A desktop health monitor

Abstract

Disclosed is a desktop health monitor, wherein the desktop health monitor can monitor and evaluate the human health status of the monitored individual under the accompaniment of a small number of medical staff or under the self-operation of the monitored individual. Specifically, a convolutional neural network model is used to extract high-dimensional local implicit correlation implicit features of blood sugar and blood pressure in the blood oxygen association matrix and heart rate and pulse in the electrocardiogram association matrix in different time and space domains to obtain a multi-scale blood oxygen feature matrix and a multi-scale electrocardiogram feature matrix, and then the classification feature matrix obtained by fusion is passed through a classifier to obtain a classification result. In this way, the health status assessment is performed based on multiple health monitoring parameters of the monitored individual to improve the confidence and accuracy of the health status assessment.

Description

Technical Field

[0001] The present application relates to the field of smart healthcare, and more specifically to a desktop health monitor. Background Art

[0002] With the development of social economy and the accelerated pace of life, people are paying more and more attention to their physical health. Regular health checks can help people understand their health status, detect early disease risks in time, and cooperate with medical treatment to obtain better and more adequate treatment. Going to a hospital or related medical institution for a physical examination is the most common way for people to have a physical examination in their daily lives, and it is also one of the ways to obtain a more professional and comprehensive examination.

[0003] At present, the problems brought about by this type of health examination method are long waiting times, long examination processes, long waiting times for results, and high fees. At the same time, medical staff are required to pay attention to and participate in the physical examination at all times. Obviously, this does not conform to the current situation of unbalanced doctor-patient ratio and equipment shortage.

[0004] Therefore, a desktop health monitor is desired, which can monitor the health status of the individual to be monitored under the accompaniment of a small number of medical staff or under the self-operation of the individual to be monitored. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. An embodiment of the present application provides a desktop health monitor, wherein the desktop health monitor can monitor and evaluate the human health status of the monitored individual itself under the accompaniment of a small number of medical staff or under the self-operation of the monitored individual. Specifically, a convolutional neural network model is used to extract the high-dimensional local implicit correlation implicit features of blood sugar and blood pressure in the blood oxygen association matrix and heart rate and pulse in the electrocardiogram association matrix in different time and space domains to obtain a multi-scale blood oxygen feature matrix and a multi-scale electrocardiogram feature matrix, and then the classification feature matrix obtained by fusion is passed through a classifier to obtain a classification result. In this way, the health status assessment is performed based on multiple health monitoring parameters of the monitored individual to improve the confidence and accuracy of the health status assessment.

[0006] According to one aspect of the present application, there is provided a desktop health monitor, comprising:

[0007] A monitoring unit, used to obtain a plurality of health parameters of a human body to be monitored at a plurality of predetermined time points within a predetermined time period, wherein the plurality of health parameters include blood sugar value, blood pressure value, heart rate value and pulse value;

[0008] A data construction unit, used for arranging the multiple health parameters of the human body to be monitored at multiple predetermined time points within a predetermined time period into input vectors according to the time dimension to obtain a blood sugar vector, a blood pressure vector, a heart rate vector and a pulse vector;

[0009] A correlation unit, used to construct a blood oxygen correlation matrix between the blood sugar vector and the blood pressure vector and to construct an electrocardiogram correlation matrix between the heart rate vector and the pulse vector;

[0010] A multi-scale convolutional encoding unit, used for respectively passing the blood oxygen correlation matrix and the electrocardiogram correlation matrix through a convolutional neural network model with a multi-scale convolutional structure to obtain a multi-scale blood oxygen feature matrix and a multi-scale electrocardiogram feature matrix;

[0011] A feature fusion unit, used for fusing the multi-scale blood oxygen feature matrix and the multi-scale electrocardiogram feature matrix to obtain a classification feature matrix; and

[0012] The monitoring result generating unit is used to pass the classification feature matrix through a classifier to obtain a classification result, and the classification result is used to indicate whether the health condition of the human body to be monitored is normal.

[0013] In the above-mentioned desktop health monitor, the association unit includes: a first association subunit, used to calculate the product between the transposed vector of the blood glucose vector and the blood pressure vector to obtain the blood oxygen association matrix; and a second association subunit, used to calculate the product between the transposed vector of the heart rate vector and the pulse vector to obtain the electrocardiogram association matrix.

[0014] In the above-mentioned desktop health monitor, the multi-scale convolution encoding unit includes: a blood oxygen correlation feature extraction subunit, which is used to use each layer of the convolutional neural network model to perform convolution processing on the input data in the forward transfer of the layer: using the first convolution kernel of the multi-scale convolution structure to convolve the input data to obtain a first convolution feature map; using the second convolution kernel of the multi-scale convolution structure to convolve the input data to obtain a second convolution feature map; using the third convolution kernel of the multi-scale convolution structure to convolve the input data to obtain a third convolution feature map; using the fourth convolution kernel of the multi-scale convolution structure to convolve the input data to obtain a fourth convolution feature map; to cascade the first convolution feature map, the second convolution feature map, the third convolution feature map and the fourth convolution feature map to obtain a multi-scale convolution feature map; to perform pooling processing on the multi-scale convolution feature map to obtain a pooled feature map; and to perform nonlinear activation processing on the pooled feature map to obtain an activation feature map; wherein the activation feature map output by the last layer of the convolution neural network model is the multi-scale blood oxygen feature matrix; wherein the size of the first convolution kernel is 7×7, the size of the second convolution kernel is 5×5, the size of the third convolution kernel is 3×3, and the size of the fourth convolution kernel is 1×1.

[0015] In the above-mentioned desktop health monitor, the multi-scale convolution encoding unit also includes: an ECG-related feature extraction subunit, which uses each layer of the convolutional neural network model to perform the following operations on the input data in the forward transfer of the layer: using the first convolution kernel of the multi-scale convolution structure to perform convolution processing on the input data to obtain a first convolution feature map; using the second convolution kernel of the multi-scale convolution structure to perform convolution processing on the input data to obtain a second convolution feature map; using the third convolution kernel of the multi-scale convolution structure to perform convolution processing on the input data to obtain a third convolution feature map; using the fourth convolution kernel of the multi-scale convolution structure to perform convolution processing on the input data; to obtain a fourth convolution feature map; to cascade the first convolution feature map, the second convolution feature map, the third convolution feature map and the fourth convolution feature map to obtain a multi-scale convolution feature map; to perform pooling processing on the multi-scale convolution feature map to obtain a pooled feature map; and to perform nonlinear activation processing on the pooled feature map to obtain an activation feature map; wherein the activation feature map output by the last layer of the convolution neural network model is the multi-scale ECG feature matrix; wherein the size of the first convolution kernel is 7×7, the size of the second convolution kernel is 5×5, the size of the third convolution kernel is 3×3, and the size of the fourth convolution kernel is 1×1.

[0016] In the above-mentioned desktop health monitor, the feature fusion unit is further used to: fuse the multi-scale blood oxygen feature matrix and the multi-scale electrocardiogram feature matrix to obtain a classification feature matrix according to the following formula; wherein the formula is:

[0017]

[0018] Among them, M c is the classification feature matrix, M a is the multi-scale blood oxygen feature matrix, M b is the multi-scale ECG feature matrix, represents the addition of elements at corresponding positions of the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix, and α and β are weighted parameters for controlling the balance between the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix in the classification feature matrix.

