Health assessment system and method based on wearable device multi-source data fusion analysis

Abstract

The application discloses a health assessment system and method based on multi-source data fusion analysis of a wearable device. The health assessment features of the wearable device are obtained based on a multi-source data fusion analysis method by using the obtained hearing feature vector, heart rate and blood pressure features, and the wearable device health assessment model is constructed by using the wearable device health assessment features and the adopted wearable health assessment results. The application adjusts the convolutional neural network model based on the hearing features collected and processed by the wearable device in combination with physiological parameters, thereby efficiently processing highly structured feature data. By imaging the multi-source data, the wearable device health assessment features more suitable for the multi-source data fusion analysis of the wearable device are obtained, the trained model has less parameters and stronger generalization ability, and under the multi-source processing of the generated health assessment features on the model, the health assessment of the wearable device on the wearer is more objective.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of artificial intelligence, and particularly relates to a health assessment system and method based on wearable device multi-source data fusion analysis. BACKGROUND

[0002] Hearing is a kind of ability to start the auditory organ to receive voice information, and hearing is very important to a person, and people communicate quickly through this ability, therefore, the detection method of hearing is gradually improved, and the hearing ability is an important index for health assessment, and many literatures assess the health status of personnel according to hearing data; for example, the patent with publication number CN107358034A comprehensively monitors the health of children through hearing indicators.

[0003] The prior art shows that, on the one hand, the detection of hearing at present is through artificial operation of equipment, and hearing is detected according to the equipment, but when the hearing of the equipment to be tested needs to be analyzed, personnel are used to analyze, which consumes time, reduces work efficiency, and is prone to errors; on the other hand, the health is assessed by using a model, the hearing ability data is not detected and expressed, and the wearable device is used to comprehensively process the basic physiological parameters to adapt to the characteristic processing to correct the model in a detailed improvement direction, so as to obtain more objective wearable device convenient health assessment results.

[0004] Therefore, how to use the existing hearing detection data to adapt to the processing mode of the input traditional CNN neural network model, how to correct the CNN neural network model by image processing the feature vector value of the traditional input model, so as to obtain highly structured feature data, and how to obtain multi-source image data, obtain wearable device health assessment features more suitable for wearable device multi-source data fusion analysis, so that the model parameters trained are less, the generalization ability is stronger, and the health assessment of the wearable device to the wearer is more objective under the multi-source processing of the health assessment features generated by fusion, therefore, the present application proposes a health assessment system and method based on wearable device multi-source data fusion analysis. SUMMARY

[0005] To solve the above technical problems, the present application provides a health assessment system and method based on wearable device multi-source data fusion analysis.

[0006] In a first aspect of the present application, a health assessment method based on wearable device multi-source data fusion analysis is provided, and the method comprises:

[0007] S1, collect and process the hearing feature vector of the wearer, obtain the heart rate and blood pressure features of the wearer based on the wearable device, obtain the wearable device health evaluation features based on the multi-source data fusion analysis method, and obtain the health evaluation result of the wearer;

[0008] S2, the multi-source data image of the hearing feature vector and the heart rate and blood pressure features is obtained by the multi-source data image, and the wearable device image health evaluation feature with image format characteristics is obtained;

[0009] S3, based on the wearable device health evaluation features and the health evaluation result, input into the neural network model improved based on the wearable device image health evaluation features to construct the wearable device health evaluation model;

[0010] S4, according to the wearable device health evaluation model, the personnel health evaluation of the wearable device wearer is carried out.

[0011] Further, the wearable device health evaluation model is a CNN neural network model improved based on the wearable device image health evaluation features.

[0012] Further, the hearing feature vector is calculated by detection difference and analysis difference, age data, detection distance data, noise data, analysis influence factor, detection influence factor, age influence factor, noise influence factor and distance influence factor.

[0013] Further, the heart rate and blood pressure features of the wearer are composed of a two-dimensional feature matrix composed of the maximum heart rate, minimum heart rate, minimum blood pressure and maximum blood pressure of the wearer.

