich time domain and frequency domain feature information of the heartbeat signal, making the heartbeat signal quality information richer; through richer heartbeat signal quality information and multiple preset time-frequency diagrams with image quality labels, the AlexNet model can learn the characteristics of high-quality signals, so that the AlexNet model of the wavelet transform time-frequency diagram obtained by training the AlexNet model using multiple preset time-frequency diagrams and the transfer learning method has more accurate quality assessment scores for each channel signal output, thereby making the quality assessment of the heartbeat signal more accurate and improving the accuracy of subsequent physiological parameter extraction.

[0040] As a refinement and expansion of the above embodiment, Figure 2 The process of a heartbeat signal quality assessment method for smart clothing in an embodiment of the present invention is as follows Figure 2 , see Figure 2 As shown, a heartbeat signal quality assessment method for smart clothing provided by an embodiment of the present invention may include:

[0041] S201. Using smart clothing to collect target multi-channel human cardiopulmonary vibration signals in real time.

[0042] Step S201 is the same as step S101 and will not be described again here.

[0043] S202: extracting a target heartbeat signal from the target multi-channel human heart and lung vibration signal, and segmenting the target heartbeat signal to obtain a plurality of target heartbeat signal segments.

[0044] Specifically, a target heartbeat signal is extracted from the target multi-channel human heart and lung vibration signal, and the target heartbeat signal is segmented to obtain multiple target heartbeat signal segments, including:

[0045] Step A1: removing the DC component from the target multi-channel human cardiopulmonary vibration signal to obtain a signal after the DC component is removed.

[0046] In order not to affect the subsequent heartbeat signal processing, the DC component in the target multi-channel human heart and lung vibration signal is removed using the following formula:

[0047]

[0048] in, is the signal after removing the DC component, x k1 [i] is the signal value of the i-th point of the human cardiopulmonary vibration signal of the target k1-th channel, is the average value of the human cardiopulmonary vibration signal of the target k1th channel.

[0049] Step A2: Using a bandpass filter, remove the signal outside the preset frequency range from the signal after the DC component is removed, and extract the target heartbeat signal.

[0050] The preset frequency range is 5-30Hz.

[0051] A bandpass filter is used to remove signals outside a preset frequency range from the signal after the DC component is removed, that is, the respiratory signal and power frequency interference are removed to extract the target heartbeat signal.

[0052] The target multi-channel human cardiopulmonary vibration signal is mixed with multiple signals. The target multi-channel human cardiopulmonary vibration signal collected in real time by smart clothing mainly includes heartbeat signal and breathing signal. The main frequency of the breathing signal is below 1Hz, and the main frequency of the heartbeat signal is 5-30Hz. In addition, the power supply will cause 50Hz power frequency interference during the acquisition process, so a bandpass filter is used to remove low-frequency interference, namely the breathing signal, and high-frequency interference, namely the power frequency interference, to extract the target heartbeat signal.

[0053] An infinite impulse response (IIR) bandpass filter is used with the lower cutoff frequency set to 5 Hz, the upper cutoff frequency set to 30 Hz, the order set to 4, and the sampling frequency set to fs Hz. The IIR bandpass filter can effectively suppress noise and retain the frequency band of the heartbeat signal, thereby extracting a relatively pure heartbeat signal.

[0054] Step A3: Segment the target heartbeat signal according to a preset time length to obtain a plurality of target heartbeat signal segments.

[0055] The preset time length t may be 10 seconds. There are various preset time lengths, and no specific limitation is made here.

[0056] Exemplarily, the target heartbeat signal is segmented according to a preset time length of 10 seconds, and the length of each target heartbeat signal segment obtained is 10 seconds.

[0057] The following steps S203 to S206 are specific operations of converting multiple target heartbeat signal segments into corresponding multiple target time-frequency graphs according to continuous wavelet transform.

[0058] S203. Determine multiple groups of target wavelet coefficients corresponding to multiple target heartbeat signal segments according to continuous wavelet transform.