[0019] In the above-mentioned desktop health monitor, the monitoring result generating unit is further used to: use the classifier to process the classification feature matrix according to the following formula to obtain the classification result; wherein the formula is:

[0020] softmax{(W n ,B n ):…:(W1,B1)|Project(M)}

[0021] Project(M) represents the projection of the classification feature matrix into a vector, W1 to W n is the weight matrix of each fully connected layer, B1 to B n Represents the bias vector of each fully connected layer.

[0022] The above-mentioned desktop health monitor also includes a training module for training the convolutional neural network model with a multi-scale convolution structure and the classifier.

[0023] In the above-mentioned desktop health monitor, the training module includes: a training monitoring unit, which is used to obtain training data, wherein the training data includes a plurality of training health parameters of the human body to be monitored at multiple predetermined time points within a predetermined time period, and a real value of whether the health status of the human body to be monitored is normal, wherein the plurality of training health parameters include training blood sugar value, training blood pressure value, training heart rate value and training pulse value; a training data construction unit, which is used to arrange the plurality of training health parameters of the human body to be monitored at multiple predetermined time points within a predetermined time period as input vectors according to the time dimension to obtain training blood sugar vectors, training blood pressure vectors, training heart rate vectors and training pulse vectors; a training association unit, which is used to construct a training blood oxygen association matrix between the training blood sugar vector and the training blood pressure vector and to construct a training electrocardiogram association matrix between the training heart rate vector and the training pulse vector; a training multi-scale convolutional coding unit, which is used to respectively convert the training blood oxygen association matrix and the training electrocardiogram association matrix into a multi-scale convolutional coding matrix. A convolutional neural network model with a convolutional structure is used to obtain a training multi-scale blood oxygen feature matrix and a training multi-scale ECG feature matrix; a training feature fusion unit is used to fuse the training multi-scale blood oxygen feature matrix and the training multi-scale ECG feature matrix to obtain a training classification feature matrix; a classification loss unit is used to pass the training classification feature matrix through a classifier to obtain a classification loss function value; a consistency rule internalization learning loss unit is used to expand the training multi-scale blood oxygen feature matrix and the training multi-scale ECG feature matrix into training multi-scale blood oxygen feature vectors and training multi-scale ECG feature vectors respectively, and calculate the sequence-to-sequence consistency rule internalization learning loss function value between the training multi-scale blood oxygen feature vectors and the training multi-scale ECG feature vectors; and a training unit is used to calculate the weighted sum of the classification loss function value and the consistency rule internalization learning loss function value as the loss function value to train the convolutional neural network model with a multi-scale convolutional structure and the classifier.

[0024] In the above-mentioned desktop health monitor, the consistency rule internalization learning loss unit is further used to: calculate the sequence-to-sequence consistency rule internalization learning loss function value between the training multi-scale blood oxygen feature vector and the training multi-scale electrocardiogram feature vector using the following formula; wherein the formula is:

[0025]

[0026]

[0027]

[0028] Among them, V1 is the training multi-scale blood oxygen feature vector, V2 is the training multi-scale ECG feature vector, and W1 and W2 are the weight matrices of the classifier for V1 and V2 respectively.

[0029] According to another aspect of the present application, a desktop health monitoring method is also provided, which includes:

[0030] Acquire multiple health parameters of the human body to be monitored at multiple predetermined time points within a predetermined time period, wherein the multiple health parameters include blood sugar value, blood pressure value, heart rate value and pulse value;

[0031] Arranging the multiple health parameters of the human body to be monitored at multiple predetermined time points within a predetermined time period as input vectors according to the time dimension to obtain a blood sugar vector, a blood pressure vector, a heart rate vector and a pulse vector;

[0032] Constructing a blood oxygen correlation matrix between the blood sugar vector and the blood pressure vector and constructing an electrocardiogram correlation matrix between the heart rate vector and the pulse vector;

[0033] The blood oxygen correlation matrix and the electrocardiogram correlation matrix are respectively passed through a convolutional neural network model with a multi-scale convolution structure to obtain a multi-scale blood oxygen feature matrix and a multi-scale electrocardiogram feature matrix;

[0034] fusing the multi-scale blood oxygen feature matrix and the multi-scale electrocardiogram feature matrix to obtain a classification feature matrix; and

[0035] The classification feature matrix is ​​passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the health condition of the human body to be monitored is normal.

[0036] In the above-mentioned desktop health monitoring method, the construction of the blood oxygen correlation matrix between the blood glucose vector and the blood pressure vector and the construction of the electrocardiogram correlation matrix between the heart rate vector and the pulse vector include: calculating the product between the transposed vector of the blood glucose vector and the blood pressure vector to obtain the blood oxygen correlation matrix; and, calculating the product between the transposed vector of the heart rate vector and the pulse vector to obtain the electrocardiogram correlation matrix.

[0037] In the above-mentioned desktop health monitoring method, the blood oxygen correlation matrix and the electrocardiogram correlation matrix are respectively passed through a convolutional neural network model with a multi-scale convolution structure to obtain a multi-scale blood oxygen feature matrix and a multi-scale electrocardiogram feature matrix, including: using each layer of the convolutional neural network model to perform convolution processing on the input data in the forward transfer of the layer: using the first convolution kernel of the multi-scale convolution structure to perform convolution processing on the input data to obtain a first convolution feature map; using the second convolution kernel of the multi-scale convolution structure to perform convolution processing on the input data to obtain a second convolution feature map; using the third convolution kernel of the multi-scale convolution structure to perform convolution processing on the input data to obtain a third Convolution feature map; use the fourth convolution kernel of the multi-scale convolution structure to convolve the input data to obtain a fourth convolution feature map; cascade the first convolution feature map, the second convolution feature map, the third convolution feature map and the fourth convolution feature map to obtain a multi-scale convolution feature map; pool the multi-scale convolution feature map to obtain a pooled feature map; and perform nonlinear activation processing on the pooled feature map to obtain an activation feature map; wherein the activation feature map output by the last layer of the convolution neural network model is the multi-scale blood oxygen feature matrix; wherein the size of the first convolution kernel is 7×7, the size of the second convolution kernel is 5×5, the size of the third convolution kernel is 3×3, and the size of the fourth convolution kernel is 1×1.

[0038] In the above-mentioned desktop health monitoring method, the blood oxygen correlation matrix and the electrocardiogram correlation matrix are respectively passed through a convolutional neural network model with a multi-scale convolution structure to obtain a multi-scale blood oxygen feature matrix and a multi-scale electrocardiogram feature matrix, and also includes: using each layer of the convolutional neural network model to perform convolution processing on the input data in the forward transfer of the layer: using the first convolution kernel of the multi-scale convolution structure to perform convolution processing on the input data to obtain a first convolution feature map; using the second convolution kernel of the multi-scale convolution structure to perform convolution processing on the input data to obtain a second convolution feature map; using the third convolution kernel of the multi-scale convolution structure to perform convolution processing on the input data to obtain a third convolution feature map; using the multi-scale convolution structure to perform convolution processing on the input data The fourth convolution kernel of the convolution structure performs convolution processing on the input data to obtain a fourth convolution feature map; the first convolution feature map, the second convolution feature map, the third convolution feature map and the fourth convolution feature map are cascaded to obtain a multi-scale convolution feature map; the multi-scale convolution feature map is pooled to obtain a pooled feature map; and the pooled feature map is nonlinearly activated to obtain an activation feature map; wherein the activation feature map output by the last layer of the convolution neural network model is the multi-scale ECG feature matrix; wherein the size of the first convolution kernel is 7×7, the size of the second convolution kernel is 5×5, the size of the third convolution kernel is 3×3, and the size of the fourth convolution kernel is 1×1.

