Pulse data detection method and system

Abstract

The pulse data detection method and system of the present invention obtain the pulse discrimination result and pulse rate monitoring result corresponding to the pulse data to be detected through the constructed pulse detection model. The scheme of the present invention can analyze whether it is valid data while detecting the pulse, and predict the value of the pulse rate when it is judged to be a correct pulse signal; it can also exit the invalid collection process in advance, greatly shorten the invalid collection time, and play a positive role in improving the user experience of pulse detection products.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a pulse data detection method and system. Background Art

[0002] In the field of professional Chinese medicine pulse detection, a pulse detection process usually takes a long time, often more than 5 minutes. However, due to the extremely high positioning requirements of the device for the detection position, there are often problems such as incorrect wearing, and it is impossible to obtain effective pulse signals. Conventional pulse signal processing methods require analysis after the signal is obtained. If the correct signal is not detected successfully, a lot of invalid experience time will be wasted.

[0003] In most of the previous research and patents, the main attention has been focused on the processing of the characteristics such as the position, number, shape, potential of the pulse from the pulse signal, and the processing procedure of valid data is rarely mentioned. In the dynamic Chinese medicine three-part pulse electrical signal continuous monitoring and real-time analysis system disclosed in CN106859608A, the signal detected by the default device is all correct and effective pulse signal, ignoring the unsuccessful processing procedure of obtaining the signal. In a kind of pulse recognition method based on the fusion of dual-path convolutional neural network disclosed in CN112487945A, although the processing of the pulse signal has been processed by artificial intelligence technology, it is also only to think correct and effective pulse signal for pre-treatment, and its network structure cannot process abnormal signal. Summary of the invention

[0004] In view of the above-mentioned shortcomings of the prior art, the purpose of this application is to provide a pulse data detection method and system to solve the problems arising from the above-mentioned prior art.

[0005] To achieve the above-mentioned objectives and other related objectives, the present application provides a pulse data detection method, comprising: obtaining pulse data to be detected; based on a deployed pulse detection model, obtaining a pulse detection result corresponding to the data according to the pulse data; wherein the pulse detection result comprises: a pulse discrimination result and a pulse rate monitoring result; and wherein the pulse discrimination result comprises: a valid pulse result or an invalid pulse result; wherein the pulse detection model is constructed in a manner comprising: constructing a pulse detection framework model; training the pulse detection framework model using a training data set to obtain a pulse detection model; and wherein the training data set comprises: a valid training data set and an invalid training data set.

[0006] In one or more embodiments of the present application, the valid training data set includes: multiple valid pulse data samples with valid classification labels and the pulse rate corresponding to each sample; the invalid training data set includes: multiple invalid pulse data samples with invalid classification labels.

[0007] In one or more embodiments of the present application, the training data set is obtained by Z-Score standardization.

[0008] In one or more embodiments of the present application, the pulse detection framework model includes: a feature coarse extraction structure, which is used to perform coarse feature extraction on the input pulse data to output coarse pulse feature extraction data; a feature fine extraction structure, which is connected to the feature coarse extraction structure, which is used to perform fine feature extraction on the coarse pulse feature extraction data to output fine pulse feature extraction data; a pulse detection result output structure, which is connected to the feature fine extraction structure, which is used to obtain a pulse detection result corresponding to the pulse data based on the input fine feature extraction data; wherein the pulse detection result output structure includes: a data validity judgment output structure, which is used to output a valid pulse result or an invalid pulse result based on the input pulse fine feature extraction data; and a pulse rate output structure, which is used to output a pulse rate monitoring result based on the input pulse fine feature extraction data.

[0009] In one or more embodiments of the present application, the data validity determination output structure includes: a two-layer connected Dense structure; and / or the pulse rate output structure includes: a two-layer connected Dense structure.

[0010] In one or more embodiments of the present application, the feature coarse extraction structure includes: two connected feature coarse sub-extraction structures; wherein each feature coarse sub-extraction structure includes: a convolutional layer, a Dropout layer connected to the convolutional layer, and a BatchNormalization layer connected to the Dropout layer.

[0011] In one or more embodiments of the present application, the feature fine extraction structure includes: a branch feature extraction structure, including: three branch structures corresponding to different convolution kernels, which are used to output corresponding branch feature data according to the input coarse feature extraction data; wherein each branch structure includes: a three-layer connected ConvBNRelu_x structure, which is used to perform three residual calculations on the input coarse feature extraction data; a global average pooling layer, which is used to average the data after three residual processing; a fusion structure, which is connected to the branch feature extraction structure, and is used to fuse the branch feature data output by the three branch structures to output fine feature extraction data.

[0012] In one or more embodiments of the present application, the activation function used by the data validity determination output structure is a sigmoid function; the activation function used by the pulse rate output structure is a Relu function.

[0013] In one or more embodiments of the present application, the pulse detection framework model also includes: a loss weighted calculation layer, connected to the pulse detection result output structure, including: a first loss weighted calculation module, used to use a binary cross entropy loss function to perform loss weighted calculation on the pulse discrimination result; a second loss weighted calculation module, used to use an MSE loss function to perform loss weighted calculation on the pulse rate monitoring result.