[0059] CWT is suitable for time-frequency analysis of non-stationary signals and can provide localized information of heartbeat signals in the time domain and frequency domain. CWT analyzes the frequency components of heartbeat signals by scaling and translating a basic wavelet function (mother wavelet). According to the continuous wavelet transform, multiple target heartbeat signal segments are converted into corresponding multiple target time-frequency graphs, and a comprehensive representation of the time domain and frequency domain characteristics of multiple target heartbeat signal segments can be obtained, providing richer heartbeat signal quality information.

[0060] Specifically, according to the continuous wavelet transform, multiple groups of target wavelet coefficients corresponding to multiple target heartbeat signal segments are determined, including:

[0061] According to the continuous wavelet transform, multiple groups of target wavelet coefficients corresponding to multiple target heartbeat signal segments are determined using the following formula:

[0062]

[0063] Among them, W k (a,b) is the target wavelet coefficient of the kth group at scale a and position b, where a is the scale, b is the position, and s k [p] is the signal value of the pth point in the kth target heartbeat signal segment, k∈[1,N], N is the number of target heartbeat signal segments, ψ is the Morlet wavelet, ψ * is the complex conjugate of the Morlet wavelet, j is the imaginary unit, ω0 is the center frequency, and z is the time coordinate of the Morlet wavelet.

[0064] ψ is the Morlet wavelet, that is, the mother wavelet, ω0=10π.

[0065] S204 , respectively normalize the absolute values ​​of the multiple groups of target wavelet coefficients to obtain the corresponding multiple groups of target normalized wavelet coefficients.

[0066] Specifically, the absolute values ​​of multiple groups of target wavelet coefficients are normalized to the range of [0,1] using the following formula to obtain the corresponding multiple groups of target normalized wavelet coefficients:

[0067]

[0068] Among them, W k ′2 is the target normalized wavelet coefficient of the k2th group, that is, the target normalized wavelet coefficient corresponding to the target wavelet coefficient of the k2th group, |Wk2(a,b)| is the absolute value of the target wavelet coefficient of the k2th group, W k2 (a,b) is the k2th group of target wavelet coefficients, that is, the k2th group of target wavelet coefficients at scale a and position b, a is the scale, b is the position, min(Wk2(a,b)) is the minimum absolute value of the k2th group of target wavelet coefficients, max(Wk2(a,b)) is the maximum absolute value of the k2th group of target wavelet coefficients, k2∈[1,N2], N2 is the number of groups of target wavelet coefficients.

[0069] S205 , converting the multiple groups of target normalized wavelet coefficients into 8-bit unsigned integer types respectively to obtain the corresponding multiple groups of target conversion results.

[0070] An 8-bit unsigned integer type is represented by the uint8 type.

[0071] Use the following formula to convert multiple sets of target normalized wavelet coefficients into uint8 type respectively to obtain the corresponding multiple sets of target conversion results:

[0072]

[0073] in, is the target conversion result of the k2th group, that is, the target conversion result corresponding to the normalized wavelet coefficient of the k2th group of targets, round(y) is the rounding of y, y = W k '2×255,W k ′2 is the normalized wavelet coefficient of the k2th group of targets.

[0074] S206 , converting the multiple groups of target conversion results into RGB images respectively to obtain corresponding multiple target time-frequency graphs.

[0075] Will By converting it into RGB images, we can get multiple target time-frequency graphs, that is, the wavelet transform time-frequency graph of N heartbeat signals C = {c1, c2, ..., c N}.

[0076] S207, inputting the multiple target time-frequency graphs into the AlexNet model of the wavelet transformed time-frequency graph, so that the AlexNet model of the wavelet transformed time-frequency graph outputs the quality assessment score of each channel signal.