[0039] In the above-mentioned desktop health monitoring method, the fusion of the multi-scale blood oxygen feature matrix and the multi-scale electrocardiogram feature matrix to obtain a classification feature matrix includes: fusing the multi-scale blood oxygen feature matrix and the multi-scale electrocardiogram feature matrix to obtain a classification feature matrix according to the following formula; wherein the formula is:

[0040]

[0041] Among them, M c is the classification feature matrix, M a is the multi-scale blood oxygen feature matrix, M b is the multi-scale ECG feature matrix, represents the addition of elements at corresponding positions of the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix, and α and β are weighted parameters for controlling the balance between the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix in the classification feature matrix.

[0042] In the above-mentioned desktop health monitoring method, the step of passing the classification feature matrix through a classifier to obtain a classification result includes: using the classifier to process the classification feature matrix using the following formula to obtain the classification result; wherein the formula is:

[0043] softmax{(W n ,B n ):…:(W1,B1)|Project(M)}

[0044] Project(M) represents the projection of the classification feature matrix into a vector, W1 to W n is the weight matrix of each fully connected layer, B1 to B n Represents the bias vector of each fully connected layer.

[0045] In the above-mentioned desktop health monitoring method, it also includes training the convolutional neural network model with a multi-scale convolution structure and the classifier.

[0046] In the above-mentioned desktop health monitoring method, the training of the convolutional neural network model with a multi-scale convolution structure and the classifier includes: obtaining training data, the training data including a plurality of training health parameters of the human body to be monitored at a plurality of predetermined time points within a predetermined time period, and a real value of whether the health status of the human body to be monitored is normal, wherein the plurality of training health parameters include training blood glucose value, training blood pressure value, training heart rate value and training pulse value; arranging the plurality of training health parameters of the human body to be monitored at a plurality of predetermined time points within a predetermined time period as input vectors according to the time dimension to obtain training blood glucose vector, training blood pressure vector, training heart rate vector and training pulse vector; constructing a training blood oxygen association matrix between the training blood glucose vector and the training blood pressure vector and constructing a training electrocardiogram association matrix between the training heart rate vector and the training pulse vector; arranging the training blood oxygen association matrix and the training The ECG correlation matrix is ​​respectively passed through a convolutional neural network model with a multi-scale convolution structure to obtain a training multi-scale blood oxygen feature matrix and a training multi-scale ECG feature matrix; the training multi-scale blood oxygen feature matrix and the training multi-scale ECG feature matrix are fused to obtain a training classification feature matrix; the training classification feature matrix is ​​passed through a classifier to obtain a classification loss function value; the training multi-scale blood oxygen feature matrix and the training multi-scale ECG feature matrix are respectively expanded into a training multi-scale blood oxygen feature vector and a training multi-scale ECG feature vector, and a sequence-to-sequence consistency rule internalized learning loss function value between the training multi-scale blood oxygen feature vector and the training multi-scale ECG feature vector is calculated; and a weighted sum of the classification loss function value and the consistency rule internalized learning loss function value is calculated as a loss function value to train the convolutional neural network model with a multi-scale convolution structure and the classifier.

[0047] In the above-mentioned desktop health monitoring method, the training multi-scale blood oxygen feature matrix and the training multi-scale ECG feature matrix are respectively expanded into training multi-scale blood oxygen feature vectors and training multi-scale ECG feature vectors, and the sequence-to-sequence consistency rule internalization learning loss function value between the training multi-scale blood oxygen feature vectors and the training multi-scale ECG feature vectors is calculated, including: calculating the sequence-to-sequence consistency rule internalization learning loss function value between the training multi-scale blood oxygen feature vectors and the training multi-scale ECG feature vectors by the following formula; wherein the formula is:

[0048]

[0049]

[0050]

[0051] Among them, V1 is the training multi-scale blood oxygen feature vector, V2 is the training multi-scale ECG feature vector, and W1 and W2 are the weight matrices of the classifier for V1 and V2 respectively.

[0052] Compared with the prior art, the desktop health monitor provided by the present application can monitor and evaluate the human health status of the individual to be monitored, accompanied by a small number of medical staff or operated by the individual to be monitored. Specifically, a convolutional neural network model is used to extract high-dimensional local implicit correlation implicit features of blood sugar and blood pressure in the blood oxygen association matrix and heart rate and pulse in the electrocardiogram association matrix in different time and space domains to obtain a multi-scale blood oxygen feature matrix and a multi-scale electrocardiogram feature matrix, and then the classification feature matrix obtained by fusion is passed through a classifier to obtain a classification result. In this way, the health status assessment is performed based on multiple health monitoring parameters of the monitored individual to improve the confidence and accuracy of the health status assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0054] Figure 1 is a block diagram of a desktop health monitor according to an embodiment of the present application.

[0055] Figure 2 is an architectural diagram of a desktop health monitor according to an embodiment of the present application.

[0056] Figure 3 4 is a block diagram of a multi-scale convolutional coding unit in a desktop health monitor according to an embodiment of the present application.

[0057] Figure 4 4 is a block diagram of a training module in a desktop health monitor according to an embodiment of the present application.

[0058] Figure 5 Flow chart of a desktop health monitoring method according to an embodiment of the present application. DETAILED DESCRIPTION

[0059] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0060] Application Overview

[0061] Accordingly, in the technical solution of the present application, a desktop health monitor is expected, which can monitor and evaluate the human health status of the individual to be monitored under the accompaniment of a small number of medical staff or under the self-operation of the individual to be monitored.

[0062] Those skilled in the art should know that there are many health monitors that can collect health parameters of the monitored individual and perform health status assessment based on the collected health parameters. However, when performing health status assessment, existing health monitors often evaluate health status based on predetermined judgment rules, for example, based on random forests to process the collected health parameters to obtain health status assessment results. However, there are complex nonlinear correlations between various health parameters of the human body, and such nonlinear correlations cannot be logically deduced through predetermined judgment rules, resulting in low accuracy in health status assessment.

[0063] In recent years, deep learning and neural networks have been widely used in computer vision, natural language processing, text signal processing and other fields. In addition, deep learning and neural networks have also demonstrated a level close to or even beyond that of humans in image classification, object detection, semantic segmentation, text translation and other fields.

[0064] The development of deep learning and neural networks provides new solutions and solutions for the health status assessment of desktop health monitors.

[0065] Specifically, first, a plurality of health parameters of the monitored human body at a plurality of predetermined time points within a predetermined time period are obtained, wherein the plurality of health parameters include blood sugar value, blood pressure value, heart rate value and pulse value. That is, in the technical solution of the present application, the desktop health monitor can perform health status assessment based on the plurality of health monitoring parameters of the monitored individual to improve the confidence and accuracy of the health status assessment.

[0066] Next, the multiple health parameters of the human body to be monitored at multiple predetermined time points within a predetermined time period are arranged as input vectors according to the time dimension to obtain a blood glucose vector, a blood pressure vector, a heart rate vector and a pulse vector. Furthermore, a blood oxygen correlation matrix between the blood glucose vector and the blood pressure vector is constructed, and an electrocardiogram correlation matrix between the heart rate vector and the pulse vector is constructed. In a specific example of the present application, the blood oxygen correlation matrix between the blood glucose vector and the blood pressure vector and the electrocardiogram correlation matrix between the heart rate vector and the pulse vector are constructed by the following formula: the product between the transposed vector of the blood glucose vector and the blood pressure vector is calculated to obtain the blood oxygen correlation matrix; and the product between the transposed vector of the heart rate vector and the pulse vector is calculated to obtain the electrocardiogram correlation matrix. That is, in this specific example, a full temporal spatial correlation between blood pressure and blood glucose and a full temporal spatial correlation between heart rate and pulse are established.