[0014] Further, the multi-source data image refers to filling the four-dimensional feature matrix with the hearing feature vector as the reference of the heart rate and blood pressure features of the wearer.

[0015] Further, the detection difference is obtained by the difference between the wearing detection start time and the wearing detection end time of the wearer, and the analysis difference is obtained by the difference between the wearing evaluation analysis start time data and the wearing evaluation analysis end time data.

[0016] Further, the wearable device health evaluation model uses the filling vector value based on the hearing feature vector to correct the activation function of the CNN neural network model.

[0017] The application further provides a health assessment system based on wearable device multi-source data fusion analysis, comprising an auditory feature processing generation module, a wearable device heart rate and blood pressure feature acquisition module, a multi-source data imaging generation module, a wearable device health assessment model construction module, and a personnel health assessment module.

[0018] The auditory feature processing generation module is used for collecting and processing auditory feature vectors of a wearer.

[0019] The wearable device heart rate and blood pressure feature acquisition module is used for acquiring heart rate and blood pressure features of the wearer according to a wearable device.

[0020] The multi-source data imaging generation module is used for obtaining wearable device health assessment features based on a multi-source data fusion analysis method.

[0021] The wearable device health assessment model construction module is used for constructing a wearable device health assessment model based on the wearable device health assessment features and the health assessment results input into a neural network model improved based on wearable device image health assessment features.

[0022] The personnel health assessment module is used for performing personnel health assessment on a wearable device wearer who wears a wearable device again or for the first time according to the wearable device health assessment model.

[0023] Further, the wearable device health assessment model is a CNN neural network model improved based on wearable device image health assessment features.

[0024] Further, the auditory feature vector is calculated from a detection difference and an analysis difference, age data, detection distance data, noise data, analysis influence factors, detection influence factors, age influence factors, noise influence factors and distance influence factors.

[0025] On one hand, when analyzing the hearing of a wearable device wearer, the application uses simply collected hearing related data, thereby improving work efficiency and reducing errors.

[0026] Therefore, the beneficial effects of the present application are that the existing hearing detection data is used to fuse basic physiological parameters in a multi-modal manner, and then the improved CNN neural network model based on the input feature image format characteristics is used for processing, that is, the feature vector values of the traditional input model are image processed to correct the CNN neural network model, so as to obtain highly structured feature data, and the multi-source image data is obtained by using the hearing vector filling value based on the feature convolution mode, and the wearable device health evaluation characteristics more suitable for wearable device multi-source data fusion analysis are obtained, so that the trained model has less parameter quantity and stronger generalization ability, and under the multi-source processing of the health evaluation characteristics generated by fusion on the model, the wearable device is more objective in health evaluation of the wearer.

[0027] More embodiments and improved effects of the present application will be further introduced in combination with the drawings and specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 is a health evaluation method flowchart based on wearable device multi-source data fusion analysis of the present application;

[0029] Figure 2 is a health evaluation system schematic diagram based on wearable device multi-source data fusion analysis of the present application;

[0030] Figure 3 is a regular padding convolution example diagram in the embodiment of the present application;

[0031] Figure 4 is a schematic diagram of an activation function of a CNN neural network model in the embodiment of the present application;

[0032] Figure 5 is a schematic diagram of an electronic device structure for implementing the method of the present application in the embodiment of the present application. DETAILED DESCRIPTION

[0033] In the following, the application will be further described in combination with the drawings and specific embodiments, and the model used in the present application is an improved version of the CNN neural network model.

[0034] As shown in Figure 2 , the system of the present application belongs to the field of artificial intelligence such as application software development of biometric recognition software, and therefore belongs to artificial intelligence, and belongs to the next generation information network industry.

[0035] In a first aspect of the present application, a health evaluation method based on wearable device multi-source data fusion analysis is provided, and the method comprises:

[0036] S1, collect and process the hearing feature vector of the wearer, obtain the heart rate and blood pressure features of the wearer according to the wearable device, obtain the wearable device health evaluation feature based on a multi-source data fusion analysis method, and obtain the health evaluation result of the wearer;

[0037] S2, the multi-source data fusion analysis method is to image the hearing feature vector, the heart rate and the blood pressure feature, and obtain the wearable device image health evaluation feature with image format characteristics;

[0038] S3, based on the wearable device health evaluation feature and the health evaluation result, input into the neural network model improved based on the wearable device image health evaluation feature to construct a wearable device health evaluation model;

[0039] S4, according to the wearable device health evaluation model, the health of the person wearing or newly wearing the wearable device is evaluated.