[0077] Among them, the AlexNet model of the wavelet transform time-frequency graph is obtained by training the AlexNet model using multiple preset time-frequency graphs and the transfer learning method. The multiple preset time-frequency graphs are time-frequency graphs with image quality labels. The multiple preset time-frequency graphs are time-frequency graphs obtained by performing preset heartbeat signal extraction, segmentation, image conversion and image quality label acquisition on the preset multi-channel human cardiopulmonary vibration signal in sequence.

[0078] Multiple target time-frequency graphs C = {c1, c2, ..., c N} is input to the AlexNet-CWT model so that the confidence Conf of the AlexNet-CWT model outputting the time-frequency graph as a high-quality signal is Conf = {p1, p2, ..., p N}, confidence is the quality assessment score of each channel signal, Conf is the quality assessment score of each channel signal, p k is the quality assessment score of the channel signal corresponding to the kth target heartbeat signal segment.

[0079] Specifically, before inputting the plurality of target time-frequency graphs into the AlexNet model of the wavelet transformed time-frequency graph so that the AlexNet model of the wavelet transformed time-frequency graph outputs the quality assessment score of each channel signal, the method further includes:

[0080] Step B1: Adjust the AlexNet model based on the output preset categories of the target fully connected layer, the class labels of the target classification layer, and the target training parameters.

[0081] The AlexNet model is a pre-trained neural network model. The AlexNet model has been trained on approximately 1.2 million images in the ImageNet dataset. The original neural network model has 23 layers and can classify images into 1,000 categories.

[0082] Specifically, in order to make the AlexNet model suitable for the time-frequency diagram of the heartbeat signal, the AlexNet model is adjusted: the target fully connected layer of the AlexNet model, that is, the last fully connected layer, is modified, the output preset category, that is, the number of output nodes, is set to the number of categories 2, and the learning rate factor of the weight and bias is set to 5. After the setting, the weight and bias of the fully connected layer will be updated 5 times faster during the training process; the target classification layer, that is, the last output classification layer, is modified, and the classification layer is replaced with a classification layer without class labels. The adjusted AlexNet model can be trained using the data set of the time-frequency diagram; the training parameters of the AlexNet model are adjusted.

[0083] Step B2: According to a plurality of preset time-frequency graphs, the AlexNet model is trained by a transfer learning method to obtain an AlexNet model of the wavelet transform time-frequency graph.

[0084] The preset time-frequency diagram dataset includes multiple preset time-frequency diagrams.

[0085] The preset time-frequency graph dataset can be divided into training set, validation set and test set in a ratio of 6:2:2. The training uses the Stochastic Gradient Descent with Momentum (SGDM) optimizer, the training cycle is set to 10, the size of the small batch sample is 15, and the initial learning rate is 0.0001.

[0086] Use the preset time-frequency graph dataset to train the adjusted AlexNet model through transfer learning method, export the trained AlexNet model and save it as AlexNet-CWT model.

[0087] The trained AlexNet model, namely the AlexNet-CWT model, achieved an accuracy of over 94% on the test set.

[0088] Specifically, before adjusting the AlexNet model according to the output preset category of the target fully connected layer, the class label of the target classification layer, and the target training parameters, the method further includes:

[0089] Step C1: Using smart clothing to collect preset multi-channel human cardiopulmonary vibration signals.

[0090] The operation of step C1 is the same as that of step S101 , so it will not be described again here.

[0091] Step C2: performing preset heartbeat signal extraction, segmentation, image conversion and image quality label acquisition on the preset multi-channel human cardiopulmonary vibration signal in sequence to obtain a plurality of preset time-frequency graphs.

[0092] Specifically, the preset multi-channel human cardiopulmonary vibration signal is sequentially subjected to preset heartbeat signal extraction, segmentation, image conversion, and image quality label acquisition to obtain a plurality of preset time-frequency graphs, including:

[0093] Step C21: extracting a preset heartbeat signal from the preset multi-channel human heart and lung vibration signal, and segmenting the preset heartbeat signal to obtain a plurality of preset heartbeat signal segments.

