Method and system for acquiring detection data based on diabetic pulse characteristics

Abstract

The application provides a detection data acquisition method based on diabetes pulse characteristics, which extracts multiple pulse signal characteristics, realizes multiple pulse characteristic fusion representation according to a fusion strategy, and the fusion representation is the finally obtained pulse diagnosis data. The application extracts pulse characteristics from three different perspectives, obtains a fusion matrix through complementary learning, and realizes the fusion representation of the pulse signal by relying on the fusion matrix. Thus, the application solves the limitation of single pulse characteristic in pulse representation, sets independent projection matrices for various characteristics, projects each characteristic to a public subspace, fully utilizes the complementary information between multiple characteristics, and retains the difference. The representation information in the pulse signal is effectively retained. Meanwhile, the application also provides a detection data acquisition system based on diabetes pulse characteristics.

Description

Technical Field

[0001] This invention relates to the field of digital healthcare. Specifically, it relates to a method and system for acquiring detection data based on the pulse characteristics of diabetic patients. Background Technology

[0002] In the diagnosis of diabetes, extracting pulse characteristics is the primary basis for obtaining sufficient pulse data. Common features include wavelet features and Gaussian features. However, each feature has certain differences and complementarities. Existing single features cannot fully represent pulse signals, thus requiring multiple features to work together. Therefore, after feature extraction, machine learning methods are needed to classify and identify pulse signals. However, current methods using single pulse features for diabetes detection have limitations and neglect the complementarity of other features, failing to fully preserve the effective characteristics of the pulse signal. Summary of the Invention

[0003] The purpose of this invention is to provide a method for acquiring detection data based on the pulse characteristics of diabetes. By setting independent projection matrices for various features, each feature is projected onto a subspace, fully utilizing the complementary information between multiple features while preserving their differences. This effectively preserves the representational information in the pulse signal.

[0004] A first aspect of the present invention provides a method for acquiring detection data based on the pulse characteristics of diabetes, comprising:

[0005] Step S101: Obtain the pulse training sample set. The pulse training sample set contains a predetermined number of training samples. Each pulse training sample set has a predetermined sample collection time. Each pulse training sample in the pulse training sample set includes a sample label Y that serves as an identifier. train The pulse training sample set includes training samples of pulses from diabetic and healthy individuals, with a set sample size.

[0006] Step S102: Extract waveform features, periodicity characteristics, and STFT features from the pulse training sample set.

[0007] Step S103: Project the waveform features, periodicity characteristics, and STFT features of the pulse training sample set to a common projection subspace using an optimization function. The optimization function includes the projection matrix D of the waveform features, periodicity characteristics, and STFT features. v The fusion vector matrix Z is obtained by analyzing the waveform characteristics, periodicity, and STFT features after projection. train .

[0008] Step S104, establish fusion feature Z train With sample label Y trianThe mapping function between them. Prior knowledge is used to preserve the pulse data structure and sample distribution information, and the vector matrix Z is fused through dimensionality reduction using the projection matrix. train And projected onto a plane that can match the sample label Y train The corresponding sequence.

[0009] Step S105: Construct the objective function using the optimization function and the mapping function.

[0010] Step S106: Based on the pulse training sample set and pulse training sample labels, obtain the fusion representation through the objective function. Learn to obtain the projection matrix D in the fusion representation. v .

[0011] Through having a projection matrix D v A fusion pulse feature model is constructed to acquire pulse sample data from diabetic patients. Diabetes detection data is then obtained based on this fusion pulse feature model and the currently collected pulse data from diabetic patients.

[0012] In one embodiment of the method for acquiring detection data based on the pulse characteristics of diabetes of the present invention, the sample collection time is 30s.

[0013] In one embodiment of the detection data acquisition method based on diabetic pulse characteristics of the present invention, the fusion vector matrix Z in step S103 train This is represented by Formula 1.

[0014]

[0015] Where η is the weight coefficient of the regularization term, and Z is the weight coefficient of the regularization term. train This is the common subspace after projection.