[0040] Further, the wearable device health evaluation model is a CNN neural network model improved based on the wearable device image health evaluation feature, the CNN neural network is a commonly used model for processing image features, the convolution filling improvement of the CNN neural network model is carried out by using the image format feature, and the calculation method of the activation function is further corrected to improve the generalization ability of the model for classification processing of multi-source fusion monitoring data, so as to improve the accurate and objective health evaluation accuracy of the model.

[0041] In another embodiment of the application, the hearing feature vector is calculated by detecting difference and analyzing difference, age data, detection distance data, noise data, analysis influence factor, detection influence factor, age influence factor, noise influence factor and distance influence factor:

[0042]

[0043] Among them, is the hearing feature vector, J is the detection distance data, Z is the noise data, G is the age data, D is the detection difference, Y is the analysis difference, J, Z, G, D and Y are one-dimensional feature vector values respectively represent the distance influence factor, the noise influence factor, the age influence factor, the detection influence factor and the analysis influence factor, the above influence factors are used to remove the dimensional influence, E represents the calculation bias adjustment factor of the hearing feature vector, the value of E is generally taken between 1.05-1.1, which is to ensure that the bias of the hearing feature vector is not too large, resulting in a large result bias in the input model calculation, and increasing the stability of the trained model.

[0044] The calculation of each influence factor is obtained by the innovative calculation of the person wearing the wearable device, and each influence factor respectively influences the calculation and acquisition of the hearing feature vector. Through the calculation of the control variables of the same person wearing the wearable device, the characteristic value which has a more obvious influence on the hearing feature vector can be obtained, thereby contributing more accurate input feature vector data to the subsequent training of the CNN neural network model.

[0045] S1: Select the wearable device wearers whose detection distance data, detection difference, analysis difference, noise data, age data and actual hearing data are all the same, wherein the detection distance data is not the same, obtain the corresponding hearing data, and mark two different hearing data as T1 and T2 respectively, mark two different detection distance data as J1 and J2 respectively, calculate the difference between the two detection distance data, and substitute J1, J2 and T1, T2 into the calculation formula: (J1-J2)* = (T1-T2), calculate the value of , and represent the influence factor of the detection distance data on the hearing detection, and mark it as the distance influence factor;

[0046] S2: Select the wearable device wearers whose detection distance data, detection difference, analysis difference, noise data, age data and actual hearing data are all the same, wherein the noise data is not the same, obtain the corresponding hearing data, and mark two different hearing data as T3 and T4 respectively, mark two different noise data as Z1 and Z2 respectively, substitute T3, T4 and Z1, Z2 into the calculation formula: (Z1-Z2)* = (T3-T4), calculate the value of , and represent the influence factor of the noise data on the hearing detection, and mark it as the noise influence factor;

[0047] S3: Select the wearable device wearers whose detection distance data, detection difference, analysis difference, noise data and actual hearing data are all the same, wherein the age data is not the same, obtain the corresponding hearing data, and mark two different hearing data as T6 and T5 respectively, mark two different age data as N1 and N2 respectively, substitute N1 and N2 and T6 and T5 into the calculation formula: (N1-N2)* = (T5-T6), calculate the value of , and represent the influence factor of the age data on the hearing detection, and mark it as the age influence factor;

[0048] S4: Select the wearable device wearer whose detection distance data, analysis difference, noise data, age data and actual hearing data are the same, wherein the detection difference is not the same, obtain the corresponding hearing data, and mark the two different hearing data as T7 and T8 respectively, mark the two different detection differences as C1 and C2 respectively, and substitute C1 and C2 and T7 and T8 into the calculation formula: (C1-C2)* =(T7-T8), calculate the value of , , which is expressed as the influence factor of detection difference on hearing detection, and is marked as detection influence factor;