[0094] Specifically, a preset heartbeat signal is extracted from a preset multi-channel human heart and lung vibration signal, and the preset heartbeat signal is segmented to obtain a plurality of preset heartbeat signal segments, including:

[0095] Step C211: removing a preset DC component from a preset multi-channel human cardiopulmonary vibration signal to obtain a signal after the preset DC component is removed.

[0096] The operation of step C211 is the same as that of step A1, so it will not be repeated here.

[0097] Step C212: Using a bandpass filter, remove the signal outside the preset frequency range from the signal after removing the preset DC component, and extract the preset heartbeat signal.

[0098] The operation of step C212 is the same as that of step A2, so it will not be repeated here.

[0099] Step C22: converting the plurality of preset heartbeat signal segments into a plurality of corresponding first time-frequency graphs according to continuous wavelet transform.

[0100] Specifically, according to continuous wavelet transform, a plurality of preset heartbeat signal segments are converted into a plurality of corresponding first time-frequency graphs, including:

[0101] Step C221: Determine multiple groups of preset wavelet coefficients corresponding to multiple preset heartbeat signal segments according to continuous wavelet transform.

[0102] The operation of step C221 is the same as that of step S203, so it will not be repeated here.

[0103] Step C222: performing normalization processing on the absolute values ​​of the plurality of preset wavelet coefficients respectively to obtain corresponding sets of preset normalized wavelet coefficients.

[0104] The operation of step C222 is the same as that of step S204, so it will not be repeated here.

[0105] Step C223: Convert the multiple groups of preset normalized wavelet coefficients into 8-bit unsigned integer types respectively to obtain the corresponding multiple groups of preset conversion results.

[0106] The operation of step C223 is the same as that of step S205, so it will not be repeated here.

[0107] Step C224: converting the plurality of sets of preset conversion results into RGB images respectively to obtain a plurality of corresponding first time-frequency graphs.

[0108] Step C23: Acquire a plurality of preset time-frequency graphs, where the plurality of preset time-frequency graphs are time-frequency graphs with image quality labels annotated on each first time-frequency graph according to characteristic peaks of heartbeats and motion artifacts.

[0109] Specifically, the multiple preset time-frequency diagrams obtained are the time-frequency diagrams corresponding to each preset heartbeat signal segment that is manually annotated. It is necessary to manually annotate the multiple first time-frequency diagrams obtained after the conversion. If one or more of the multiple preset heartbeat signal segments have significant and continuous heartbeat characteristic peaks, the amplitudes of these heartbeat characteristic peaks are higher than the background noise, that is, the peak amplitude of the heartbeat characteristic peaks exceeds the amplitude threshold corresponding to the background noise, and there is no motion artifact in the preset heartbeat signal segment, then the preset time-frequency diagram is annotated as high quality; if one or more of the multiple preset heartbeat signal segments do not have significant and continuous heartbeat characteristic peaks, or the amplitudes of these heartbeat characteristic peaks are not higher than the background noise, or there are motion artifacts in the preset heartbeat signal segment, then the preset time-frequency diagram is annotated as low quality.

[0110] The preset heartbeat signal is converted into multiple first time-frequency graphs through continuous wavelet transform, providing rich time domain and frequency domain feature information of the preset heartbeat signal; the AlexNet model is used for transfer learning, and the AlexNet model is adjusted to adapt to the characteristics of the time-frequency graph; combined with the data set of time-frequency graphs with image quality labels, the AlexNet model learns the characteristics of high-quality signals on a large number of labeled samples, thereby having higher accuracy in real-time evaluation of heartbeat signals.