[0016] In another embodiment of the method for acquiring detection data based on diabetic pulse characteristics of the present invention, step S104 includes: using the mapping function shown in Formula 2, preserving the pulse wave data characteristics and distribution with the help of prior knowledge, and fusion vector matrix Z through dimensionality reduction using projection matrix P. train And projected onto a plane that can match the sample label Y train The corresponding sequence:

[0017] min||Z train PY train || F +tr(Z train T LZ train )+β||P|| F Formula 2

[0018] Where β is the weight coefficient of the regularization term, L is the graph Laplacian matrix of the training set, and Z... train T Z representstrain The transpose of the matrix, tr() represents the trace of the matrix.

[0019] In another embodiment of the method for acquiring detection data based on diabetic pulse characteristics of the present invention, step S105 includes: constructing an objective function as shown in formula 3 using the optimization function shown in formula 1 and the mapping function shown in formula 2.

[0020]

[0021] In another embodiment of the method for acquiring detection data based on diabetic pulse characteristics of the present invention, the fused pulse characteristic model in step S107 is as shown in Formula 4:

[0022]

[0023] In another embodiment of the method for acquiring detection data based on diabetic pulse characteristics of the present invention, step S106 further includes:

[0024] Step S107: Receive a set number of pulse data points from the patient to be diagnosed. Obtain the current fused pulse characteristic data of the patient to be diagnosed using Formula 4. Compare the fused pulse characteristic data of the diabetic patient with the current fused data to obtain a comparison difference value.

[0025] A second aspect of the present invention provides a detection data acquisition system based on the pulse characteristics of diabetes, comprising:

[0026] The sample set acquisition unit is configured to acquire a pulse training sample set. The pulse training sample set contains a predetermined number of training samples. Each pulse training sample set has a predetermined sample acquisition time. Each pulse training sample in the pulse training sample set includes a sample label Y that serves as an identifier. train The pulse training sample set includes training samples of pulses from diabetic and healthy individuals, with a set sample size.

[0027] The feature extraction unit is configured to extract waveform features, periodicity characteristics, and STFT features from the pulse training sample set.

[0028] The projection unit is configured to project the waveform features, periodicity characteristics, and STFT features of the pulse training sample set to a common projection subspace through an optimization function. The optimization function includes the projection matrix D of the waveform features, periodicity characteristics, and STFT features. v The fusion vector matrix Z is obtained by analyzing the waveform characteristics, periodicity, and STFT features after projection. train .

[0029] Mapping unit, configured to establish fused feature Z trainWith sample label Y trian The mapping function between them. The vector matrix Z is fused by dimensionality reduction through the projection matrix. train And projected onto a plane that can match the sample label Y train The corresponding sequence.

[0030] The objective function acquisition unit is configured to construct the objective function through an optimization function and a mapping function.

[0031] The diabetes detection data acquisition unit is configured to obtain a fusion representation based on a pulse training sample set and pulse training sample labels, using an objective function. It then learns to obtain the projection matrix D within the fusion representation. v .

[0032] Through having a projection matrix D v A fusion pulse feature model is constructed to acquire pulse sample data from diabetic patients. Diabetes detection data is then obtained based on this fusion pulse feature model and the currently collected pulse data from diabetic patients.

[0033] In another embodiment of the detection data acquisition system based on the pulse characteristics of diabetes of the present invention, the sample collection time is 30s.

[0034] In another embodiment of the detection data acquisition system based on diabetic pulse characteristics of the present invention, the fusion vector matrix Z in the projection unit train This is represented by Formula 5.

[0035]

[0036] Where η is the weight coefficient of the regularization term, and Z is the weight coefficient of the regularization term. train This is the common subspace after projection.

[0037] The mapping unit includes the mapping function shown in Equation 6. It preserves data structure and sample distribution information using prior knowledge, and fuses the vector matrix Z through dimensionality reduction using the projection matrix P. train And projected onto a plane that can match the sample label Y train The corresponding sequence:

[0038] min||Z train PY train || F +tr(Z train T LZ train )+β||P|| F Formula 6

[0039] Where β is the weight coefficient of the regularization term, L is the graph Laplacian matrix of the training set, and Z... train T Z represents trainThe transpose of the matrix, tr() represents the trace of the matrix.

[0040] The following text will further explain the characteristics, technical features, advantages, and implementation of the detection data acquisition method and system based on the pulse characteristics of diabetes in a clear and easy-to-understand manner, with the aid of accompanying figures. Attached Figure Description

[0041] Figure 1 This is a flowchart illustrating the method for acquiring detection data based on the pulse characteristics of diabetes in an embodiment of the present invention.