[0049] S5: Select the wearable device wearer whose detection distance data, detection difference, noise data, age data and actual hearing data are the same, wherein the analysis difference is not the same, obtain the corresponding hearing data, and mark the two different hearing data as T9 and T10 respectively, mark the two different analysis differences as F1 and F2 respectively, and substitute F1 and F2 and T9 and T10 into the calculation formula: (F1-F2)* =(T9-T10), calculate the value of , , which is expressed as the influence factor of analysis difference on hearing detection, and is marked as analysis influence factor;

[0050] Further, the heart rate and blood pressure characteristics of the wearer are composed of a two-dimensional feature matrix composed of the maximum heart rate, minimum heart rate, and minimum blood pressure and maximum blood pressure of the wearer.

[0051] Further, the multi-source data is imaged, specifically, the hearing feature vector is filled into a four-dimensional feature matrix with the heart rate and blood pressure characteristics of the wearer as the benchmark. The filling operation can keep the spatial size of the two-dimensional feature matrix unchanged, which helps to increase the hearing health related information, and also can better preserve the edge information, and the calculation method of the filling value is:

[0052]

[0053] In the formula, represents the filling vector value, represents the hearing feature vector, D represents the detection difference, and Y represents the analysis difference. The specific four-dimensional feature matrix is the image health evaluation feature of the wearable device, that is, the four-dimensional feature matrix is .

[0054] Further, the detection difference is obtained by the difference between the wearing detection start time and the wearing detection end time of the wearer, and the analysis difference is obtained by the difference between the wearing evaluation analysis start time data and the wearing evaluation analysis end time data.

[0055] Further, the wearable device health assessment model utilizes the filling vector value based on the hearing feature vector to correct the activation function of the CNN neural network model, and the specific expression of the activation function is:

[0056]

[0057] In the formula, is the activation function value, is the filling vector value, respectively represent a distance influence factor, a noise influence factor, an age influence factor, a detection influence factor and an analysis influence factor, and M is the vector value after weighted summation of the input four-dimensional feature matrix in the neural network.

[0058] The present application adopts the supplementary input of the filling vector value of the input feature, which can be more suitable for the processing efficiency of the image feature input of the input feature, and the filling method of the feature vector matrix based on the convolution filling can generalize the activation function of the CNN neural network, thereby obtaining a more objective classification for the subsequent output health assessment result, thereby improving the applicability of the wearable device, increasing the robustness of the health assessment of the wearable device, and making the trained model have stable and reliable endorsement for the classification result.

[0059] A health assessment system based on wearable device multi-source data fusion analysis is also provided, comprising a hearing feature processing generation module, a wearable device heart rate and blood pressure feature acquisition module, a multi-source data image generation module, a wearable device health assessment model construction module, and a personnel health assessment module.

[0060] The hearing feature processing generation module is used to collect and process the hearing feature vector of the wearer.

[0061] The wearable device heart rate and blood pressure feature acquisition module is used to acquire the heart rate and blood pressure features of the wearer according to the wearable device.

[0062] The multi-source data image generation module is used to obtain wearable device health assessment features based on a multi-source data fusion analysis method, wherein the multi-source data fusion analysis method is to perform multi-source data imaging on the hearing feature vector and the heart rate and blood pressure features to obtain wearable device image health assessment features with image format features.

[0063] The wearable device health assessment model construction module is used to input the wearable device health assessment features and the health assessment results into a neural network model improved based on the wearable device image health assessment features to construct a wearable device health assessment model.

[0064] The personnel health assessment module: according to the wearable device health assessment model, the personnel health of the personnel wearing or newly wearing the wearable device is assessed.

[0065] Further, the wearable device health assessment model is a CNN neural network model improved based on wearable device image health assessment features, the CNN neural network is a commonly used model for processing image features, the generalization ability of the model for classification processing of multi-source fusion monitoring data is improved by improving the CNN neural network model through convolution filling of image format features and further correcting the calculation method of the activation function, so that the accurate and objective health assessment accuracy of the model is improved.