[0111] Based on the same inventive concept, as an implementation of the above-mentioned heartbeat signal quality assessment method for smart clothing, an embodiment of the present invention further provides a heartbeat signal quality assessment device for smart clothing. Figure 3 FIG. 1 is a structural diagram of a heartbeat signal quality assessment device for smart clothing in an embodiment of the present invention, see Figure 3 As shown, the device may include:

[0112] The real-time acquisition module 301 is used to collect the target multi-channel human cardiopulmonary vibration signal in real time by using the smart clothing;

[0113] An extraction and segmentation module 302 is used to extract a target heartbeat signal from a target multi-channel human heart and lung vibration signal, and segment the target heartbeat signal to obtain a plurality of target heartbeat signal segments;

[0114] A conversion module 303, used for converting a plurality of target heartbeat signal segments into a corresponding plurality of target time-frequency graphs according to continuous wavelet transform;

[0115] The quality assessment module 304 is used to input multiple target time-frequency graphs into the AlexNet model of the wavelet transform time-frequency graph, so that the AlexNet model of the wavelet transform time-frequency graph outputs the quality assessment score of each channel signal. The AlexNet model of the wavelet transform time-frequency graph is obtained by training the AlexNet model using multiple preset time-frequency graphs and a transfer learning method. The multiple preset time-frequency graphs are time-frequency graphs with image quality labels. The multiple preset time-frequency graphs are time-frequency graphs obtained by sequentially performing preset heartbeat signal extraction, segmentation, image conversion, and image quality label acquisition on a preset multi-channel human cardiopulmonary vibration signal.

[0116] The extraction and segmentation module 302 is specifically used to remove the DC component from the target multi-channel human cardiopulmonary vibration signal to obtain a signal after the DC component is removed; use a bandpass filter to remove the signal outside the preset frequency range in the signal after the DC component is removed to extract the target heartbeat signal; and segment the target heartbeat signal according to a preset time length to obtain multiple target heartbeat signal segments.

[0117] The conversion module 303 is specifically used to determine multiple groups of target wavelet coefficients corresponding to multiple target heartbeat signal segments based on continuous wavelet transform; normalize the absolute values ​​of the multiple groups of target wavelet coefficients respectively to obtain multiple groups of corresponding target normalized wavelet coefficients; convert the multiple groups of target normalized wavelet coefficients into 8-bit unsigned integer types respectively to obtain multiple groups of corresponding target conversion results; convert the multiple groups of target conversion results into RGB images respectively to obtain multiple corresponding target time-frequency diagrams.

[0118] The conversion module 303 determines multiple groups of target wavelet coefficients corresponding to multiple target heartbeat signal segments according to continuous wavelet transform, including: determining multiple groups of target wavelet coefficients corresponding to multiple target heartbeat signal segments according to continuous wavelet transform using the following formula:

[0119]

[0120] Among them, W k (a,b) is the target wavelet coefficient of the kth group at scale a and position b, where a is the scale, b is the position, and s k [p] is the signal value of the pth point in the kth target heartbeat signal segment, k∈[1,N], N is the number of target heartbeat signal segments, ψ is the Morlet wavelet, ψ * is the complex conjugate of the Morlet wavelet, j is the imaginary unit, ω0 is the center frequency, and z is the time coordinate of the Morlet wavelet.

[0121] It should be pointed out here that the above description of the embodiment of the heartbeat signal quality assessment device for smart clothing is similar to the description of the embodiment of the heartbeat signal quality assessment method for smart clothing, and has similar beneficial effects as the embodiment of the heartbeat signal quality assessment method for smart clothing. For technical details not disclosed in the embodiment of the heartbeat signal quality assessment device for smart clothing of the embodiment of the present invention, please refer to the description of the embodiment of the heartbeat signal quality assessment method for smart clothing of the present invention for understanding.

[0122] Based on the same inventive concept, an embodiment of the present invention further provides a computer-readable storage medium, which includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute the heartbeat signal quality assessment method for smart clothing in one or more of the above embodiments.

