Pulse condition classification model determination method, pulse condition determination method and electronic equipment

Abstract

The embodiment of the invention is suitable for the technical field of biomedical signal processing, and provides a pulse condition classification model determining method, a pulse condition determining method and electronic device.The method comprises the steps that multiple sets of pulse sample signals are obtained, each set of pulse sample signals has a corresponding pulse condition category, and the pulse sample signals are sent to a processor; the pulse sample signal is acquired through a bone conduction microphone; extracting pulse features of each group of pulse sample signals; determining a multi-dimensional feature vector corresponding to the pulse sample signal according to the pulse feature; determining a training data set and a test data set based on the multiple multi-dimensional feature vectors and the pulse condition category corresponding to each multi-dimensional feature vector; and training a preset model through the training data set and the test data set to obtain a pulse condition classification model. By means of the method, the pulse condition classification model can be determined, and therefore automatic classification of pulse conditions can be achieved based on the pulse condition classification model.

Description

Technical Field

[0001] The present application belongs to the technical field of biomedical signal processing, and in particular, relates to a method for determining a pulse classification model, a pulse determination method, and an electronic device. Background Art

[0002] Pulse diagnosis is a method of palpation that detects changes in the pulse by touching the pulse of different parts of the human body. Pulse diagnosis can be classified according to the pulse signal. At present, the classification of pulse signals is mostly based on the personal experience of traditional Chinese medicine. When there is insufficient experience, the pulse may be inaccurate, which may lead to misjudgment. Summary of the invention

[0003] In view of this, an embodiment of the present application provides a method for determining a pulse classification model, a pulse determination method and an electronic device to realize automatic classification of pulses.

[0004] A first aspect of an embodiment of the present application provides a method for determining a pulse classification model, comprising:

[0005] Acquire multiple groups of pulse sample signals, each group of the pulse sample signals has a pre-calibrated pulse category, and the pulse sample signals are collected by a bone conduction microphone;

[0006] Extracting pulse features from each group of pulse sample signals;

[0007] Determine a corresponding multidimensional feature vector of the pulse sample signal according to the pulse feature;

[0008] Determine a training data set and a test data set based on a plurality of the multidimensional feature vectors and the pulse category corresponding to each of the multidimensional feature vectors;

[0009] The preset model is trained by using the training data set and the test data set to obtain a pulse classification model.

[0010] In a possible implementation, the method further includes:

[0011] Determine the pulse category corresponding to each group of pulse sample signals.

[0012] In a possible implementation, the above-mentioned determining the pulse category corresponding to each group of the pulse sample signals includes:

[0013] Based on the characteristic parameter values ​​of each group of pulse sample signals, the pulse category corresponding to the pulse sample signals is determined, wherein the characteristic parameter values ​​include one or more of the pulse position characteristic value, the pulse rate characteristic value, the pulse shape characteristic value, and the pulse strength characteristic value.

[0014] In a possible implementation, determining the pulse category corresponding to the pulse sample signal based on the characteristic parameter value of each group of pulse sample signals includes:

[0015] Determine a pulse category mapping table, wherein the pulse category mapping table includes a characteristic parameter value range corresponding to each pulse category;

[0016] According to the pulse category mapping table and the characteristic parameter value of each group of pulse sample signals, the pulse category corresponding to the characteristic parameter value of each group of pulse sample signals is determined.

[0017] A second aspect of the embodiment of the present application provides a pulse condition determination method, comprising:

[0018] The pulse signal to be classified is collected through a bone conduction microphone;

[0019] extracting a plurality of pulse features of the pulse signal;

[0020] Determine a multidimensional feature vector according to the plurality of pulse features;

[0021] Based on the pulse classification model and the multidimensional feature vector, the pulse category corresponding to the pulse signal is determined, and the pulse classification model is determined by the method described in the first aspect above.

[0022] In a possible implementation, the method further includes:

[0023] A diagnosis report is generated based on the pulse category and the TCM pulse diagnosis and treatment database.

[0024] A third aspect of the embodiment of the present application provides a device for determining a pulse classification model, comprising:

[0025] A signal acquisition module, used for acquiring multiple groups of pulse sample signals, each group of the pulse sample signals has a pre-calibrated pulse category, and the pulse sample signals are acquired through a bone conduction microphone;

[0026] A feature extraction module, used for extracting pulse features of each group of pulse sample signals;

[0027] A vector determination module, used to determine the corresponding multi-dimensional feature vector of the pulse sample signal according to the pulse feature;

[0028] A data set determination module is used to determine a training data set and a test data set based on a plurality of the multidimensional feature vectors and the pulse condition category corresponding to each of the multidimensional feature vectors;

[0029] The model training module is used to train the preset model through the training data set and the test data set to obtain the pulse classification model.

[0030] A fourth aspect of the embodiments of the present application provides a pulse condition determination device, comprising:

[0031] A collection module, used for collecting pulse signals to be classified through a bone conduction microphone;

[0032] An extraction module, used for extracting the pulse feature of the pulse signal;

[0033] A determination module, used for determining a multi-dimensional feature vector according to the pulse feature;

[0034] A classification module is used to determine the pulse category corresponding to the pulse signal based on a pulse classification model and the multidimensional feature vector, wherein the pulse classification model is determined by the method described in the first aspect above.

[0035] A fifth aspect of an embodiment of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the first or second aspect above when executing the computer program.

[0036] A sixth aspect of an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect or the second aspect above is implemented.

[0037] A seventh aspect of the embodiments of the present application provides a computer program product. When the computer program product is run on an electronic device, the electronic device executes the method described in the first aspect or the second aspect.

[0038] Compared with the prior art, the embodiments of the present application have the following advantages:

[0039] By applying the method for determining the pulse classification model of the embodiment of the present application, the pulse classification model can be determined. When determining the pulse classification model, a bone conduction microphone can be used to collect signals, so that more accurate and objective data can be collected, so that the accuracy of the pulse classification model finally generated is higher. When performing model training, pulse features can be extracted from the accurate pulse signal to obtain a data set used for the training model, and the pulse classification model can be obtained by training the model based on the data set.

[0040] Based on the pulse classification model, pulse determination can be achieved. When classifying, the bone conduction microphone is used to collect signals, so that more accurate and objective data can be collected; using the trained pulse classification model to determine the pulse can realize automatic classification of the pulse, reduce manual labor, and thus reduce subjective errors in the classification process. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art.