[0067] Next, the blood oxygen correlation matrix and the electrocardiogram correlation matrix are respectively passed through a convolutional neural network model with a multi-scale convolution structure to obtain a multi-scale blood oxygen feature matrix and a multi-scale electrocardiogram feature matrix. That is, a convolutional neural network model with excellent performance in the field of local feature extraction is used as a feature extractor to extract the high-dimensional local implicit correlation implicit features of blood sugar and blood pressure in different spatiotemporal domains in the blood oxygen correlation matrix and the high-dimensional local implicit correlation implicit features of heart rate and pulse in different spatiotemporal domains in the electrocardiogram correlation matrix.

[0068] In particular, in the technical solution of the present application, considering that blood sugar and blood pressure will present different correlation patterns in different time and space domains, and heart rate and pulse will present different correlation patterns in different time and space domains, the various layer structures of the convolutional neural network model are improved. Specifically, a multi-scale convolution structure is used to replace the traditional single-scale convolution structure to perform multi-scale convolution encoding on the blood oxygen correlation matrix and the electrocardiogram correlation matrix to obtain the multi-scale blood oxygen feature matrix and the multi-scale electrocardiogram feature matrix.

[0069] Then, the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix are fused to obtain a classification feature matrix. For example, in a specific example of the present application, the position-weighted sum between the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix is ​​calculated to obtain a classification feature matrix containing blood oxygen features and ECG features. Then, the classification feature matrix is ​​passed through a classifier to obtain a classification result, which is used to indicate whether the health status of the monitored human body is normal.

[0070] In particular, in the technical solution of the present application, although the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix respectively represent feature correlation values ​​on different time-space distribution dimensions, it is still expected that the intrinsic correlation feature expression of the distribution of the original parameters along the time dimension is consistent, so as to improve the fusion effect of the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix.

[0071] Therefore, in the technical solution of the present application, the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix are respectively expanded into multi-scale blood oxygen feature vectors and multi-scale ECG feature vectors, so as to represent their intrinsic feature distribution across time-space dimensions in the form of feature value sequences, and the sequence-to-sequence consistency rule internalization learning loss function between the multi-scale blood oxygen feature vectors and the multi-scale ECG feature vectors is calculated, which is specifically expressed as:

[0072]

[0073]

[0074]

[0075] Among them, V1 is the multi-scale blood oxygen feature vector, V2 is the multi-scale ECG feature matrix, and W1 and W2 are the weight matrices of the classifier for V1 and V2 respectively.

[0076] Here, the sequence-to-sequence consistency rule internalized learning loss function can obtain enhanced distinguishing ability between sequences through the classifier's squeeze-excitation channel attention mechanism for weight matrices of different sequences. In this way, by training the network with this loss function, it is possible to achieve the recovery of more distinguishable correlation features between the multi-scale blood oxygen feature vector and the multi-scale ECG feature vector, so as to internalize the correlation consistency between the feature distributions of the multi-scale blood oxygen feature vector and the multi-scale ECG feature vector. As a result, the consistency of the intrinsic correlation feature expression of the multi-scale blood oxygen feature vector and the multi-scale ECG feature vector for the distribution of the original parameters along the time series dimension is enhanced, thereby improving the fusion effect of the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix.

[0077] Based on this, the present application proposes a desktop health monitor, which includes: a monitoring unit, used to obtain multiple health parameters of a human body to be monitored at multiple predetermined time points within a predetermined time period, and the multiple health parameters include blood sugar value, blood pressure value, heart rate value and pulse value; a data construction unit, used to arrange the multiple health parameters of the human body to be monitored at multiple predetermined time points within a predetermined time period as input vectors according to the time dimension to obtain a blood sugar vector, a blood pressure vector, a heart rate vector and a pulse vector; an association unit, used to construct a blood oxygen association matrix between the blood sugar vector and the blood pressure vector and to construct an electrocardiogram association matrix between the heart rate vector and the pulse vector; a multi-scale convolution encoding unit, used to pass the blood oxygen association matrix and the electrocardiogram association matrix through a convolutional neural network model with a multi-scale convolution structure to obtain a multi-scale blood oxygen feature matrix and a multi-scale electrocardiogram feature matrix; a feature fusion unit, used to fuse the multi-scale blood oxygen feature matrix and the multi-scale electrocardiogram feature matrix to obtain a classification feature matrix; and a monitoring result generation unit, used to pass the classification feature matrix through a classifier to obtain a classification result, and the classification result is used to indicate whether the health status of the human body to be monitored is normal.

[0078] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be described in detail with reference to the accompanying drawings.

[0079] Exemplary Systems

[0080] Figure 1 FIG. 1 is a block diagram of a desktop health monitor according to an embodiment of the present application. Figure 1As shown, according to the desktop health monitor 100 of the embodiment of the present application, it includes: a monitoring unit 110, which is used to obtain a plurality of health parameters of a human body to be monitored at a plurality of predetermined time points within a predetermined time period, wherein the plurality of health parameters include a blood glucose value, a blood pressure value, a heart rate value, and a pulse value; a data construction unit 120, which is used to arrange the plurality of health parameters of the human body to be monitored at a plurality of predetermined time points within a predetermined time period as input vectors according to the time dimension to obtain a blood glucose vector, a blood pressure vector, a heart rate vector, and a pulse vector; an association unit 130, which is used to construct a blood oxygen association matrix between the blood glucose vector and the blood pressure vector and to construct the blood oxygen association matrix between the blood glucose vector and the blood pressure vector. An ECG correlation matrix between a heart rate vector and a pulse vector; a multi-scale convolutional coding unit 140, used to pass the blood oxygen correlation matrix and the ECG correlation matrix through a convolutional neural network model with a multi-scale convolution structure to obtain a multi-scale blood oxygen feature matrix and a multi-scale ECG feature matrix respectively; a feature fusion unit 150, used to fuse the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix to obtain a classification feature matrix; and a monitoring result generating unit 160, used to pass the classification feature matrix through a classifier to obtain a classification result, and the classification result is used to indicate whether the health status of the monitored human body is normal.

[0081] Figure 2 FIG. 1 is a diagram of the architecture of a desktop health monitor according to an embodiment of the present application. Figure 2 As shown, in the architecture of the desktop health monitor, first, a plurality of health parameters of the human body to be monitored at a plurality of predetermined time points within a predetermined time period are obtained, and the plurality of health parameters include blood sugar value, blood pressure value, heart rate value and pulse value. Then, the plurality of health parameters of the human body to be monitored at a plurality of predetermined time points within a predetermined time period are arranged as input vectors according to the time dimension to obtain blood sugar vector, blood pressure vector, heart rate vector and pulse vector. Then, a blood oxygen association matrix between the blood sugar vector and the blood pressure vector is constructed, and an electrocardiogram association matrix between the heart rate vector and the pulse vector is constructed. Furthermore, the blood oxygen association matrix and the electrocardiogram association matrix are respectively passed through a convolutional neural network model with a multi-scale convolution structure to obtain a multi-scale blood oxygen feature matrix and a multi-scale electrocardiogram feature matrix. Then, the multi-scale blood oxygen feature matrix and the multi-scale electrocardiogram feature matrix are fused to obtain a classification feature matrix. Then, the classification feature matrix is ​​passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the health status of the human body to be monitored is normal.