[0066] Further, the hearing feature vector is calculated by the detection difference and the analysis difference, age data, detection distance data, noise data, analysis influence factors, detection influence factors, age influence factors, noise influence factors and distance influence factors.

[0067]

[0068] Among them, Indicates the hearing feature vector, J indicates the detection distance data, Z indicates the noise data, G indicates the age data, D indicates the detection difference, Y indicates the analysis difference, Indicates the distance influence factor, the noise influence factor, the age influence factor, the detection influence factor and the analysis influence factor respectively, and e indicates the calculation bias adjustment factor of the hearing feature vector.

[0069] On the one hand, when analyzing the hearing of the personnel of the wearable device, the hearing related data collected simply improves the work efficiency and is not prone to errors; on the other hand, the health assessment by using the model is based on the multi-source data fusion analysis method to obtain the wearable device health assessment features by using the hearing feature vector, the heart rate and the blood pressure features obtained by processing, the wearable device health assessment features and the wearable health assessment results are used to construct the wearable device monitoring evaluation model, so as to obtain more objective and convenient wearable device health assessment results.

[0070] Therefore, the beneficial effect of the present application is to utilize the existing hearing detection data to fuse basic physiological parameters in a multi-modal manner, to adapt to the input of a CNN neural network model improved based on a feature imaging format feature, that is, to image the feature vector value of the traditional input model to correct the CNN neural network model, thereby obtaining highly structured feature data, and to obtain multi-source imaging data based on the feature convolution mode, to obtain wearable device health evaluation features more suitable for wearable device multi-source data fusion analysis, to make the trained model have less parameters and stronger generalization ability, and to make the wearable device more objective in health evaluation of the wearer under the multi-source processing of the model based on the generated health evaluation features.

[0071] Of course, it can be understood that each embodiment of the present application can realize one of the effects, and the combination of multiple embodiments of the present application can realize all the effects described above, but it is not required that each embodiment of the present application realizes all the advantages and effects described above, because each embodiment of the present application can constitute a separate technical solution and make one or more contributions to the prior art, and each embodiment of the present application is not affected, and the scheme can still be implemented by deleting one embodiment.

[0072] The part of the module structure of the present application which is not particularly clear is subject to the content recorded in the prior art. The prior art mentioned in the foregoing background section and the specific embodiment section of the present application can be used as a part of the present application to understand the meaning of some technical features or parameters.

Claims

1. A health assessment method based on multi-source data fusion analysis of a wearable device, characterized in that, The method comprises: S1, collecting and processing the hearing feature vector of the wearer, obtaining the heart rate and blood pressure features of the wearer based on the wearable device, obtaining the wearable device health evaluation features based on the multi-source data fusion analysis method, and obtaining the health evaluation result of the wearer; the hearing feature vector is calculated by detection difference and analysis difference, age data, detection distance data, noise data, analysis influence factor, detection influence factor, age influence factor, noise influence factor and distance influence factor: ； wherein, is expressed as an auditory feature vector, is a specific feature vector value, J is expressed as detection distance data, Z is expressed as noise data, G is expressed as age data, D is expressed as a detection difference value, Y is expressed as an analysis difference value, J, Z, G, D, and Y are all one-dimensional feature vector values, respectively expressed as a distance influence factor, a noise influence factor, an age influence factor, a detection influence factor, and an analysis influence factor, the above influence factors are used to remove dimensional influence, E is expressed as a calculation bias adjustment factor of the auditory feature vector; the detection difference value is obtained by the difference between the wearing detection start time and the wearing detection end time of the wearer, and the analysis difference value is obtained by the difference between the wearing evaluation analysis start time data and the wearing evaluation analysis end time data. S2, the multi-source data image generation based on the hearing feature vector and the heart rate and blood pressure features is a wearable device image health evaluation feature with image format characteristics; S3, based on the wearable device health evaluation features and the health evaluation result, input into the neural network model improved based on the wearable device image health evaluation features to construct a wearable device health evaluation model; S4, according to the wearable device health evaluation model, the health of the wearer of the wearable device is evaluated.