[0042] Figure 1 It is a schematic flow chart of the steps of a method for determining a pulse classification model provided by an embodiment of the present application;

[0043] Figure 2 It is a schematic diagram of the steps of another pulse condition determination method provided by an embodiment of the present application;

[0044] Figure 3 It is a schematic diagram of another pulse condition determination method provided by an embodiment of the present application;

[0045] Figure 4 It is a schematic diagram of a device for determining a pulse classification model provided in an embodiment of the present application;

[0046] Figure 5 is a schematic diagram of a pulse condition determination device provided in an embodiment of the present application;

[0047] Figure 6 It is a schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0048] In the following description, specific details such as specific system structures, technologies, etc. are proposed for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from hindering the description of the present application.

[0049] As one of the "four diagnoses" of traditional Chinese medicine, pulse diagnosis plays an important role in the diagnosis and prognosis of diseases. However, traditional pulse diagnosis relies on the subjective feeling of doctors and lacks objectivity and standardization. With the development of sensor technology and signal processing technology, it has become possible to obtain and analyze pulse signals objectively and accurately using modern technical means. Based on this, the present embodiment provides a pulse classification model and a pulse determination method.

[0050] The technical solution of the present application is described below through specific embodiments.

[0051] Reference Figure 1 , shows a schematic flow chart of the steps of a method for determining a pulse classification model provided by an embodiment of the present application, which may specifically include the following steps:

[0052] S101, obtaining multiple groups of pulse sample signals, each group of the pulse sample signals having a pre-calibrated pulse category, and the pulse sample signals are collected by a bone conduction microphone.

[0053] The execution subject of the embodiments of the present application may be an electronic device, and the electronic device may be a mobile phone, a tablet computer, a wearable device, a laptop computer, an ultra-mobile personal computer (UMPC), a netbook, a personal digital assistant (PDA), etc. The embodiments of the present application do not impose any restrictions on the specific type of the electronic device.

[0054] In this embodiment, the microphone can be used to capture the tiny vibration and sound wave signals caused by the pulse. For example, a bone conduction microphone can be used. The bone conduction microphone is a sensitive vibration sensor that is sensitive to the tiny vibrations of the skin and blood vessels, thereby being able to capture the high-frequency components of the pulse and effectively capture the physiological vibration signals of the human body. In this embodiment, the pulse signal is collected by the bone conduction microphone, so that objective and accurate signal data can be obtained.

[0055] When collecting signals, bone conduction microphones can be set at multiple parts of the human body. The bone conduction microphone can be placed at the radial artery of the human body to collect pulse vibration signals in real time. As an example, multiple bone conduction microphones can be placed at the three parts of "Cun, Guan, Chi" to collect pulse sample signals. Each group of pulse sample signals includes multiple part signals, and each part signal corresponds to a part of the human body.

[0056] Exemplarily, high-quality pulse signals can be collected at the six parts of the left and right hands, namely, "Cun, Guan, and Chi". The bone conduction microphone can be set at the Cun pulse of the left hand, which corresponds to the heart and small intestine; the bone conduction microphone can be set at the Guan pulse of the left hand, which corresponds to the liver and gallbladder; the bone conduction microphone can be set at the Chi pulse of the left hand, which corresponds to the kidney (left), bladder, and reproductive system; the bone conduction microphone can be set at the Cun pulse of the right hand, which corresponds to the lung and large intestine; the bone conduction microphone can be set at the Guan pulse of the right hand, which corresponds to the spleen and stomach; the bone conduction microphone can be set at the Chi pulse of the right hand, which corresponds to the kidney (right), Mingmen, and Sanjiao. A high-sensitivity bone conduction microphone is installed at each position of the left and right hands, for a total of 6 bone conduction microphones; in order to ensure that the bone conduction microphone is in good contact with the skin, a one-dimensional pressure sensor can be used for automatic contact, thereby fixing the bone conduction microphone firmly.

[0057] When sampling, the pulse signal within the preset time length can be collected based on the preset sampling frequency, so as to obtain the time series signals of the six bone conduction microphones. Based on multiple bone conduction microphones, each group of pulse sample signals includes multiple part signals. For example, each group of pulse sample signals can include 6 part signals, and the 6 part signals correspond to 6 human body parts respectively, that is, each group of pulse sample signals can include the left hand inch part signal S 左寸 (t), left hand joint signal S 左关 (t), left hand ruler signal S 左尺 (t), right hand Cun position signal S 右寸 (t), right hand closed part signal S 右关 (t), signal S of the right hand 右尺 (t). For example, the sampling frequency can be f s =1000Hz, the preset time for collection can be T=60S.

[0058] In order to carry out model training, a sufficient number of groups of pulse sample signals can be collected, and each group of pulse sample signals can be calibrated with a corresponding pulse category.

[0059] After the pulse sample signal is collected, the pulse sample signal can be preprocessed to eliminate interference of the pulse sample signal and improve the signal-to-noise ratio, thereby facilitating feature extraction from the pulse sample signal.

[0060] As an example, the collected pulse sample signal may be filtered, denoised and normalized.

[0061] When performing filtering, a bandpass filter may be used to filter the signal based on a preset cutoff frequency. For example, the bandpass filter may be a Butterworth bandpass filter, and the cutoff frequency may be: f low =0.5Hz, f high =15Hz. For each group of pulse sample signal S 左寸 (t), S 左关 (t), S 左尺 (t), S 右寸 (t), S 右关 (t), S 右尺 (t) are filtered respectively to obtain S 左寸滤 (t), S 左关滤 (t), S 左尺滤 (t), S 右寸滤 (t), S 右关滤 (t), S 右尺滤 (t). That is, S 部位滤 (t) = BandpassFilter(S 部位 (t)).

[0062] When performing noise reduction processing, wavelet denoising or adaptive filtering can be used to perform threshold processing on high-frequency detail coefficients to eliminate noise. 部位降 (t) = WaveletDenoise(S 部位滤 (t)). After filtering, each group of sample signal data is subjected to noise reduction processing, and S 左寸降 (t), S 左关降 (t), S 左尺降 (t), S 右寸降 (t), S 右关降 (t), S 右尺降 (t).

[0063] When performing normalization processing, each group of sample signal data can be normalized based on a preset formula to obtain S 左寸归 (t), S 左关归 (t), S 左尺归 (t), S 右寸归 (t), S 右关归 (t), S 右尺归 (t). For example,

[0064]

[0065] S102, extracting a plurality of pulse features from each group of pulse sample signals.