[0123] It should be noted here that the description of the above computer-readable storage medium embodiment is similar to the description of the above-mentioned heartbeat signal quality assessment method embodiment for smart clothing, and has similar beneficial effects as the heartbeat signal quality assessment method embodiment for smart clothing. For technical details not disclosed in the embodiment of the computer-readable storage medium of the embodiment of the present invention, please refer to the description of the heartbeat signal quality assessment method embodiment for smart clothing of the present invention for understanding.

[0124] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A heartbeat signal quality assessment method for smart clothing, characterized in that: The heartbeat signal quality assessment method for smart clothing comprises: Using smart clothing, real-time collection of target multi-channel human cardiopulmonary vibration signals; Extracting a target heartbeat signal from the target multi-channel human cardiopulmonary vibration signal, and segmenting the target heartbeat signal to obtain a plurality of target heartbeat signal segments; According to continuous wavelet transform, the plurality of target heartbeat signal segments are converted into a plurality of corresponding target time-frequency graphs; The multiple target time-frequency graphs are input into the AlexNet model of the wavelet transform time-frequency graph, so that the AlexNet model of the wavelet transform time-frequency graph outputs the quality assessment score of each channel signal. The AlexNet model of the wavelet transform time-frequency graph is obtained by training the AlexNet model using multiple preset time-frequency graphs and a transfer learning method. The multiple preset time-frequency graphs are time-frequency graphs with image quality labels, and the multiple preset time-frequency graphs are time-frequency graphs obtained by sequentially performing preset heartbeat signal extraction, segmentation, image conversion, and image quality label acquisition on a preset multi-channel human cardiopulmonary vibration signal.

2. The heartbeat signal quality assessment method for smart clothing according to claim 1, characterized in that: The target heartbeat signal is extracted from the target multi-channel human cardiopulmonary vibration signal, and the target heartbeat signal is segmented to obtain a plurality of target heartbeat signal segments, including: Removing a DC component from the target multi-channel human cardiopulmonary vibration signal to obtain a signal after the DC component is removed; Using a bandpass filter, the signal outside a preset frequency range is removed from the signal after the DC component is removed, so as to extract the target heartbeat signal; The target heartbeat signal is segmented according to a preset time length to obtain the multiple target heartbeat signal segments.

3. The heartbeat signal quality assessment method for smart clothing according to claim 1, characterized in that: The step of converting the plurality of target heartbeat signal segments into a plurality of corresponding target time-frequency graphs according to continuous wavelet transform comprises: Determining, according to the continuous wavelet transform, a plurality of groups of target wavelet coefficients corresponding to the plurality of target heartbeat signal segments; Normalizing the absolute values ​​of the multiple groups of target wavelet coefficients respectively to obtain corresponding multiple groups of target normalized wavelet coefficients; Convert the multiple groups of target normalized wavelet coefficients into 8-bit unsigned integer types respectively to obtain corresponding multiple groups of target conversion results; The multiple groups of target conversion results are converted into RGB images respectively to obtain the corresponding multiple target time-frequency diagrams.

4. The heartbeat signal quality assessment method for smart clothing according to claim 3, characterized in that: Determining, according to the continuous wavelet transform, a plurality of groups of target wavelet coefficients corresponding to the plurality of target heartbeat signal segments comprises: According to the continuous wavelet transform, multiple groups of target wavelet coefficients corresponding to the multiple target heartbeat signal segments are determined using the following formula: Among them, W k (a,b) is the target wavelet coefficient of the kth group at scale a and position b, where a is the scale, b is the position, and s k [p] is the signal value of the pth point in the kth target heartbeat signal segment, k∈[1,N], N is the number of target heartbeat signal segments, ψ is the Morlet wavelet, ψ * is the complex conjugate of the Morlet wavelet, j is the imaginary unit, ω0 is the center frequency, and z is the time coordinate of the Morlet wavelet.