[0042] Figure 2 This is a flowchart illustrating a method for acquiring detection data based on diabetic pulse characteristics in another embodiment of the present invention.

[0043] Figure 3 This is a flowchart illustrating the detection data acquisition system based on the pulse characteristics of diabetes in a further embodiment of the present invention.

[0044] Figure 4 This is a waveform diagram used to illustrate the pulse training sample in an embodiment of the present invention.

[0045] Figure 5 This is a schematic diagram illustrating the Gabor waveform feature in an embodiment of the present invention.

[0046] Figure 6 This is a schematic diagram illustrating the short-time Fourier transform (STFT) characteristics in an embodiment of the present invention.

[0047] Figure 7 This is a schematic diagram illustrating the periodicity characteristic of a 2D matrix in an embodiment of the present invention.

[0048] Figure 8 This is a schematic diagram illustrating the fusion representation features of pulse signals in an embodiment of the present invention.

[0049] Figure 9 This is a flowchart illustrating a method for acquiring detection data based on diabetic pulse characteristics in another embodiment of the present invention. Detailed Implementation

[0050] To provide a clearer understanding of the technical features, objectives, and effects of the invention, specific embodiments of the invention are now described with reference to the accompanying drawings. In the drawings, the same reference numerals indicate components with the same or similar structures but the same function.

[0051] In this document, "illustrative" means "serving as an example, illustration, or description," and any illustrations or embodiments described herein as "illustrative" should not be construed as preferred or more advantageous technical solutions. For the sake of brevity, each figure only schematically shows the parts relevant to this exemplary embodiment, and they do not represent the actual structure or true proportions of the product.

[0052] In a first aspect, the present invention provides a method for acquiring detection data based on the pulse characteristics of diabetic patients, such as... Figure 1 As shown, the methods for acquiring detection data based on diabetic pulse characteristics include:

[0053] Step S101: Obtain the pulse training sample set.

[0054] In this step, a pulse training sample set is obtained. The pulse training sample set contains a predetermined number of training samples. Each pulse training sample set has a predetermined sample collection time. Each pulse training sample in the pulse training sample set includes a sample label Y that serves as an identifier. train The pulse training sample set includes training samples of pulses from diabetic and healthy individuals, with a set sample size.

[0055] Step S102: Extract features.

[0056] In this step, waveform features (Gabor features), periodicity characteristics (2DMatrix), and STFT features (STFT short-time Fourier transform features) of the pulse training sample set are extracted.

[0057] Step S103: Project to the common subspace.

[0058] In this step, the waveform features, periodicity characteristics, and STFT features of the pulse training sample set are projected to a common projection subspace using an optimization function. The optimization function includes the projection matrix D of the waveform features, periodicity characteristics, and STFT features. v The fusion vector matrix Z is obtained by analyzing the waveform characteristics, periodicity, and STFT features after projection. train .

[0059] Step S104: Establish the mapping function.

[0060] In this step, the fusion feature Z is established. train With sample label Y trian The mapping function between them. Prior knowledge is used to preserve the pulse data structure and sample distribution information, and the vector matrix Z is fused through dimensionality reduction using the projection matrix. train And projected onto a plane that can match the sample label Y train The corresponding sequence.

[0061] Step S105: Construct the objective function.

[0062] In this step, the objective function is constructed by optimizing the function and mapping the function.

[0063] Step S106: Obtain the projection matrix and acquire diabetes detection data.

[0064] In this step, based on the pulse training sample set and its labels, a fusion representation is obtained through an objective function. The projection matrix D in the fusion representation is then learned. v .

[0065] Through having a projection matrix D v A fusion pulse feature model is constructed to acquire pulse sample data from diabetic patients. Diabetes detection data is then obtained based on this fusion pulse feature model and the currently collected pulse data from diabetic patients.

[0066] In one embodiment of the method for acquiring detection data based on the pulse characteristics of diabetes of the present invention, the sample collection time is 30s.

[0067] In one embodiment of the detection data acquisition method based on diabetic pulse characteristics of the present invention, the fusion vector matrix Z in step S103 train This is represented by Formula 1.