[0066] As mentioned above, each group of pulse sample signals may include multiple position signals. For each group of pulse sample signals, feature extraction may be performed to obtain pulse features. Pulse features may include one or more of pulse position features, pulse rate features, pulse shape features, and pulse strength features. The pulse position feature may include at least one pulse position feature value; the pulse rate feature may include at least one pulse position feature value; the pulse shape feature may include at least one pulse position feature value; and the pulse strength feature may include at least one pulse position feature value.

[0067] Exemplarily, the pulse characteristics include pulse position characteristics, and each part signal has at least one pulse position characteristic value. Extracting the pulse characteristics corresponding to each group of pulse sample signals may include: when different pressures are applied to any part of the human body, obtaining the maximum amplitude of the part signal under different pressures; taking the sum of the maximum amplitudes of the part signals under different pressures as the target amplitude; calculating the ratio of the first amplitudes corresponding to the part signals of multiple parts of the human body divided by the target amplitude, and obtaining the pulse position characteristic values ​​corresponding to the part signals, wherein the first amplitude is the amplitude under a preset first pressure.

[0068] For example, signals can be collected under different pressures, such as light pressure (floating), medium pressure (medium), and heavy pressure (sinking). Then, the light pressure amplitude of the position signal is calculated from the collected pulse sample data: A浮 =max(S 部位浮 (t)), medium voltage amplitude: A 中 =max(S 部位中 (t)), pressure amplitude: A 沉 =max(S 部位沉 (t)). Afterwards, based on the light pressure amplitude, the medium pressure amplitude and the heavy pressure amplitude, the floating index can be calculated and used as the pulse position characteristic value. The sum of the light pressure amplitude, the medium pressure amplitude and the heavy pressure amplitude can be the target amplitude, and the floating index calculation formula can be:

[0069]

[0070] V P It is a pulse position characteristic value. In this embodiment, the floating index is used as a pulse position characteristic value of the position signal.

[0071] Exemplarily, the pulse feature includes a pulse rate feature, each part signal has at least one pulse rate feature value, and the pulse feature corresponding to each group of pulse sample signals is extracted, including:

[0072] The part signal may have peaks and troughs, and based on the time of the peaks and troughs, the electronic device may determine the period of the part signal. For example, the electronic device may determine the peak time corresponding to the peak of the part signal; based on two adjacent peak times, the average pulse period of the part signal may be calculated; thus, based on the average pulse period, the heart rate and heart rate variability of the part signal may be calculated; and the heart rate and heart rate variability are both used as pulse feature values. That is, a part signal may have two pulse feature values, and the two pulse feature values ​​are the heart rate and heart rate variability values, respectively.

[0073] As an example, peak detection may be performed on the site signal to identify each pulse peak in the site signal, and each pulse peak may have a corresponding peak time t i , then the pulse period can be Ti = t i+1 -ti, the average pulse period can be calculated

[0074]

[0075] Based on the average pulse period, the heart rate HR can be calculated:

[0076]

[0077] The calculation formula of heart rate variability (HRV) can be:

[0078]

[0079] Among them, SDNN is the above heart rate variability value.

[0080] Exemplarily, the pulse feature may also include a pulse shape feature, which may be determined based on the waveform of the site signal. Each site signal has at least one pulse shape feature value, and the pulse feature corresponding to each group of pulse sample signals is extracted, including:

[0081] The electronic device can determine the waveform rise time and waveform fall time of the part signal according to the waveform corresponding to the part signal; calculate the pulse index based on the waveform rise time and waveform fall time; calculate the rising slope, falling slope and curvature of the waveform; and use the pulse index, rising slope, falling slope and curvature as pulse characteristic values. That is, the pulse feature can include 4 pulse characteristic values, which are the pulse index, rising slope, falling slope and curvature.

[0082] As an example, the pulse shape feature can be based on waveform analysis to identify the peak time points t in the position signal. peak , trough time point t valley , waveform slope S and curvature K. Then the rise time T can be calculated rise , fall time T fall And pulse index M:

[0083] T rise =t peak -t valley

[0084] T fall =t next_valley -t peak

[0085]

[0086] Based on the above waveform rise time and waveform fall time, the rising slope, falling slope and curvature of the waveform can be calculated:

[0087]

[0088]

[0089]

[0090] Among them, S rise is the rising slope, S fall is the descending slope, K is the curvature, S peak is the peak value of the waveform slope, S valley is the valley value of the waveform slope.

[0091] In a possible implementation, for a waveform within a cycle, the waveform slope corresponding to each preset point can be calculated, and then the maximum value of the waveform slope is used as the peak value of the waveform slope of the cycle, and the maximum value of the waveform slope is used as the valley value of the waveform slope of the cycle. Among them, each preset point can be preset, for example, a cycle is divided into 100 segments, and an endpoint is determined in each segment as a preset point.

[0092] Exemplarily, the pulse feature includes a pulse feature, which is used to characterize the pulse intensity feature. Each part signal has at least one pulse feature value, and the pulse feature corresponding to each group of pulse sample signals is extracted, including: determining the strength index corresponding to the part signal according to the waveform rise time and waveform fall time of the part signal; integrating the square of the part signal to obtain the pulse wave energy of the part signal; calculating the amplitude of the part signal according to the maximum and minimum values ​​in the part signal; and taking the amplitude, pulse wave energy and strength index as pulse feature values. That is, each part signal has three pulse feature values, and the three pulse feature values ​​are amplitude, pulse wave energy and strength index.

[0093] As an example, the pulse characteristics can be characterized by parameters such as the amplitude A and energy E of the position signal to characterize the strength of the pulse. The calculation formula of the amplitude A is:

[0094] A=max(S 部位归一 (t))-min(S 部位归一 (t))

[0095] The pulse wave energy E is:

[0096]

[0097] The force index L is:

[0098]

[0099] S103: Determine a plurality of multi-dimensional feature vectors according to the plurality of pulse features.

[0100] Each group of pulse sample signals includes multiple part signals, and each part signal has a corresponding at least one pulse position characteristic value, at least one pulse rate characteristic value, at least one pulse shape characteristic value and at least one pulse strength characteristic value. Therefore, the pulse position characteristic values, pulse rate characteristic values, pulse shape characteristic values ​​and pulse strength characteristic values ​​corresponding to each part signal can be arranged in order to obtain a multi-dimensional feature vector.