5. The heartbeat signal quality assessment method for smart clothing according to claim 1, characterized in that: Before inputting the plurality of target time-frequency graphs into the AlexNet model of the wavelet transformed time-frequency graph so that the AlexNet model of the wavelet transformed time-frequency graph outputs the quality assessment score of each channel signal, the method further includes: Adjusting the AlexNet model according to the output preset category of the target fully connected layer, the class label of the target classification layer, and the target training parameters; According to the multiple preset time-frequency graphs, the AlexNet model is trained by the transfer learning method to obtain the AlexNet model of the wavelet transform time-frequency graph.

6. The heartbeat signal quality assessment method for smart clothing according to claim 5, characterized in that: Before adjusting the AlexNet model according to the output preset category of the target fully connected layer, the class label of the target classification layer and the target training parameters, the method further includes: Using the smart clothing, collecting the preset multi-channel human cardiopulmonary vibration signal; The preset multi-channel human cardiopulmonary vibration signal is sequentially subjected to preset heartbeat signal extraction, segmentation, image conversion and image quality label acquisition to obtain the plurality of preset time-frequency graphs.

7. The heartbeat signal quality assessment method for smart clothing according to claim 6, characterized in that: The preset multi-channel human heart and lung vibration signal is sequentially subjected to preset heartbeat signal extraction, segmentation, image conversion and image quality label acquisition to obtain the plurality of preset time-frequency graphs, including: Extracting the preset heartbeat signal from the preset multi-channel human heart and lung vibration signal, and segmenting the preset heartbeat signal to obtain a plurality of preset heartbeat signal segments; According to the continuous wavelet transform, converting the plurality of preset heartbeat signal segments into a corresponding plurality of first time-frequency graphs; The plurality of preset time-frequency graphs are obtained, where the plurality of preset time-frequency graphs are time-frequency graphs with image quality labels annotated on each first time-frequency graph according to characteristic peaks of heartbeats and motion artifacts.

8. The heartbeat signal quality assessment method for smart clothing according to claim 7, characterized in that: The converting the plurality of preset heartbeat signal segments into a plurality of corresponding first time-frequency graphs according to the continuous wavelet transform comprises: Determining, according to the continuous wavelet transform, a plurality of preset wavelet coefficients corresponding to the plurality of preset heartbeat signal segments; Normalizing the absolute values ​​of the plurality of preset wavelet coefficients respectively to obtain a plurality of corresponding preset normalized wavelet coefficients; Converting the plurality of preset normalized wavelet coefficients into 8-bit unsigned integer types respectively to obtain corresponding plurality of preset conversion results; The plurality of preset conversion results are converted into RGB images respectively to obtain the corresponding plurality of first time-frequency graphs.

9. A heartbeat signal quality assessment device for smart clothing, characterized in that: The heartbeat signal quality assessment device for smart clothing comprises: A real-time acquisition module is used to collect target multi-channel human cardiopulmonary vibration signals in real time using smart clothing; An extraction and segmentation module, used for extracting a target heartbeat signal from the target multi-channel human cardiopulmonary vibration signal, and segmenting the target heartbeat signal to obtain a plurality of target heartbeat signal segments; A conversion module, used for converting the plurality of target heartbeat signal segments into a plurality of corresponding target time-frequency graphs according to continuous wavelet transform; A quality assessment module is used to input the multiple target time-frequency graphs into the AlexNet model of the wavelet transform time-frequency graph, so that the AlexNet model of the wavelet transform time-frequency graph outputs the quality assessment score of each channel signal, the AlexNet model of the wavelet transform time-frequency graph is obtained by training the AlexNet model using multiple preset time-frequency graphs and a transfer learning method, the multiple preset time-frequency graphs are time-frequency graphs with image quality labels, and the multiple preset time-frequency graphs are time-frequency graphs obtained by sequentially performing preset heartbeat signal extraction, segmentation, image conversion and image quality label acquisition on a preset multi-channel human cardiopulmonary vibration signal.

10. A computer-readable storage medium, characterized in that: The storage medium includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute the heartbeat signal quality assessment method for smart clothing according to any one of claims 1 to 8.