[0014] To achieve the above-mentioned objectives and other related objectives, the present application provides a pulse data detection system, comprising: a data acquisition module, used to acquire pulse data to be detected; a pulse detection module, connected to the data acquisition module, used to obtain pulse detection results corresponding to the data according to the pulse data based on a deployed pulse detection model; wherein the pulse detection results include: pulse discrimination results and pulse rate monitoring results; and wherein the pulse discrimination results include: valid pulse results or invalid pulse results; wherein the pulse detection model is constructed in a manner including: constructing a pulse detection framework model; training the pulse detection framework model using a training data set to obtain a pulse detection model; and wherein the training data set includes: a valid training data set and an invalid training data set.

[0015] As described above, the pulse data detection method and system of the present application obtain the pulse discrimination result and pulse rate monitoring result corresponding to the pulse data to be detected through the constructed pulse detection model. The scheme of the present invention can analyze whether it is valid data while detecting the pulse, and predict the value of the pulse rate when it is judged to be a correct pulse signal; it can also exit the invalid collection process in advance, greatly shorten the invalid collection time, and play a positive role in improving the user experience of pulse detection products. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Shown is a flow chart of the pulse data detection method in the embodiment of the present application.

[0017] Figure 2 Shown is a structural schematic diagram of the pulse detection framework model in an embodiment of the present application.

[0018] Figure 3 Shown is a structural schematic diagram of a feature rough extraction structure in an embodiment of the present application.

[0019] Figure 4 Shown is a structural schematic diagram of a feature fine extraction structure in an embodiment of the present application.

[0020] Figure 5 Shown is a structural schematic diagram of the ConvBNRelu_x structure in an embodiment of the present application.

[0021] Figure 6Shown is a flow chart of the pulse data detection method in the embodiment of the present application.

[0022] Figure 7 Shown is a structural schematic diagram of the pulse detection framework model in an embodiment of the present application.

[0023] Figure 8 Shown is a structural schematic diagram of the pulse data detection system in an embodiment of the present application. DETAILED DESCRIPTION

[0024] The following describes the implementation of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in this specification. The present application can also be implemented or applied through other different specific implementations, and the details in this specification can also be modified or changed in various ways based on different viewpoints and application systems without departing from the spirit of the present application. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0025] The following is a detailed description of the embodiments of the present application with reference to the accompanying drawings so that those skilled in the art can easily implement the present application. The present application can be embodied in a variety of different forms and is not limited to the embodiments described herein.

[0026] In order to clearly describe the present application, parts not related to the description are omitted, and the same reference numerals are given to the same or similar components throughout the specification.

[0027] Throughout the specification, when a component is said to be "connected" to another component, this includes not only the case of "direct connection" but also the case of "indirect connection" by placing other elements therebetween. In addition, when a component is said to "include" a certain constituent element, unless otherwise stated, it does not exclude other constituent elements, but means that other constituent elements may be included.

[0028] When a component is said to be "on" another component, it may be directly on the other component, but other components may also be present between it. In contrast, when a component is said to be "directly" "on" another component, no other components are present between it.

[0029] Although the terms first, second, etc. are used to describe various elements in this article in some examples, these elements should not be limited by these terms. These terms are only used to distinguish one element from another element. For example, the first interface and the second interface are described. Moreover, as used in this article, the singular forms "one", "one" and "the" are intended to also include plural forms, unless there is an opposite indication in the context. It should be further understood that the terms "comprising" and "including" indicate that there are the described features, steps, operations, elements, components, projects, kinds, and / or groups, but do not exclude the existence, occurrence or addition of one or more other features, steps, operations, elements, components, projects, kinds, and / or groups. The terms "or" and "and / or" used herein are interpreted as inclusive, or mean any one or any combination. Therefore, "A, B or C" or "A, B and / or C" means "any of the following: A; B; C; A and B; A and C; B and C; A, B and C". Only when the combination of elements, functions, steps or operations is inherently mutually exclusive in some way, will there be an exception to this definition.

[0030] The technical terms used herein are only used to refer to specific embodiments and are not intended to limit the present application. The singular form used herein also includes the plural form unless the sentence clearly indicates the contrary meaning. The meaning of "including" used in the specification is to specify specific characteristics, regions, integers, steps, operations, elements and / or components, and does not exclude the existence or addition of other characteristics, regions, integers, steps, operations, elements and / or components.

[0031] Relative spatial terms such as "below" and "above" may be used to more easily describe the relationship of one component to another component illustrated in the drawings. Such terms are intended to include not only the meaning indicated in the drawings, but also other meanings or operations of the device in use. For example, if the device in the drawings is turned over, a component that was previously described as being "below" other components is described as being "above" the other components. Therefore, the exemplary term "below" includes both above and below. The device may be rotated 90° or other angles, and the relative spatial terms may be interpreted accordingly.