[0101] Specifically, the computer device can vertically arrange at least one pulse position characteristic value, at least one pulse rate characteristic value, at least one pulse shape characteristic value and at least one pulse potential characteristic value of the part signal in sequence to obtain a part characteristic matrix corresponding to the part signal. Arrange multiple part characteristic matrices corresponding to multiple part signals in each group of pulse sample signals in sequence from left to right to obtain a multi-dimensional feature vector corresponding to the pulse sample signal.

[0102] As an example, each group of pulse sample signals may include 6 part signals, and one part signal may have 1 pulse position characteristic value, 2 pulse number characteristic values, 4 pulse shape characteristic values ​​and 3 pulse potential characteristic values, that is, one part signal may have pulse position characteristic value, heart rate, heart rate variability value, pulse shape index, rising slope, falling slope, curvature, amplitude, pulse wave energy, and strength index. The pulse position characteristic value, heart rate, heart rate variability value, pulse shape index, rising slope, falling slope, curvature, amplitude, pulse wave energy, and strength index corresponding to one part signal are arranged vertically from top to bottom to obtain a part feature matrix. Six part signals can obtain six part feature matrices, and thus the six part feature matrices are arranged in order from left to right to obtain a multidimensional feature vector.

[0103] In the embodiment of the present application, each group of pulse sample signals can determine a multidimensional feature vector, and the multidimensional feature vector can be a 10*6 matrix. In the multidimensional feature vector, each row can represent an eigenvalue, and the multidimensional feature vector includes 10 rows, each row corresponding to: pulse position feature value, heart rate, heart rate variability, pulse shape index, rising slope, falling slope, curvature, amplitude, pulse wave energy, and force index. In the multidimensional feature vector, each column can represent a part, for example, it can include 6 columns, each column corresponding to the left hand Cun part, the left hand Guan part, the left hand Chi part, the right hand Cun part, the right hand Guan part, and the right hand Chi part.

[0104] For example, a multidimensional feature vector X = [X 左寸 , X 左关 , X 左尺 , X 右寸 , X 右尺 , X 右关 ]. Among them, X 左寸 , X 左关 , X 左尺 , X 右寸 , X 右尺 , X 右关 is the above-mentioned part feature matrix.

[0105] Assume that X 部位 Characterization X 左寸 , X 左关 , X 左尺 , X 右寸 , X 右尺, X 右关 Any one of , then:

[0106]

[0107] Among them, V P is the above-mentioned floating index, that is, the above-mentioned pulse position characteristic value, HR is the above-mentioned heart rate, SDNN is the above-mentioned heart rate variability value, M is the above-mentioned pulse shape index, S rise is the above rising slope, S fall is the above-mentioned descending slope, K is the above-mentioned curvature, A is the above-mentioned amplitude, E is the above-mentioned pulse wave energy, and L is the above-mentioned force index.

[0108] In a possible implementation, in order to reduce the complexity of calculation, the multidimensional feature vector can be reduced in dimension. For example, the weight corresponding to each eigenvalue in the multidimensional feature vector can be determined, and the weight can represent the contribution of the eigenvalue to the pulse classification; based on the weight, the multidimensional feature vector is reduced in dimension. Wherein, each eigenvalue is any one of the above-mentioned at least one pulse position eigenvalue, at least one pulse rate eigenvalue, at least one pulse shape eigenvalue and at least one pulse strength eigenvalue.

[0109] When reducing the dimensionality of a multidimensional feature vector, feature importance evaluation can be performed. The electronic device can use statistical methods or algorithms (such as feature importance of random forest) to evaluate the contribution of each eigenvalue in the multidimensional feature vector to the determination of the pulse condition. Based on the contribution of the eigenvalue to the determination of the pulse condition, the electronic device can use methods such as principal component analysis (PCA) or linear discriminant analysis (LDA) to reduce the feature dimension and reduce the complexity of the model.

[0110] S104, determining a training data set and a test data set based on the plurality of multidimensional feature vectors and the pulse category corresponding to each multidimensional feature vector.

[0111] Traditional Chinese medicine can include 28 kinds of pulse conditions, which usually refer to: floating, sinking, slow, rapid, slippery, astringent, weak, strong, long, short, loud, weak, tight, slow, stringy, twisted, leathery, firm, moist, weak, scattered, thin, hidden, moving, hurried, knotted, intermittent, and large. For each pulse condition, it can be classified into mild, moderate, and strong. Therefore, quantitative classification of pulse conditions can obtain 28*3=84 pulse condition categories. Each group of pulse sample signals can correspond to a pulse condition category.

[0112] When determining the data set, it is necessary to determine the pulse category corresponding to each multidimensional feature vector, that is, to determine the pulse type corresponding to each group of pulse sample signals.

[0113] In a possible implementation, manual calibration can be performed. That is, a Chinese medicine pulse diagnosis expert can determine the pulse category corresponding to each group of pulse sample signals based on the pulse signal. Through expert labeling, the label of the training data is ensured to be accurate and the influence of noise data is reduced. Various pulse types and classifications are covered, the model's ability to recognize different pulses is improved, and a unified labeling standard is provided for the quantitative classification of pulse types.

[0114] In another possible implementation, a dedicated annotation software can be developed to facilitate experts to view signal waveforms and characteristic parameters and to annotate them. Experts use traditional pulse diagnosis and sample pulse signals and extracted characteristic parameters to determine the pulse type and its classification based on pulse characteristics and TCM theory, and select corresponding labels. The annotation results are stored together with the corresponding characteristic parameters and signal data to form a complete data set.

[0115] As an example, the extracted characteristic parameters can be mapped to the TCM pulse type to construct a multidimensional recognition model. The extracted characteristic parameters are mapped to a quantitative classification model of 28 types of pulse.

[0116] For example, the quantitative classification map of floating pulse can be:

[0117] The floating index V 浮高1 、V 浮高2 、V 浮高3 is the preset threshold, and the floating pulse is quantitatively graded according to the preset threshold. For example, the floating pulse, i.e., slightly floating pulse, corresponds to the floating index range of: V p >V 浮高1 , and V p ≤V 浮高2 Moderate floating pulse, that is, floating pulse corresponding to the floating index range: V p >V 浮高2 , and V p ≤V 浮高3 Severe floating pulse, i.e. extremely floating pulse, corresponds to the floating index range of: V p >V 浮高3 .