[0082] In the above-mentioned desktop health monitor 100, the monitoring unit 110 is used to obtain multiple health parameters of the monitored human body at multiple predetermined time points within a predetermined time period, and the multiple health parameters include blood sugar value, blood pressure value, heart rate value and pulse value. In the technical solution of the present application, a desktop health monitor is expected, which can monitor and evaluate the human health status of the monitored individual under the accompaniment of a small number of medical staff or under the self-operation of the monitored individual.

[0083] Those skilled in the art should know that there are many health monitors that can collect health parameters of the monitored individual and perform health status assessment based on the collected health parameters. However, when performing health status assessment, existing health monitors often evaluate health status based on predetermined judgment rules, for example, based on random forests to process the collected health parameters to obtain health status assessment results. However, there are complex nonlinear correlations between various health parameters of the human body, and such nonlinear correlations cannot be logically deduced through predetermined judgment rules, resulting in low accuracy in health status assessment.

[0084] In recent years, deep learning and neural networks have been widely used in computer vision, natural language processing, text signal processing and other fields. In addition, deep learning and neural networks have also shown a level close to or even beyond that of humans in image classification, object detection, semantic segmentation, text translation and other fields. The development of deep learning and neural networks provides new solutions and solutions for the health status assessment of desktop health monitors.

[0085] Specifically, first, a plurality of health parameters of the monitored human body at a plurality of predetermined time points within a predetermined time period are obtained, wherein the plurality of health parameters include blood sugar value, blood pressure value, heart rate value and pulse value. That is, in the technical solution of the present application, the desktop health monitor can perform health status assessment based on the plurality of health monitoring parameters of the monitored individual to improve the confidence and accuracy of the health status assessment.

[0086] In the above-mentioned desktop health monitor 100, the data construction unit 120 is used to arrange the multiple health parameters of the monitored human body at multiple predetermined time points within a predetermined time period as input vectors according to the time dimension to obtain a blood sugar vector, a blood pressure vector, a heart rate vector and a pulse vector. That is, the multiple health parameters are arranged as input vectors according to the time dimension to facilitate subsequent data processing in the convolutional neural network model.

[0087] In the above-mentioned desktop health monitor 100, the association unit 130 is used to construct a blood oxygen association matrix between the blood sugar vector and the blood pressure vector and to construct an electrocardiogram association matrix between the heart rate vector and the pulse vector.

[0088] In a specific example of the present application, the blood oxygen correlation matrix between the blood glucose vector and the blood pressure vector and the electrocardiogram correlation matrix between the heart rate vector and the pulse vector are constructed by the following formula:

[0089]

[0090]

[0091] in, represents the transposed vector of the blood glucose vector, V b represents the blood pressure vector, M1 represents the blood oxygen correlation matrix, represents the transposed vector of the heart rate vector, V d represents the pulse vector, M2 represents the ECG correlation matrix, Represents matrix multiplication.

[0092] That is, the product between the transposed vector of the blood glucose vector and the blood pressure vector is calculated to obtain the blood oxygen correlation matrix; and the product between the transposed vector of the heart rate vector and the pulse vector is calculated to obtain the electrocardiogram correlation matrix. Here, in this specific example, the full temporal spatial correlation between blood pressure and blood glucose and the full temporal spatial correlation between heart rate and pulse are established.

[0093] More specifically, in an embodiment of the present application, the association unit 130 includes: a first association subunit, used to calculate the product between the transposed vector of the blood glucose vector and the blood pressure vector to obtain the blood oxygen association matrix; and a second association subunit, used to calculate the product between the transposed vector of the heart rate vector and the pulse vector to obtain the electrocardiogram association matrix.

[0094] In the above-mentioned desktop health monitor 100, the multi-scale convolutional coding unit 140 is used to pass the blood oxygen correlation matrix and the electrocardiogram correlation matrix through a convolutional neural network model with a multi-scale convolution structure to obtain a multi-scale blood oxygen feature matrix and a multi-scale electrocardiogram feature matrix. That is, a convolutional neural network model with excellent performance in the field of local feature extraction is used as a feature extractor to extract the high-dimensional local implicit correlation implicit features of blood sugar and blood pressure in different spatiotemporal domains in the blood oxygen correlation matrix and the high-dimensional local implicit correlation implicit features of heart rate and pulse in different spatiotemporal domains in the electrocardiogram correlation matrix.

[0095] In particular, in the technical solution of the present application, considering that blood sugar and blood pressure will present different correlation patterns in different time and space domains, and heart rate and pulse will present different correlation patterns in different time and space domains, the various layer structures of the convolutional neural network model are improved. Specifically, a multi-scale convolution structure is used to replace the traditional single-scale convolution structure to perform multi-scale convolution encoding on the blood oxygen correlation matrix and the electrocardiogram correlation matrix to obtain the multi-scale blood oxygen feature matrix and the multi-scale electrocardiogram feature matrix.

[0096] Specifically, in an embodiment of the present application, the multi-scale convolution encoding unit 140 includes: a blood oxygen correlation feature extraction subunit 141, which is used to use each layer of the convolutional neural network model to perform convolution processing on the input data in the forward pass of the layer: using the first convolution kernel of the multi-scale convolution structure to convolve the input data to obtain a first convolution feature map; using the second convolution kernel of the multi-scale convolution structure to convolve the input data to obtain a second convolution feature map; using the third convolution kernel of the multi-scale convolution structure to convolve the input data to obtain a third convolution feature map; using the fourth convolution kernel of the multi-scale convolution structure to convolve the input data Convolution processing is performed to obtain a fourth convolution feature map; the first convolution feature map, the second convolution feature map, the third convolution feature map and the fourth convolution feature map are cascaded to obtain a multi-scale convolution feature map; the multi-scale convolution feature map is pooled to obtain a pooled feature map; and the pooled feature map is nonlinearly activated to obtain an activation feature map; wherein the activation feature map output by the last layer of the convolution neural network model is the multi-scale blood oxygen feature matrix; wherein the size of the first convolution kernel is 7×7, the size of the second convolution kernel is 5×5, the size of the third convolution kernel is 3×3, and the size of the fourth convolution kernel is 1×1.

[0097] Specifically, in the embodiment of the present application, the multi-scale convolution encoding unit 140 also includes: an ECG-related feature extraction subunit 142, which is used to use each layer of the convolutional neural network model to perform the following operations on the input data in the forward pass of the layer: convolution processing on the input data using the first convolution kernel of the multi-scale convolution structure to obtain a first convolution feature map; convolution processing on the input data using the second convolution kernel of the multi-scale convolution structure to obtain a second convolution feature map; convolution processing on the input data using the third convolution kernel of the multi-scale convolution structure to obtain a third convolution feature map; convolution processing on the input data using the third convolution kernel of the multi-scale convolution structure to obtain a third convolution feature map; convolution processing on the input data using the third convolution kernel of the multi-scale convolution structure to obtain a third convolution feature map; convolution processing on the input data using the third convolution kernel of the multi-scale convolution structure to obtain a third convolution feature map. The fourth convolution kernel of the convolution neural network model performs convolution processing on the input data to obtain a fourth convolution feature map; the first convolution feature map, the second convolution feature map, the third convolution feature map and the fourth convolution feature map are cascaded to obtain a multi-scale convolution feature map; the multi-scale convolution feature map is pooled to obtain a pooled feature map; and the pooled feature map is nonlinearly activated to obtain an activation feature map; wherein the activation feature map output by the last layer of the convolution neural network model is the multi-scale ECG feature matrix; wherein the size of the first convolution kernel is 7×7, the size of the second convolution kernel is 5×5, the size of the third convolution kernel is 3×3, and the size of the fourth convolution kernel is 1×1.