2. The health evaluation method based on multi-source data fusion analysis of wearable device according to claim 1, wherein: The wearable device health evaluation model is a CNN neural network model improved based on the wearable device image health evaluation features.

3. The health evaluation method based on multi-source data fusion analysis of wearable device according to claim 2, wherein: The heart rate and blood pressure features of the wearer are composed of a two-dimensional feature matrix composed of the maximum heart rate, minimum heart rate, minimum blood pressure and maximum blood pressure of the wearer.

4. The health evaluation method based on multi-source data fusion analysis of wearable device according to claim 3, wherein: The multi-source data image generation is specifically filling the four-dimensional feature matrix with the hearing feature vector as the reference of the heart rate and blood pressure features of the wearer, and the calculation method of the filling value is: ； In the formula, represents a filling vector value, represents an auditory feature vector, D represents a detection difference value, Y represents an analysis difference value, and the specific four-dimensional feature matrix is a wearable device image health evaluation feature, that is, the four-dimensional feature matrix is 。 5. The health evaluation method based on multi-source data fusion analysis of wearable device according to claim 4, wherein: The detection difference is obtained by the difference between the detection start time and the detection end time of the wearer, and the analysis difference is obtained by the difference between the analysis start time data and the analysis end time data of the wearable evaluation.

6. The health evaluation method based on multi-source data fusion analysis of wearable device according to claim 5, wherein: The wearable device health evaluation model uses the filling vector value based on the hearing feature vector to correct the activation function of the CNN neural network model, and the specific expression of the activation function is: ； In the formula, is an activation function value, is a padding vector value, respectively represent distance influence factor, noise influence factor, age influence factor, detection influence factor and analysis influence factor, and M is a vector value after weighted summation of the input four-dimensional feature matrix in the neural network.

7. A health evaluation system based on multi-source data fusion analysis of wearable device, which realizes the method according to claim 1, comprising an auditory feature processing generation module, a wearable device heart rate and blood pressure feature acquisition module, a multi-source data image generation module, a wearable device health evaluation model construction module, and a personnel health evaluation module, wherein: The auditory feature processing generation module is used for collecting and processing the hearing feature vector of the wearer. The wearable device heart rate and blood pressure feature acquisition module: acquires the heart rate and blood pressure features of the wearer according to the wearable device; The multi-source data imaging generation module: obtains the wearable device health evaluation features based on a multi-source data fusion analysis method, which is obtained by imaging the hearing feature vector and the heart rate and blood pressure features to obtain wearable device image health evaluation features with image format features; The wearable device health evaluation model construction module: constructs a wearable device health evaluation model based on the wearable device health evaluation features and health evaluation results input into a neural network model improved based on the wearable device image health evaluation features; The personnel health evaluation module: performs personnel health evaluation on the personnel wearing the wearable device again or newly wearing the wearable device according to the wearable device health evaluation model.

8. The health evaluation system based on wearable device multi-source data fusion analysis according to claim 7, wherein: The wearable device health evaluation model is a CNN neural network model improved based on the wearable device image health evaluation features.

9. The health evaluation system based on wearable device multi-source data fusion analysis according to claim 8, wherein: The hearing feature vector is calculated from the detection difference and the analysis difference, age data, detection distance data, noise data, analysis influence factors, detection influence factors, age influence factors, noise influence factors, and distance influence factors. ； wherein, is expressed as an auditory feature vector, is a specific feature vector value, J is expressed as detection distance data, Z is expressed as noise data, G is expressed as age data, D is expressed as a detection difference value, Y is expressed as an analysis difference value, J, Z, G, D, and Y are all one-dimensional feature vector values, respectively expressed as a distance influence factor, a noise influence factor, an age influence factor, a detection influence factor, and an analysis influence factor, the above influence factors are used to remove dimensional influence, E is expressed as a calculation bias adjustment factor of the auditory feature vector, the detection difference value is obtained by a difference value of a wearing detection start time and a wearing detection end time of the wearer, and the analysis difference value is obtained by a difference value of wearing evaluation analysis start time data and wearing evaluation analysis end time data.

Images