[0118] The pulse is related to the heart rate. Therefore, the quantitative classification of the pulse can be determined based on the heart rate to determine the preset threshold HR. 正常低 , HR 迟高 , HR 迟低 The quantitative classification of slow pulse is mapped as follows: Slightly slow pulse: HR 迟高 <HR≤HR 正常低 . Slow pulse (moderate slow pulse): HR 迟低 <HR≤HR 迟高 Extremely slow pulse (severe slow pulse): HR≤HR 迟低 .

[0119] Based on clinical data, expert experience, and model training calibration measurements, the threshold values of each characteristic parameter can be determined. For example, V 浮高1 = 0.6, V 浮高2 = 0.7, V 浮高3 = 0.8; HR 正常低 = 60 bpm, HR 迟高 = 55 bpm, HR 迟低 = 50 bpm.

[0120] Each pulse condition category can have a corresponding range of characteristic parameter values. Therefore, a characteristic mapping table can be established, and the characteristic mapping table can include the corresponding range of characteristic parameter values that each pulse condition category can have. A specific example:

[0121] Pulse type Pulse type level Characteristic parameter value range Micro-float Floating pulse Mild <![CDATA[0.6<V p ≤0.7]]> Floating pulse Floating pulse Moderate <![CDATA[0.7<V p ≤0.8 <!-- 8 -->]]> Extremely Floating Floating pulse Severe <![CDATA[0.8≤V p ]]> Slightly delayed Slow pulse Mild 55bpm<HR≤60bpm Slow pulse Slow pulse Moderate 50bpm<HR≤55bpm Very late Slow pulse Severe HR≤50bpm

[0122] To facilitate the automatic execution of the algorithm, mapping algorithm logic can be written for each pulse condition category.

[0123] As an example:

[0124] For each part in [left cun, left guan, left chi, right cun, right guan, right chi]:

[0125] Calculate characteristic parameters V_p, HR, M, K, L, E, SDNN, etc.

[0126] If V_p is within the range of the floating pulse:

[0127] If V_p > V_{\text{floating high 3}}:

[0128] Pulse condition[part] = 'extremely floating'

[0129] Else If V_p > V_{\text{floating high 2}}:

[0130] Pulse condition[part] = 'floating pulse'

[0131] Else If V_p > V_{\text{floating high 1}}:

[0132] Pulse condition[part] ='slightly floating'

[0133] Else If V_p is within the range of the sinking pulse:

[0134] ... (similarly)

[0135] If HR is within the range of the slow pulse:

[0136] If HR ≤ HR_{\text{slow low}}:

[0137] Pulse condition[part] += 'extremely slow'

[0138] Else If HR ≤ HR_{\text{late high}}:

[0139] Pulse condition[location] += 'Slow pulse'

[0140] Else:

[0141] Pulse condition[location] += 'Slightly slow'

[0142] ...(The same applies to other pulse condition types)

[0143] Based on the mapping algorithm logic for each pulse condition category, the comprehensive mapping algorithm can be determined:

[0144] For each location:

[0145] Pulse condition[location] = ""

[0146] Pulse condition[location] += MapFloatingPulse(V_p)

[0147] Pulse condition[location] += MapSlowPulse(HR)

[0148] Pulse condition[location] += MapPulseShape(M, K,...)

[0149] Pulse condition[location] += MapPulseForce(L, E,...)

[0150] S105. Use the training data set and the test data set to train the preset model to obtain a pulse condition classification model.

[0151] The above preset model can be a machine learning algorithm, such as, support vector machine (SVM), random forest (RF), multi-layer perceptron (MLP), etc.

[0152] After determining the model, the preset model can be trained based on the training data set; after each training, the trained preset model can be tested based on the test data set to obtain a test result, and thus it can be determined whether the training is completed based on the test result.

[0153] If the test result is a pass, it can be determined that the current preset model is a pulse condition classification model.

[0154] If the test result is that the test fails, the preset model and the training data set can be updated; the target step and the steps after the target are executed until the pulse classification model is obtained, and the target step is to train the preset model based on the training data set. Wherein, the model has hyperparameters, and during the training process, updating the preset model can be to adjust the hyperparameters of the model, thereby optimizing the model. For example, the hyperparameters of the model can be optimized by methods such as grid search (Grid Search) or random search (Random Search).

[0155] In model training (especially classification models, such as binary classification models), the number of true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN) are key indicators for evaluating model performance. These indicators are mainly determined based on the model's prediction results for samples and the true labels of the samples.

[0156] Among them, when the model is trained for classification tasks, a true positive refers to the model correctly predicting a positive class sample as positive. For example, in a model that identifies whether there is a cat in a picture, if a picture does have a cat (actually positive), and the model also determines that the picture has a cat (predicted as positive), then this prediction result is recorded as a true positive. This shows that the model successfully identified the positive class features on this sample.

[0157] True negative refers to when the model correctly predicts negative samples as negative. Taking the cat image recognition model as an example, when a picture does not have a cat (actually negative), the model also determines that the picture does not have a cat (predicted as negative), this is a true negative. This shows that the model can correctly distinguish samples that do not have positive features.

[0158] A false positive is when the model mistakenly predicts a negative sample as a positive sample. For example, in a cat image recognition model, a picture without a cat (actually a negative class) but the model judges that there is a cat (predicts it as a positive class) is a false positive. This may be because the model misjudged some cat-like features (such as objects with cat-like shapes).

[0159] A false negative is when the model mistakenly predicts a positive sample as a negative sample. In the case of cat image recognition, a picture with a cat (actually a positive sample) but the model judges that there is no cat (predicts a negative sample) is a false negative. This may be due to certain features of the cat (such as a large number of occluded parts) that prevent the model from correctly recognizing the cat.

[0160] After training the model once, the number of true positives, true negatives, false positives, and false negatives can be determined based on the training results, and the accuracy (Accuracy), precision (Precision), recall (Recall) and F1 value (F1-score) of the model can be determined based on the number of true positives, true negatives, false positives, and false negatives.

[0161] in:

[0162]

[0163]

[0164]

[0165]

[0166] The confusion matrix can be displayed through various visualization tools (such as heat maps), where the depth of color can indicate the number of samples or the error rate. This allows you to more intuitively observe which categories the model is prone to error and which categories are accurately predicted. Its advantage is that it can fully display the classification performance of the model. Not only can you see the correct classification of the model (through the elements on the diagonal), but you can also understand the misclassification in detail (the elements on the off-diagonal). This helps model developers determine the direction of model improvement. For example, if a certain category is found to have a large number of false positives or false negatives, you can adjust the model parameters or improve feature engineering in a targeted manner.