[0068]

[0069] Where η is the weight coefficient of the regularization term, and Z is the weight coefficient of the regularization term. train This is the common subspace after projection.

[0070] In another embodiment of the method for acquiring detection data based on diabetic pulse characteristics of the present invention, step S104 includes: using the mapping function shown in Formula 2, preserving the pulse data structure and sample distribution information with the help of prior knowledge, and fusion of the vector matrix Z by reducing the dimension of the projection matrix P. train And projected onto a plane that can match the sample label Y train The corresponding sequence:

[0071] min||Z train PY train || F +tr(Z train T LZ train )+β||P|| F Formula 2

[0072] Where β is the weight coefficient of the regularization term, L is the graph Laplacian matrix of the training set, and Z... train T Z represents trainThe transpose of the matrix, tr() represents the trace of the matrix.

[0073] In another embodiment of the method for acquiring detection data based on diabetic pulse characteristics of the present invention, step S105 includes: constructing an objective function as shown in formula 3 using the optimization function shown in formula 1 and the mapping function shown in formula 2.

[0074]

[0075] In another embodiment of the method for acquiring detection data based on diabetic pulse characteristics of the present invention, the fused pulse characteristic model in step S107 is as shown in Formula 4:

[0076]

[0077] In another embodiment of the method for acquiring detection data based on diabetic pulse characteristics of the present invention, such as Figure 2 As shown, after step S106, the following is also included:

[0078] Step S107: Obtain the comparison difference value.

[0079] In this step, a set number of pulse data points are received from the patient to be diagnosed. The current fused pulse characteristic data of the patient to be diagnosed is obtained using Formula 4. The fused pulse characteristic data of the diabetic patient is compared with the current fused data to obtain a comparison difference value.

[0080] A second aspect of the present invention provides a detection data acquisition system based on the pulse characteristics of diabetes, such as... Figure 3 As shown, the detection data acquisition system based on diabetes pulse characteristics includes: a sample set acquisition unit 101, a feature extraction unit 102, a projection unit 103, a mapping unit 104, an objective function acquisition unit 105, and a diabetes detection data acquisition unit 106. Wherein:

[0081] The sample set acquisition unit 101 is configured to acquire a pulse training sample set. The pulse training sample set contains a predetermined number of training samples. Each pulse training sample set has a predetermined sample acquisition time. Each pulse training sample in the pulse training sample set includes a sample label Y that serves as an identifier. train The pulse training sample set includes training samples of pulses from diabetic and healthy individuals, with a set sample size.

[0082] The feature extraction unit 102 is configured to extract waveform features, periodicity characteristics and STFT features from the pulse training sample set.

[0083] Projection unit 103 is configured to project the waveform features, periodicity characteristics, and STFT features of the pulse training sample set to a common projection subspace through an optimization function. The optimization function includes the projection matrix D of the waveform features, periodicity characteristics, and STFT features. v The fusion vector matrix Z is obtained by analyzing the waveform characteristics, periodicity, and STFT features after projection. train .

[0084] Mapping unit 104, configured to establish fusion feature Z train With sample label Y trian The mapping function between them. The vector matrix Z is fused by dimensionality reduction through the projection matrix. train And projected onto a plane that can match the sample label Y train The corresponding sequence.

[0085] The objective function acquisition unit 105 is configured to construct the objective function through an optimization function and a mapping function.

[0086] The diabetes detection data acquisition unit 106 is configured to obtain a fusion representation based on a pulse training sample set and pulse training sample labels through an objective function. It then learns to obtain the projection matrix D in the fusion representation. v .

[0087] Through having a projection matrix D v A fusion pulse feature model is constructed to acquire pulse sample data from diabetic patients. Diabetes detection data is then obtained based on this fusion pulse feature model and the currently collected pulse data from diabetic patients.

[0088] In another embodiment of the detection data acquisition system based on the pulse characteristics of diabetes of the present invention, the sample collection time is 30s.

[0089] In another embodiment of the detection data acquisition system based on diabetic pulse characteristics of the present invention, the fusion vector matrix Z in the projection unit 103 train This is represented by Formula 5.