[0098] Figure 3 FIG. 1 is a block diagram of a multi-scale convolutional coding unit in a desktop health monitor according to an embodiment of the present application. Figure 3 As shown, the multi-scale convolutional coding unit 140 includes the blood oxygen related feature extraction subunit 141 and the electrocardiogram related feature extraction subunit 142.

[0099] In the above-mentioned desktop health monitor 100, the feature fusion unit 150 is used to fuse the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix to obtain a classification feature matrix. For example, in a specific example of the present application, the position-weighted sum between the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix is ​​calculated to obtain a classification feature matrix containing blood oxygen features and ECG features.

[0100] Specifically, in the embodiment of the present application, the feature fusion unit 150 is further used to: fuse the multi-scale blood oxygen feature matrix and the multi-scale electrocardiogram feature matrix according to the following formula to obtain a classification feature matrix; wherein the formula is:

[0101]

[0102] Among them, M c is the classification feature matrix, M a is the multi-scale blood oxygen feature matrix, M bis the multi-scale ECG feature matrix, It represents the addition of the elements at the corresponding positions of the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix, and α and β are weighted parameters for controlling the balance between the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix in the classification feature matrix. Here, the weight value of the multi-scale ECG feature matrix is ​​greater than the weight value of the multi-scale blood oxygen feature matrix, that is, in the health monitoring of the monitored individual, ECG data is more important than blood oxygen data.

[0103] In the above-mentioned desktop health monitor 100, the monitoring result generating unit 160 is used to pass the classification feature matrix through a classifier to obtain a classification result, and the classification result is used to indicate whether the health condition of the human body to be monitored is normal.

[0104] Specifically, in the embodiment of the present application, the monitoring result generating unit 160 is further used to: use the classifier to process the classification feature matrix according to the following formula to obtain the classification result; wherein the formula is: softmax{(W n ,B n ):…:(W1,B1)|Project(M)}, where Project(M) represents projecting the classification feature matrix into a vector, W1 to W n is the weight matrix of each fully connected layer, B1 to B n Represents the bias vector of each fully connected layer.

[0105] The above-mentioned desktop health monitor 100 also includes a training module 200 for training the convolutional neural network model with a multi-scale convolution structure and the classifier.

[0106] Figure 4 FIG. 1 is a block diagram of a training module in a desktop health monitor according to an embodiment of the present application. Figure 4As shown, the training module 200 includes: a training monitoring unit 210, which is used to obtain training data, wherein the training data includes a plurality of training health parameters of the human body to be monitored at a plurality of predetermined time points within a predetermined time period, and a real value of whether the health status of the human body to be monitored is normal, wherein the plurality of training health parameters include training blood glucose value, training blood pressure value, training heart rate value and training pulse value; a training data construction unit 220, which is used to arrange the plurality of training health parameters of the human body to be monitored at a plurality of predetermined time points within a predetermined time period as input vectors according to the time dimension to obtain training blood glucose vectors, training blood pressure vectors, training heart rate vectors and training pulse vectors; a training association unit 230, which is used to construct a training blood oxygen association matrix between the training blood glucose vector and the training blood pressure vector and to construct a training electrocardiogram association matrix between the training heart rate vector and the training pulse vector; a training multi-scale convolution encoding unit 240, which is used to respectively convert the training blood oxygen association matrix and the training electrocardiogram association matrix into a training vector having a multi-scale convolution structure; A convolutional neural network model with a multi-scale convolution structure is constructed to obtain a training multi-scale blood oxygen feature matrix and a training multi-scale ECG feature matrix; a training feature fusion unit 250 is used to fuse the training multi-scale blood oxygen feature matrix and the training multi-scale ECG feature matrix to obtain a training classification feature matrix; a classification loss unit 260 is used to pass the training classification feature matrix through a classifier to obtain a classification loss function value; a consistency rule internalization learning loss unit 270 is used to expand the training multi-scale blood oxygen feature matrix and the training multi-scale ECG feature matrix into training multi-scale blood oxygen feature vectors and training multi-scale ECG feature vectors respectively, and calculate the sequence-to-sequence consistency rule internalization learning loss function value between the training multi-scale blood oxygen feature vectors and the training multi-scale ECG feature vectors; and a training unit 280 is used to calculate the weighted sum of the classification loss function value and the consistency rule internalization learning loss function value as the loss function value to train the convolutional neural network model with a multi-scale convolution structure and the classifier.

[0107] In particular, in the technical solution of the present application, although the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix respectively represent feature correlation values ​​on different time-space distribution dimensions, it is still expected that the intrinsic correlation feature expression of the distribution of the original parameters along the time dimension is consistent, so as to improve the fusion effect of the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix. Therefore, in the technical solution of the present application, the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix are respectively expanded into multi-scale blood oxygen feature vectors and multi-scale ECG feature vectors, so as to represent their intrinsic feature distribution across time-space dimensions in the form of feature value sequences, and calculate the sequence-to-sequence consistency rule internalized learning loss function between the multi-scale blood oxygen feature vectors and the multi-scale ECG feature vectors.

[0108] Specifically, in the embodiment of the present application, the consistency rule internalization learning loss unit 270 is further used to calculate the sequence-to-sequence consistency rule internalization learning loss function value between the training multi-scale blood oxygen feature vector and the training multi-scale electrocardiogram feature vector using the following formula; wherein the formula is:

[0109]

[0110]

[0111]

[0112] Among them, V1 is the training multi-scale blood oxygen feature vector, V2 is the training multi-scale ECG feature vector, and W1 and W2 are the weight matrices of the classifier for V1 and V2 respectively, and d represents the calculation of the Euclidean distance between the two vectors.

[0113] Here, the sequence-to-sequence consistency rule internalized learning loss function can obtain enhanced distinguishing ability between sequences through the classifier's squeeze-excitation channel attention mechanism for weight matrices of different sequences. In this way, by training the network with this loss function, it is possible to achieve the recovery of more distinguishable correlation features between the multi-scale blood oxygen feature vector and the multi-scale ECG feature vector, so as to internalize the correlation consistency between the feature distributions of the multi-scale blood oxygen feature vector and the multi-scale ECG feature vector. As a result, the consistency of the intrinsic correlation feature expression of the multi-scale blood oxygen feature vector and the multi-scale ECG feature vector for the distribution of the original parameters along the time series dimension is enhanced, thereby improving the fusion effect of the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix.

[0114] In summary, the desktop health monitor 100 according to the embodiment of the present application is explained, wherein the desktop health monitor can monitor and evaluate the human health status of the monitored individual under the accompaniment of a small number of medical staff or under the self-operation of the monitored individual. Specifically, a convolutional neural network model is used to extract high-dimensional local implicit correlation implicit features of blood sugar and blood pressure in the blood oxygen association matrix and heart rate and pulse in the electrocardiogram association matrix in different time and space domains to obtain a multi-scale blood oxygen feature matrix and a multi-scale electrocardiogram feature matrix, and then the classification feature matrix obtained by fusion is passed through a classifier to obtain a classification result. In this way, a health status assessment is performed based on multiple health monitoring parameters of the monitored individual to improve the confidence and accuracy of the health status assessment.