[0167] In this embodiment, a confusion matrix can be constructed based on the test results to intuitively display the classification effect of the model on each pulse type, thereby facilitating the adjustment of model parameters and the improvement of feature engineering.

[0168] As an example, feature engineering improvements may include feature dimensionality reduction.

[0169] After iterative training of the preset model, a pulse classification model can be obtained. In the iterative training process, the feature set and model parameters can be adjusted according to the test results, and multiple training and evaluation can be performed. In the process of model training, in order to avoid overfitting, regularization methods or additional training data can be used to prevent the model from overfitting on the training set.

[0170] In the present embodiment, a training method for a pulse classification model is provided. The data used in the training process of the pulse classification model is the data collected by the bone conduction sensor. Since the bone conduction sensor has good sensitivity, more accurate pulse signal data can be collected, so that the pulse classification model obtained by training can be more accurate. In the model training process, the pulse signal can be processed to obtain pulse features. In the embodiment of the present application, in order to improve the accuracy of the model, feature extraction can be performed based on the pulse position, pulse rate, pulse shape, and pulse strength of the pulse signal, so that the obtained pulse features can reflect the time domain features and frequency domain features of the pulse signal. Feature extraction is performed based on multiple features to obtain a multidimensional vector that guarantees multidimensional features, and the model is trained based on the multidimensional vector, so that more feature information can be used in the model training process to make the pulse model more accurate.

[0171] Reference Figure 2 , shows a schematic flow chart of another pulse condition determination method provided by an embodiment of the present application, which may specifically include the following steps:

[0172] S201, collecting a pulse signal to be classified through a bone conduction microphone.

[0173] The execution subject of this embodiment is the above-mentioned electronic device. The pulse signal can be a signal collected by a bone conduction microphone at the cun part of the left hand and the right hand of the subject to be diagnosed.

[0174] S202: extract multiple pulse features of the pulse signal.

[0175] S203: Determine a multi-dimensional feature vector according to the plurality of pulse features.

[0176] S201 - S203 of this embodiment may refer to S101 - S103 of the previous embodiment, and may refer to each other, and will not be described in detail here.

[0177] S204, determining the pulse category corresponding to the pulse signal based on the pulse classification model and the multidimensional feature vector, wherein the pulse classification model is determined by the method described in any one of claims 1-8.

[0178] The trained pulse classification model is used to determine the corresponding pulse category based on the multidimensional feature vector. Therefore, the multidimensional feature vector is input into the pulse classification model, and the corresponding pulse category can be output.

[0179] In one possible implementation, a TCM pulse diagnosis and treatment database may be integrated into the electronic device. After determining the pulse category, the electronic device may also generate a diagnosis report based on the pulse category and the TCM pulse diagnosis and treatment database, thereby providing conditioning and treatment suggestions to the user.

[0180] In this embodiment, the pulse signal can be classified based on the trained pulse classification model. The pulse determination method in the embodiment of the present application can be encapsulated as an algorithm, so that the electronic device can call the algorithm based on the data collected from the bone conduction sensor, thereby realizing automatic pulse diagnosis, reducing manual labor, reducing the subjective influence in the pulse diagnosis process, and improving the accuracy of the pulse diagnosis results.

[0181] The method in the embodiment of the present application can be used to assist TCM diagnosis. Based on the method in the embodiment of the present application, the pulse category of the pulse diagnosis object can be output. Based on the output pulse category, TCM can make a diagnosis and give treatment suggestions.

[0182] It should be noted that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0183] In order to better illustrate the process in the embodiments of the present application, the present application scheme is described below with a specific embodiment. Figure 3 It is a schematic diagram of a pulse condition determination method provided in an embodiment of the present application.

[0184] like Figure 3 As shown, data collection can be performed first, and multiple bone conduction microphones can be placed at the three positions of "Cun, Guan, and Chi" respectively to collect pulse signals.

[0185] After data collection, the collected pulse signal can be preprocessed. Specifically, the electronic device can perform filtering, noise reduction and normalization on the pulse signal.

[0186] Feature extraction can be performed on the preprocessed pulse signal to extract the pulse position, pulse rate, pulse shape, and pulse strength.

[0187] Each pulse signal can be manually calibrated by a Chinese medicine pulse diagnosis expert to determine the pulse condition label of each pulse signal. The pulse condition label is the above-mentioned pulse condition category.

[0188] After manual calibration, the dataset can be constructed. The characteristic parameters are combined with the pulse labels annotated by experts to obtain the training dataset and the test dataset.

[0189] For the collected features, feature selection and feature engineering can be performed to screen out features that are highly correlated with pulse labels for combination; or dimensionality reduction can be performed based on the importance of feature parameters.

[0190] Afterwards, an appropriate machine learning algorithm can be selected and the model can be trained using the training dataset so that the model parameters can be adjusted.

[0191] After each model training, the model can be verified and evaluated. Electronic devices can calculate the accuracy, precision, recall, and F1 value on the test data set, so as to evaluate the model performance.

[0192] If the model performance meets the requirements, the model can be used for classification and diagnosis. The electronic device can classify the pulse condition of new data based on the model.

[0193] After classification and diagnosis, the results can be output. Electronic devices can output diagnosis results, generate pulse diagnosis reports and treatment recommendations.

[0194] If the model performance does not meet the requirements, model optimization can be performed to adjust the model structure and parameters, and iterative training can continue based on the adjusted model until the model performance meets the requirements.

[0195] Reference Figure 4 , shows a schematic diagram of a pulse classification model determination device provided by an embodiment of the present application, which may specifically include a signal acquisition module 41, a feature extraction module 42, a vector determination module 43, a data set determination module 44 and a model training module 45, wherein:

[0196] A signal acquisition module 41 is used to acquire multiple groups of pulse sample signals, each group of the pulse sample signals has a pre-calibrated pulse category, and the pulse sample signals are collected by a bone conduction microphone;

[0197] A feature extraction module 42, used for extracting pulse features of each group of pulse sample signals;

[0198] A vector determination module 43, used to determine a multi-dimensional feature vector corresponding to the pulse sample signal according to the pulse feature;

[0199] A data set determination module 44 is used to determine a training data set and a test data set based on a plurality of the multidimensional feature vectors and the pulse condition category corresponding to each of the multidimensional feature vectors;

[0200] The model training module 45 is used to train the preset model through the training data set and the test data set to obtain the pulse classification model.