[0090]

[0091] Where η is the weight coefficient of the regularization term, and Z is the weight coefficient of the regularization term. train This is the common subspace after projection.

[0092] Mapping unit 104 includes: a mapping function as shown in Equation 6. It preserves the pulse data structure and sample distribution information using prior knowledge, and fuses the vector matrix Z through dimension reduction using the projection matrix P. train And projected onto a plane that can match the sample label Y train The corresponding sequence:

[0093] min||Ztrain PY train || F +tr(Z train T LZ train )+β||P|| F Formula 6

[0094] Where β is the weight coefficient of the regularization term, L is the graph Laplacian matrix of the training set, and Z... train T Z represents train The transpose of the matrix, tr() represents the trace of the matrix.

[0095] In a preferred embodiment of the method for acquiring detection data based on diabetic pulse characteristics of the present invention, the method aims to extract diabetic diagnostic data based on multimodal feature fusion, and realizes the fusion representation of pulse signals in three steps, as follows: Figure 9 As shown, firstly, the three pulse signal features are extracted; secondly, the multiple pulse features are fused and represented according to the fusion strategy; the fused representation is the final pulse diagnosis data.

[0096] This invention proposes a method for acquiring detection data based on the pulse characteristics of diabetes. It proposes to extract pulse characteristics from three different perspectives, obtain a fusion matrix through complementary learning, and realize the fusion representation of pulse signals by relying on the fusion matrix.

[0097] Gabor features, 2D matrix, and STFT features are three commonly used features in pulse diagnosis. Gabor features represent the peaks and troughs of the pulse signal, 2D matrix reflects the periodic and non-periodic characteristics of the pulse signal, and STFT features... Therefore, these three features can fully express the characteristics of the pulse and are used for feature recognition. A schematic diagram of the waveforms of the collected pulse training samples is shown below. Figure 4 As shown. The Gabor feature of the pulse signal has a dimension of 384, as shown below. Figure 5 The dimension of the STFT feature is 150, such as Figure 6 The dimension of a 2D matrix feature is 325, such as Figure 7 Based on the above individual pulse characteristics, the fused representation of the pulse signal can be obtained as a dimension 22, such as... Figure 8 .

[0098] The STFT features are shown in Formula 7 below:

[0099]

[0100] Where Z(t) is the source signal and g(t) is the window function.

[0101] Gabor characteristics are shown in the following formula 8:

[0102]

[0103] in, Here, s is a Gaussian window function, s is the scaling factor, u is the displacement factor, and s is the frequency factor.

[0104] Suppose there are n case samples, and the dataset of all samples can be defined as X. x1 represents a single sample, and the training data can also be represented as... Let m represent the i-th feature of all pulse data. i Let be the dimension of the i-th feature, and n be the number of samples. The corresponding label can be represented as... y n is the label of the nth sample, and c is the number of categories.

[0105] The pulse fusion representation strategy in this invention is as follows:

[0106] First, due to the complementarity and differences between modal information, in order to preserve this information, each modal feature X is... V Design different projection matrices D V Each modality is projected onto a common shared subspace Z. After projection, the dimensionality-reduced data is synthesized into a fusion vector matrix Z. train Meanwhile, to ensure data sparsity, the F-norm was used to optimize the fusion effect. The resulting optimization function is shown in Equation 9:

[0107]

[0108] Secondly, in order to establish the fusion feature Z train With sample label Y train The mapping relationship between them is preserved by using prior knowledge to retain the pulse data structure and sample distribution information, and the fusion matrix Z is fused through the projection matrix P. train Dimension reduction and projection onto label Y train The specific implementation process is shown in Formula 10:

[0109] min||Z train PY train || F +tr(Z train T LZ train )+β||P|| F Formula 10

[0110] Where β is the weight coefficient of the regularization term, L is the graph Laplacian matrix of the training set, and Z... trainT Z represents train The transpose of the matrix, tr() represents the trace of the matrix.

[0111] Next, the objective function of this fusion method is constructed. The objective function is shown in Formula 11:

[0112]

[0113] To obtain each projection matrix D v With the mapping matrix P, an iterative strategy is adopted to optimize each mapping matrix, while the Shelvis equation is introduced to obtain explicit solutions for each variable.

[0114] The final fusion representation of the pulse signal can be calculated using Equation 12.