[0115] Exemplary Methods

[0116] Figure 5 FIG. 1 is a flow chart of a desktop health monitoring method according to an embodiment of the present application. Figure 5 As shown, according to the desktop health monitoring method of the embodiment of the present application, the steps include: S110, obtaining multiple health parameters of the human body to be monitored at multiple predetermined time points within a predetermined time period, and the multiple health parameters include blood sugar value, blood pressure value, heart rate value and pulse value; S120, arranging the multiple health parameters of the human body to be monitored at multiple predetermined time points within the predetermined time period as input vectors according to the time dimension to obtain blood sugar vector, blood pressure vector, heart rate vector and pulse vector; S130, constructing a blood oxygen association matrix between the blood sugar vector and the blood pressure vector and constructing an electrocardiogram association matrix between the heart rate vector and the pulse vector; S140, respectively passing the blood oxygen association matrix and the electrocardiogram association matrix through a convolutional neural network model with a multi-scale convolution structure to obtain a multi-scale blood oxygen feature matrix and a multi-scale electrocardiogram feature matrix; S150, fusing the multi-scale blood oxygen feature matrix and the multi-scale electrocardiogram feature matrix to obtain a classification feature matrix; and, S160, passing the classification feature matrix through a classifier to obtain a classification result, and the classification result is used to indicate whether the health status of the human body to be monitored is normal.

[0117] In one example, in the above-mentioned desktop health monitoring method, constructing a blood oxygen correlation matrix between the blood glucose vector and the blood pressure vector and constructing an electrocardiogram correlation matrix between the heart rate vector and the pulse vector include: calculating the product between the transposed vector of the blood glucose vector and the blood pressure vector to obtain the blood oxygen correlation matrix; and, calculating the product between the transposed vector of the heart rate vector and the pulse vector to obtain the electrocardiogram correlation matrix.

[0118] In one example, in the above-mentioned desktop health monitoring method, the blood oxygen correlation matrix and the ECG correlation matrix are respectively passed through a convolutional neural network model with a multi-scale convolution structure to obtain a multi-scale blood oxygen feature matrix and a multi-scale ECG feature matrix, including: using each layer of the convolutional neural network model to perform convolution processing on the input data in the forward transfer of the layer: using the first convolution kernel of the multi-scale convolution structure to perform convolution processing on the input data to obtain a first convolution feature map; using the second convolution kernel of the multi-scale convolution structure to perform convolution processing on the input data to obtain a second convolution feature map; using the third convolution kernel of the multi-scale convolution structure to perform convolution processing on the input data to obtain a third convolution feature map; using the multi-scale convolution structure to perform convolution processing on the input data. The fourth convolution kernel of the multi-scale convolution structure performs convolution processing on the input data to obtain a fourth convolution feature map; the first convolution feature map, the second convolution feature map, the third convolution feature map and the fourth convolution feature map are cascaded to obtain a multi-scale convolution feature map; the multi-scale convolution feature map is pooled to obtain a pooled feature map; and the pooled feature map is nonlinearly activated to obtain an activation feature map; wherein the activation feature map output by the last layer of the convolution neural network model is the multi-scale blood oxygen feature matrix; wherein the size of the first convolution kernel is 7×7, the size of the second convolution kernel is 5×5, the size of the third convolution kernel is 3×3, and the size of the fourth convolution kernel is 1×1.

[0119] In one example, in the above-mentioned desktop health monitoring method, the fusing the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix to obtain a classification feature matrix includes: fusing the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix to obtain a classification feature matrix using the following formula; wherein the formula is:

[0120]

[0121] Among them, M c is the classification feature matrix, M a is the multi-scale blood oxygen feature matrix, M b is the multi-scale ECG feature matrix, represents the addition of elements at corresponding positions of the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix, and α and β are weighted parameters for controlling the balance between the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix in the classification feature matrix.

[0122] In one example, in the above-mentioned desktop health monitoring method, it also includes training the convolutional neural network model with the multi-scale convolution structure and the classifier.

[0123] In one example, in the above-mentioned desktop health monitoring method, the training of the convolutional neural network model with a multi-scale convolution structure and the classifier includes: obtaining training data, the training data including a plurality of training health parameters of the human body to be monitored at a plurality of predetermined time points within a predetermined time period, and a true value of whether the health status of the human body to be monitored is normal, wherein the plurality of training health parameters include training blood glucose values, training blood pressure values, training heart rate values, and training pulse values; arranging the plurality of training health parameters of the human body to be monitored at a plurality of predetermined time points within a predetermined time period as input vectors according to the time dimension to obtain training blood glucose vectors, training blood pressure vectors, training heart rate vectors, and training pulse vectors; constructing a training blood oxygen association matrix between the training blood glucose vectors and the training blood pressure vectors, and constructing a training electrocardiogram association matrix between the training heart rate vectors and the training pulse vectors; and arranging the training blood oxygen association matrix and the training blood glucose vectors into input vectors according to the time dimension. The training ECG association matrix is ​​respectively passed through a convolutional neural network model with a multi-scale convolution structure to obtain a training multi-scale blood oxygen feature matrix and a training multi-scale ECG feature matrix; the training multi-scale blood oxygen feature matrix and the training multi-scale ECG feature matrix are fused to obtain a training classification feature matrix; the training classification feature matrix is ​​passed through a classifier to obtain a classification loss function value; the training multi-scale blood oxygen feature matrix and the training multi-scale ECG feature matrix are respectively expanded into a training multi-scale blood oxygen feature vector and a training multi-scale ECG feature vector, and a sequence-to-sequence consistency rule internalized learning loss function value between the training multi-scale blood oxygen feature vector and the training multi-scale ECG feature vector is calculated; and a weighted sum of the classification loss function value and the consistency rule internalized learning loss function value is calculated as a loss function value to train the convolutional neural network model with a multi-scale convolution structure and the classifier.

[0124] In one example, in the above-mentioned desktop health monitoring method, the training multi-scale blood oxygen feature matrix and the training multi-scale ECG feature matrix are respectively expanded into training multi-scale blood oxygen feature vectors and training multi-scale ECG feature vectors, and the sequence-to-sequence consistency rule internalization learning loss function value between the training multi-scale blood oxygen feature vectors and the training multi-scale ECG feature vectors is calculated, including: calculating the sequence-to-sequence consistency rule internalization learning loss function value between the training multi-scale blood oxygen feature vectors and the training multi-scale ECG feature vectors by the following formula; wherein the formula is:

[0125]

[0126]

[0127]

[0128] Among them, V1 is the training multi-scale blood oxygen feature vector, V2 is the training multi-scale ECG feature vector, and W1 and W2 are the weight matrices of the classifier for V1 and V2 respectively.

[0129] In summary, the desktop health monitoring method according to the embodiment of the present application is explained, wherein the desktop health monitor can monitor and evaluate the human health status of the monitored individual under the accompaniment of a small number of medical staff or under the self-operation of the monitored individual. Specifically, a convolutional neural network model is used to extract high-dimensional local implicit correlation implicit features of blood sugar and blood pressure in the blood oxygen association matrix and heart rate and pulse in the electrocardiogram association matrix in different time and space domains to obtain a multi-scale blood oxygen feature matrix and a multi-scale electrocardiogram feature matrix, and then the classification feature matrix obtained by fusion is passed through a classifier to obtain a classification result. In this way, a health status assessment is performed based on multiple health monitoring parameters of the monitored individual to improve the confidence and accuracy of the health status assessment.