[0201] In a possible implementation, each group of the pulse sample signals includes a plurality of part signals, each of the part signals corresponds to a human body part, the pulse features include pulse position features, each of the part signals has at least one pulse position feature value, and the extracting of the pulse features of each group of the pulse sample signals includes:

[0202] When different pressures are applied to any of the human body parts, obtaining the maximum amplitude of the signal of the part under different pressures;

[0203] The sum of the maximum amplitudes of the signals at the parts under different pressures is taken as the target amplitude;

[0204] The ratio of the first amplitude corresponding to the position signal divided by the target amplitude is calculated to obtain the pulse position characteristic value corresponding to the position signal, wherein the first amplitude is the amplitude under a preset first pressure.

[0205] In a possible implementation, each group of the pulse sample signals includes a plurality of part signals, each part signal corresponds to a human body part, the pulse feature includes a pulse count feature, each part signal has the pulse count feature including at least one pulse count feature value, and extracting the pulse feature corresponding to each group of the pulse sample signals includes:

[0206] Determine the peak time corresponding to the peak value of the part signal;

[0207] Calculating the average pulse period of the site signal according to the peak time;

[0208] Calculating the heart rate and heart rate variability of the position signal according to the average pulse cycle;

[0209] The heart rate and the heart rate variability value are both used as the pulse rate characteristic values ​​of the part signal.

[0210] In a possible implementation, each group of the pulse sample signals includes a plurality of part signals, each part signal corresponds to a human body part, the pulse feature includes a pulse shape feature, each part signal has at least one pulse shape feature value, and extracting the pulse feature corresponding to each of the pulse sample signals includes:

[0211] Determining a waveform rise time and a waveform fall time of the part signal according to a waveform corresponding to the part signal;

[0212] Calculating a pulse index based on the waveform rise time and the waveform fall time;

[0213] Calculating the rising slope, falling slope and curvature of the waveform;

[0214] The pulse index, the rising slope, the falling slope and the curvature are all used as the pulse characteristic values ​​of the part signal.

[0215] In a possible implementation, each group of the pulse sample signals includes a plurality of part signals, each part signal corresponds to a human body part, the pulse feature includes a pulse potential feature, each part signal has at least one pulse potential feature value, and extracting the pulse feature corresponding to each group of the pulse sample signals includes:

[0216] Determining a force index corresponding to the part signal according to a waveform rise time and a waveform fall time of the part signal;

[0217] Integrate the square of the site signal to obtain the pulse wave energy of the site signal;

[0218] Calculating the amplitude of the part signal according to the maximum value and the minimum value of the part signal;

[0219] The amplitude, the pulse wave energy and the strength index are all used as the pulse characteristic values ​​of the part signal.

[0220] In a possible implementation, each group of the pulse sample signals includes a plurality of site signals, each site signal has at least one corresponding pulse position characteristic value, at least one pulse rate characteristic value, at least one pulse shape characteristic value and at least one pulse strength characteristic value, and determining the multidimensional characteristic direction corresponding to the pulse sample signal according to the pulse characteristics includes:

[0221] Arrange vertically and sequentially at least one pulse position characteristic value, at least one pulse number characteristic value, at least one pulse shape characteristic value and at least one pulse potential characteristic value of the part signal to obtain a part characteristic matrix corresponding to the part signal;

[0222] Arrange the plurality of part feature matrices corresponding to the plurality of part signals in each group of the pulse sample signals in order from left to right to obtain the multi-dimensional feature vector corresponding to the pulse sample signal.

[0223] In a possible implementation, determining the multidimensional feature vector corresponding to the pulse sample signal according to the pulse feature further includes:

[0224] Determine the weight corresponding to each eigenvalue in the multidimensional eigenvector, wherein the weight is used to characterize the contribution of the eigenvalue to the pulse classification;

[0225] Based on the weights, the multi-dimensional feature vector is reduced in dimension.

[0226] In a possible implementation, the preset model is trained by the training data set and the test data set to obtain a pulse classification model, including:

[0227] Training the preset model based on the training data set;

[0228] Testing the trained preset model based on the test data set to obtain a test result;

[0229] If the test result is that the test passes, then the current preset model is determined to be the pulse classification model;

[0230] If the test result is that the test fails, the preset model and the training data set are updated; the target step and the steps after the target are executed until the pulse classification model is obtained, and the target step is to train the preset model based on the training data set.

[0231] In a possible implementation, the method further includes:

[0232] Determine the pulse category corresponding to each group of pulse sample signals.

[0233] In a possible implementation, the above-mentioned determining the pulse category corresponding to each group of the pulse sample signals includes:

[0234] Based on the characteristic parameter values ​​of each group of pulse sample signals, the pulse category corresponding to the pulse sample signals is determined, wherein the characteristic parameter values ​​include one or more of the pulse position characteristic value, the pulse rate characteristic value, the pulse shape characteristic value, and the pulse strength characteristic value.

[0235] In a possible implementation, determining the pulse category corresponding to the pulse sample signal based on the characteristic parameter value of each group of pulse sample signals includes:

[0236] Determine a pulse category mapping table, wherein the pulse category mapping table includes a characteristic parameter value range corresponding to each pulse category;

[0237] According to the pulse category mapping table and the characteristic parameter value of each group of pulse sample signals, the pulse category corresponding to the characteristic parameter value of each group of pulse sample signals is determined.

[0238] Reference Figure 5 , shows a schematic diagram of a pulse condition determination device provided by an embodiment of the present application, which may specifically include a collection module 51, an extraction module 52, a determination module 53 and a classification module 54, wherein:

[0239] A collection module 51, used for collecting the pulse signal to be classified through a bone conduction microphone;

[0240] An extraction module 52, used for extracting the pulse feature of the pulse signal;

[0241] A determination module 53, used to determine a multi-dimensional feature vector according to the pulse feature;

[0242] The classification module 54 is used to determine the pulse category corresponding to the pulse signal based on the pulse classification model and the multidimensional feature vector. The pulse classification model is determined by the above-mentioned pulse classification model determination method.

[0243] In a possible implementation, the method further includes:

[0244] A diagnosis report is generated based on the pulse category and the TCM pulse diagnosis and treatment database.

[0245] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiment part.

[0246] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 6 As shown, the electronic device 6 of this embodiment includes: at least one processor 60 ( Figure 6 Only one is shown in the figure), a memory 61, and a computer program 62 stored in the memory 61 and executable on the at least one processor 60, wherein the processor 60 implements the steps of any of the above-mentioned method embodiments when executing the computer program 62.