[0115]

[0116] First, it addresses the limitations of single pulse features in pulse representation by setting independent projection matrices for various features, projecting each feature onto a subspace, fully utilizing the complementary information between multiple features while preserving their differences. This effectively preserves the representational information in the pulse signal.

[0117] Secondly, using a mapping matrix, a mapping relationship between the fused feature vector and the sample labels was established, and the projection matrix was optimized using a regularization term, further optimizing the pulse fusion feature Z. train .

[0118] This invention selects three features to express pulse information from three perspectives, and these three types of information can be used to comprehensively express pulse signals.

[0119] A multi-feature fusion strategy is proposed, which designs an independent projection matrix for each feature and effectively fuses the features. The mapping relationship between the fusion vector and the label is established using the projection matrix P, which effectively realizes the fusion representation of the pulse signal.

[0120] This invention aims to achieve diabetes diagnosis based on multimodal feature fusion, which involves two steps to achieve feature fusion of pulse signals, as described below. Figure 9 As shown, firstly, the three pulse signal features are extracted; secondly, a fusion representation of multiple pulse features is achieved based on a fusion strategy.

[0121] 1. Select signals with a length of 30s for each sample and extract Gabor, 2D-Matrix, and STFT features according to the formula.

[0122] 2. Based on Formula 9, project each feature to a common projection subspace, and calculate the fused feature Z by summing the features.train By utilizing the F-norm, the sparsity of each matrix is ​​guaranteed.

[0123] 3. Furthermore, in order to establish the mapping relationship between the fused features and the labels, the fused features are linked to the labels L of the samples through the projection matrix P, and a regularization term is added. The process is as shown in Formula 10.

[0124] 4. Feature fusion and mapping relationship, construct the objective function, as shown in Formula 11.

[0125] 5. Optimize and solve for each variable, and obtain the fused pulse feature Z according to Formula 12. train .

[0126] It should be understood that although this specification describes various embodiments, not every embodiment contains only one independent technical solution. This way of describing the specification is only for clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

[0127] The detailed descriptions listed above are merely specific descriptions of feasible embodiments of the present invention, and are not intended to limit the scope of protection of the present invention. All equivalent embodiments or modifications made without departing from the spirit of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for acquiring detection data based on the pulse characteristics of diabetes, characterized in that, It includes: Step S101: Obtain the pulse training sample set; The pulse training sample set contains a set number of training samples; Each pulse training sample set has a set sample collection time; Each pulse training sample in the pulse training sample set includes a sample label Y that serves as an identifier. train The pulse training sample set includes a set number of training samples of pulses from diabetic and healthy individuals. Step S102: Extract the waveform features, periodicity characteristics, and STFT features of the pulse training sample set; Step S103: Project the waveform features, periodicity characteristics, and STFT features of the pulse training sample set to a common projection subspace using an optimization function. The optimization function includes a projection matrix D of the waveform features, periodicity characteristics, and STFT features. v The fusion vector matrix Z is obtained by analyzing the waveform characteristics, periodicity, and STFT features after projection. train ; Step S104, establish the fusion feature Z train With the sample label Y trian The mapping function between them; adding prior knowledge to the training sample data, and preserving the distribution characteristics of the training pulse data through the graph Laplacian matrix; further, reducing the dimensionality of the fusion vector matrix Z through the projection matrix. train And projected onto a plane that can be aligned with the sample label Y train The corresponding sequence; Step S105: Construct the objective function using the optimization function and the mapping function; Step S106: Based on the pulse training sample set and pulse training sample labels, obtain the fusion representation through the objective function; learn to obtain the projection matrix D in the fusion representation. v ; By having the projection matrix D v A fusion pulse feature model is constructed to obtain pulse sample data of diabetic patients using a fusion representation method; diabetes detection data is obtained based on the fusion pulse feature model and the currently collected pulse data of diabetic patients.

2. The method for acquiring detection data based on diabetic pulse characteristics according to claim 1, characterized in that, The sample collection time was 30 seconds.

3. The method for acquiring detection data based on diabetic pulse characteristics according to claim 1, characterized in that, The fusion vector matrix Z in step S103 train Represented as Formula 1; Where η is the weight coefficient of the regularization term, and Z is the weight coefficient of the regularization term. train This is the common subspace after projection.