Claims

1. A desktop health monitor, It is characterized in that include: A monitoring unit, used to obtain a plurality of health parameters of a human body to be monitored at a plurality of predetermined time points within a predetermined time period, wherein the plurality of health parameters include blood sugar value, blood pressure value, heart rate value and pulse value; A data construction unit, used for arranging the multiple health parameters of the human body to be monitored at multiple predetermined time points within a predetermined time period into input vectors according to the time dimension to obtain a blood sugar vector, a blood pressure vector, a heart rate vector and a pulse vector; A correlation unit, used to construct a blood oxygen correlation matrix between the blood sugar vector and the blood pressure vector and to construct an electrocardiogram correlation matrix between the heart rate vector and the pulse vector; A multi-scale convolutional encoding unit, used for respectively passing the blood oxygen correlation matrix and the electrocardiogram correlation matrix through a convolutional neural network model with a multi-scale convolutional structure to obtain a multi-scale blood oxygen feature matrix and a multi-scale electrocardiogram feature matrix; A feature fusion unit, used for fusing the multi-scale blood oxygen feature matrix and the multi-scale electrocardiogram feature matrix to obtain a classification feature matrix; as well as A monitoring result generating unit, used for passing the classification feature matrix through a classifier to obtain a classification result, wherein the classification result is used to indicate whether the health condition of the human body to be monitored is normal; The desktop health monitor also includes a training module for training a convolutional neural network model with a multi-scale convolutional structure and the classifier; The training module comprises: a training monitoring unit, configured to obtain training data, wherein the training data includes a plurality of training health parameters of a human body to be monitored at a plurality of predetermined time points within a predetermined time period, and a real value of whether the health status of the human body to be monitored is normal, wherein the plurality of training health parameters include a training blood sugar value, a training blood pressure value, a training heart rate value, and a training pulse value; A training data construction unit, used for arranging a plurality of training health parameters of the human body to be monitored at a plurality of predetermined time points within a predetermined time period as input vectors according to the time dimension to obtain a training blood sugar vector, a training blood pressure vector, a training heart rate vector and a training pulse vector; A training association unit, used to construct a training blood oxygen association matrix between the training blood sugar vector and the training blood pressure vector, and to construct a training electrocardiogram association matrix between the training heart rate vector and the training pulse vector; A training multi-scale convolutional encoding unit is used to pass the training blood oxygen correlation matrix and the training ECG correlation matrix through a convolutional neural network model with a multi-scale convolution structure to obtain a training multi-scale blood oxygen feature matrix and a training multi-scale ECG feature matrix; A training feature fusion unit, used for fusing the training multi-scale blood oxygen feature matrix and the training multi-scale electrocardiogram feature matrix to obtain a training classification feature matrix; A classification loss unit, used for passing the training classification feature matrix through a classifier to obtain a classification loss function value; A consistency rule internalization learning loss unit, used to expand the training multi-scale blood oxygen feature matrix and the training multi-scale ECG feature matrix into training multi-scale blood oxygen feature vectors and training multi-scale ECG feature vectors respectively, and calculate a sequence-to-sequence consistency rule internalization learning loss function value between the training multi-scale blood oxygen feature vectors and the training multi-scale ECG feature vectors; and A training unit, used for calculating a weighted sum of the classification loss function value and the consistency rule internalization learning loss function value as a loss function value to train the convolutional neural network model with a multi-scale convolution structure and the classifier; The consistency rule internalizes the learning loss unit, and is further used to: The sequence-to-sequence consistency rule internalization learning loss function value between the training multi-scale blood oxygen feature vector and the training multi-scale electrocardiogram feature vector is calculated by the following formula; Wherein, the formula is: in, is the training multi-scale blood oxygen feature vector, is the training multi-scale ECG feature vector, and and The classifiers are and The weight matrix of .

2. The desktop health monitor according to claim 1, It is characterized in that The association unit comprises: A first association subunit, configured to calculate the product between the transposed vector of the blood glucose vector and the blood pressure vector to obtain the blood oxygen association matrix; and The second correlation subunit is used to calculate the product between the transposed vector of the heart rate vector and the pulse vector to obtain the electrocardiographic correlation matrix.

3. The desktop health monitor according to claim 2, It is characterized in that The multi-scale convolutional coding unit comprises: The blood oxygen correlation feature extraction subunit is used to use each layer of the convolutional neural network model to perform the following operations on the input data in the forward pass of the layer: Using a first convolution kernel of the multi-scale convolution structure to perform convolution processing on the input data to obtain a first convolution feature map; Using a second convolution kernel of the multi-scale convolution structure to perform convolution processing on the input data to obtain a second convolution feature map; Using a third convolution kernel of the multi-scale convolution structure to perform convolution processing on the input data to obtain a third convolution feature map; Using a fourth convolution kernel of the multi-scale convolution structure to perform convolution processing on the input data to obtain a fourth convolution feature map; Cascading the first convolution feature map, the second convolution feature map, the third convolution feature map, and the fourth convolution feature map to obtain a multi-scale convolution feature map; Performing pooling processing on the multi-scale convolutional feature map to obtain a pooled feature map; and Performing nonlinear activation processing on the pooled feature map to obtain an activated feature map; Wherein, the activation feature map output by the last layer of the convolutional neural network model is the multi-scale blood oxygen feature matrix; Among them, the size of the first convolution kernel is 7×7, the size of the second convolution kernel is 5×5, the size of the third convolution kernel is 3×3, and the size of the fourth convolution kernel is 1×1.

4. The desktop health monitor according to claim 3, It is characterized in that The multi-scale convolutional coding unit further includes: The ECG correlation feature extraction subunit is used to use each layer of the convolutional neural network model to perform the following operations on the input data in the forward pass of the layer: Using a first convolution kernel of the multi-scale convolution structure to perform convolution processing on the input data to obtain a first convolution feature map; Using a second convolution kernel of the multi-scale convolution structure to perform convolution processing on the input data to obtain a second convolution feature map; Using a third convolution kernel of the multi-scale convolution structure to perform convolution processing on the input data to obtain a third convolution feature map; Using a fourth convolution kernel of the multi-scale convolution structure to perform convolution processing on the input data to obtain a fourth convolution feature map; Cascading the first convolution feature map, the second convolution feature map, the third convolution feature map, and the fourth convolution feature map to obtain a multi-scale convolution feature map; Performing pooling processing on the multi-scale convolutional feature map to obtain a pooled feature map; and Performing nonlinear activation processing on the pooled feature map to obtain an activated feature map; Wherein, the activation feature map output by the last layer of the convolutional neural network model is the multi-scale ECG feature matrix; Among them, the size of the first convolution kernel is 7×7, the size of the second convolution kernel is 5×5, the size of the third convolution kernel is 3×3, and the size of the fourth convolution kernel is 1×1.

5. The desktop health monitor according to claim 4, It is characterized in that The feature fusion unit is further used for: The multi-scale blood oxygen feature matrix and the multi-scale electrocardiogram feature matrix are fused using the following formula to obtain a classification feature matrix; Wherein, the formula is: in, is the classification feature matrix, is the multi-scale blood oxygen feature matrix, is the multi-scale ECG feature matrix, " " represents the addition of the elements at corresponding positions of the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix, and is a weighting parameter used to control the balance between the multi-scale blood oxygen feature matrix and the multi-scale ECG feature matrix in the classification feature matrix.

6. The desktop health monitor according to claim 5, It is characterized in that The monitoring result generating unit is further used for: Using the classifier to process the classification feature matrix according to the following formula to obtain the classification result; Wherein, the formula is: ,in Represents projecting the classification feature matrix into a vector, to is the weight matrix of each fully connected layer, to Represents the bias vector of each fully connected layer.

Images