[0247] The electronic device 6 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud electronic device. The electronic device may include, but is not limited to, a processor 60 and a memory 61. Those skilled in the art will appreciate that Figure 6 It is only an example of the electronic device 6 and does not constitute a limitation on the electronic device 6. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, it may also include input and output devices, network access devices, etc.

[0248] The processor 60 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0249] In some embodiments, the memory 61 may be an internal storage unit of the electronic device 6, such as a hard disk or memory of the electronic device 6. In other embodiments, the memory 61 may also be an external storage device of the electronic device 6, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 6. Further, the memory 61 may also include both an internal storage unit of the electronic device 6 and an external storage device. The memory 61 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program, etc. The memory 61 may also be used to temporarily store data that has been output or is to be output.

[0250] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the above-mentioned method embodiments can be implemented.

[0251] An embodiment of the present application provides a computer program product. When the computer program product runs on an electronic device, the electronic device can implement the steps in the above-mentioned method embodiments when executing the computer program product.

[0252] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application is described in detail with reference to the above-mentioned embodiments, a person skilled in the art should understand that the technical solutions described in the above-mentioned embodiments can still be modified, or some of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A method for determining a pulse classification model, characterized in that, include: Acquire multiple groups of pulse sample signals, each group of the pulse sample signals has a pre-calibrated pulse category, and the pulse sample signals are collected by a bone conduction microphone; Extracting pulse features from each group of pulse sample signals; Determining a multidimensional feature vector corresponding to the pulse sample signal according to the pulse feature; Determine a training data set and a test data set based on a plurality of the multidimensional feature vectors and the pulse category corresponding to each of the multidimensional feature vectors; The preset model is trained by using the training data set and the test data set to obtain a pulse classification model.

2. The method according to claim 1, characterized in that Each group of the pulse sample signals includes a plurality of part signals, each of the part signals corresponds to a human body part, the pulse features include pulse position features, each of the part signals has at least one pulse position feature value, and the step of extracting the pulse features of each group of the pulse sample signals includes: When different pressures are applied to any of the human body parts, obtaining the maximum amplitude of the signal of the part under different pressures; The sum of the maximum amplitudes of the signals at the parts under different pressures is taken as the target amplitude; The ratio of the first amplitude corresponding to the position signal divided by the target amplitude is calculated to obtain the pulse position characteristic value corresponding to the position signal, wherein the first amplitude is the amplitude under a preset first pressure.

3. The method according to claim 1, characterized in that Each group of the pulse sample signals includes a plurality of part signals, each part signal corresponds to a human body part, the pulse feature includes a pulse count feature, each part signal has at least one pulse count feature value, and the step of extracting the pulse feature of each group of the pulse sample signals includes: Determine the peak time corresponding to the peak value of the part signal; Calculating the average pulse period of the site signal according to the peak time; Calculating the heart rate and heart rate variability of the position signal according to the average pulse cycle; The heart rate and the heart rate variability value are both used as the pulse rate characteristic values ​​of the part signal.

4. The method according to claim 1, characterized in that Each group of the pulse sample signals includes a plurality of part signals, each part signal corresponds to a human body part, the pulse features include pulse shape features, each part signal has at least one pulse shape feature value, and extracting the pulse features corresponding to each of the pulse sample signals includes: Determining a waveform rise time and a waveform fall time of the part signal according to a waveform corresponding to the part signal; Calculating a pulse index based on the waveform rise time and the waveform fall time; Calculating the rising slope, falling slope and curvature of the waveform; The pulse index, the rising slope, the falling slope and the curvature are all used as the pulse characteristic values ​​of the part signal.

5. The method according to claim 1, characterized in that Each group of the pulse sample signals includes a plurality of part signals, each part signal corresponds to a human body part, the pulse features include pulse potential features, each part signal has at least one pulse potential feature value, and extracting the pulse features corresponding to each group of the pulse sample signals includes: Determining a force index corresponding to the part signal according to a waveform rise time and a waveform fall time of the part signal; Integrate the square of the site signal to obtain the pulse wave energy of the site signal; Calculating the amplitude of the part signal according to the maximum value and the minimum value of the part signal; The amplitude, the pulse wave energy and the strength index are all used as the pulse characteristic values ​​of the part signal.

6. The method according to any one of claims 1 to 5, characterized in that: Each group of the pulse sample signals includes a plurality of position signals, each position signal has at least one corresponding pulse position characteristic value, at least one pulse rate characteristic value, at least one pulse shape characteristic value and at least one pulse strength characteristic value, and determining the multidimensional feature vector corresponding to the pulse sample signal according to the pulse characteristics includes: Arrange vertically and sequentially at least one pulse position characteristic value, at least one pulse number characteristic value, at least one pulse shape characteristic value and at least one pulse potential characteristic value of the part signal to obtain a part characteristic matrix corresponding to the part signal; Arrange the plurality of part feature matrices corresponding to the plurality of part signals in each group of the pulse sample signals in order from left to right to obtain the multi-dimensional feature vector corresponding to the pulse sample signal.

7. The method according to claim 6, characterized in that The step of determining the multidimensional feature vector corresponding to the pulse sample signal according to the pulse feature further comprises: Determine the weight corresponding to each eigenvalue in the multidimensional eigenvector, wherein the weight is used to characterize the contribution of the eigenvalue to the pulse classification; Based on the weights, the multi-dimensional feature vector is reduced in dimension.

8. The method according to any one of claims 1 to 5 or 7, characterized in that: The preset model is trained by the training data set and the test data set to obtain a pulse classification model, including: Training the preset model based on the training data set; Testing the trained preset model based on the test data set to obtain a test result; If the test result is that the test passes, then the current preset model is determined to be the pulse classification model; If the test result is that the test fails, the preset model and the training data set are updated; the target step and the steps after the target are executed until the pulse classification model is obtained, and the target step is to train the preset model based on the training data set.

9. A pulse condition determination method, characterized in that: include: The pulse signal to be classified is collected through a bone conduction microphone; extracting a pulse feature of the pulse signal; Determining a multidimensional feature vector according to the pulse feature; Based on the pulse classification model and the multidimensional feature vector, the pulse category corresponding to the pulse signal is determined, and the pulse classification model is determined by the method described in any one of claims 1-8.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 9 is implemented.

Images