4. The method for acquiring detection data based on diabetic pulse characteristics according to claim 3, characterized in that, Step S104 includes: using the mapping function shown in Formula 2; adding prior knowledge to the training sample data; preserving the distribution characteristics of the training sample data through the graph Laplacian matrix; and reducing the dimensionality of the fusion vector matrix Z through the projection matrix P. train And projected onto a plane that can be aligned with the sample label Y train The corresponding sequence: min||Z train P - Y train || F +tr(Z train T LZ train ) + β||P|| F Equation 2 Where β is the weight coefficient of the regularization term, L is the graph Laplacian matrix of the training set, and Z... train T Z represents train The transpose of the matrix, tr() represents the trace of the matrix.

5. The method for acquiring detection data based on diabetic pulse characteristics according to claim 4, characterized in that, Step S105 includes: constructing the objective function shown in Formula 3 using the optimization function shown in Formula 1 and the mapping function shown in Formula 2. 。 6. The method for acquiring detection data based on diabetic pulse characteristics according to claim 5, characterized in that, The fusion pulse feature model in step S107 is shown in Formula 4: 。 7. The method for acquiring detection data based on diabetic pulse characteristics according to claim 1, characterized in that, Step S106 is followed by: Step S107: Receive a set number of pulse data from the current patient to be diagnosed; obtain the current fusion pulse feature data of the current patient to be diagnosed using Formula 4; compare the fusion pulse feature data of the diabetic patient with the current fusion data to obtain a comparison difference value.

8. A detection data acquisition system based on the pulse characteristics of diabetes, characterized in that, It includes: The sample set acquisition unit is configured to acquire a pulse training sample set. The pulse training sample set contains a set number of training samples; Each pulse training sample set has a set sample collection time; each pulse training sample in the pulse training sample set includes a sample label Y that serves as an identifier. train The pulse training sample set includes a set number of training samples of pulses from diabetic and healthy individuals. The feature extraction unit is configured to extract waveform features, periodicity characteristics, and STFT features from the pulse training sample set. The projection unit is configured to project the waveform features, periodicity characteristics, and STFT features of the pulse training sample set to a common projection subspace using an optimization function. The optimization function includes a projection matrix D of the waveform features, periodicity characteristics, and STFT features. v The fusion vector matrix Z is obtained by analyzing the waveform characteristics, periodicity, and STFT features after projection. train ; Mapping unit, configured to establish the fusion feature Z train With the sample label Y trian The mapping function between them; the fusion vector matrix Z is reduced in dimensionality by the projection matrix. train And projected onto a plane that can be aligned with the sample label Y train The corresponding sequence; The objective function acquisition unit is configured to construct an objective function using the optimization function and the mapping function. The diabetes detection data acquisition unit is configured to obtain a fusion representation based on a pulse training sample set and pulse training sample labels, using the objective function; and to learn and obtain the projection matrix D in the fusion representation. v ; By having the projection matrix D v A fusion pulse feature model is constructed to obtain pulse sample data of diabetic patients using a fusion representation method; diabetes detection data is obtained based on the fusion pulse feature model and the currently collected pulse data of diabetic patients.

9. The detection data acquisition system based on diabetic pulse characteristics according to claim 8, characterized in that, The sample collection time was 30 seconds.

10. The detection data acquisition system based on diabetic pulse characteristics according to claim 8, characterized in that, The fusion vector matrix Z in the projection unit train Represented as Formula 5; Where η is the weight coefficient of the regularization term, and Z is the weight coefficient of the regularization term. train This is the common subspace after projection; The mapping unit includes: a mapping function as shown in Equation 6; adding prior knowledge to the training sample data; preserving the distribution characteristics of the training sample data through the Graph Laplacian matrix; and reducing the dimensionality of the fusion vector matrix Z through the projection matrix P. train And projected onto a plane that can be aligned with the sample label Y train The corresponding sequence: min||Z train P - Y train || F +tr(Z train T LZ train ) + β||P|| F Formula 6 Where β is the weight coefficient of the regularization term, L is the graph Laplacian matrix of the training set, and Z... train T Z represents train The transpose of the matrix, tr() represents the trace of the matrix.

Images