[0032] Although not defined differently, all terms, including technical and scientific terms used herein, have the same meaning as those generally understood by those skilled in the art to which this application belongs. Terms defined in commonly used dictionaries are additionally interpreted as having meanings consistent with the relevant technical literature and the contents of the current disclosure, and unless defined, shall not be overly interpreted as ideal or very formal meanings.

[0033] In view of the shortcomings of the prior art, a pulse data detection method is provided, and a pulse discrimination result and a pulse rate monitoring result corresponding to the pulse data to be detected are obtained through a constructed pulse detection model. The scheme of the present invention can analyze whether it is valid data while detecting the pulse, and predict the value of the pulse rate when it is judged to be a correct pulse signal; it can also exit the invalid collection process in advance, greatly shorten the invalid collection time, and play a positive role in improving the user experience of pulse detection products.

[0034] The following is a detailed description of the embodiments of the present invention with reference to the accompanying drawings so that those skilled in the art can easily implement the present invention. The present invention can be embodied in many different forms and is not limited to the embodiments described herein.

[0035] like Figure 1 Shown is a schematic flow chart showing the pulse data detection method in an embodiment of the present invention.

[0036] The method comprises:

[0037] Step S11: Obtain pulse data to be detected.

[0038] Optionally, the pulse data is fixed-length data; for example, in order to quickly identify data, the input length is set to be 4 seconds of original data.

[0039] Optionally, a device with a sampling rate of 100 sps and a channel number of 1 is used to sample the pulse signal, so the dimension of the collected data is (400, 1).

[0040] Step S12: Based on the deployed pulse detection model, a pulse detection result corresponding to the pulse data is obtained according to the pulse data.

[0041] Specifically, the pulse detection result includes: a pulse discrimination result and a pulse rate monitoring result; and wherein the pulse discrimination result includes: a valid pulse result or an invalid pulse result. The pulse detection model is constructed in a manner including: constructing a pulse detection framework model; training the pulse detection framework model using a training data set to obtain a pulse detection model; and wherein the training data set includes: a valid training data set and an invalid training data set.

[0042] It should be noted that when an invalid pulse result is obtained, the corresponding pulse rate monitoring result should be ignored and its value has no actual meaning.

[0043] Optionally, the valid training data set includes: a plurality of valid pulse data samples with valid classification labels and the pulse rate corresponding to each valid pulse data sample; the invalid training data set includes: a plurality of invalid pulse data samples with invalid classification labels.

[0044] Preferably, the valid pulse data and invalid non-pulse data obtained in the past are marked separately, the valid pulse data is marked with a classification label of 1, and the invalid data is marked with a classification label of 0, for training the classification results. At the same time, in order to output the pulse rate in the model, the pulse rate in the data can also be marked, and the non-pulse data does not have a pulse rate result, and its value can be set to a value, such as 0, by default when marking the data.

[0045] Optionally, since the effective pulse data may present different amplitude heights under different pulse pressures, in order to ignore the influence of pulse pressure and effective peak-to-peak value, and to make the trained model more robust, the pulse data is subjected to Z-Score standardization. The data values ​​after the effective pulse data and the invalid pulse data are subjected to Z-Score standardization to obtain the processed training data set are:

[0046]

[0047] Among them, mean(x) represents the mean of the input data sequence x, and std(x) represents the standard deviation of the input data sequence x.

[0048] Optional, such as Figure 2 As shown, the pulse detection framework model includes: a feature coarse extraction structure 21, which is used to perform coarse feature extraction on the input pulse data to output coarse pulse feature extraction data; a feature fine extraction structure 22, which is connected to the feature coarse extraction structure and is used to perform fine feature extraction on the coarse pulse feature extraction data to output fine pulse feature extraction data; a pulse detection result output structure 23, which is connected to the feature fine extraction structure 22 and is used to obtain a pulse detection result corresponding to the pulse data based on the input fine feature extraction data; wherein the pulse detection result output structure 23 includes: a data validity judgment output structure 231, which is used to output a valid pulse result or an invalid pulse result based on the input pulse fine feature extraction data; and a pulse rate output structure 232, which is used to output a pulse rate monitoring result based on the input pulse fine feature extraction data.

[0049] Optional, such as Figure 3As shown, the feature rough extraction structure includes: two connected feature rough sub-extraction structures 311 and 312; that is, the pulse data to be detected is input into the feature rough sub-extraction structure 311 to output the first rough feature extraction data, and then the first rough feature extraction data is input into the feature rough sub-extraction structure 312 to output the rough feature extraction data; wherein each feature rough sub-extraction structure includes: a convolution layer Conv, used to map the data to the feature space; a Dropout layer, connected to the convolution layer, used to remove the training unit of the neural network from the network according to a certain probability, so that the activation value of its neuron is suspended, so that the model generalization is stronger; preferably, the probability value of its removal is set to 0.4. A BatchNormalization layer, connected to the Dropout layer, used to normalize the upper layer output value to solve the influence of input data transmission offset and increase.

[0050] Optional, such as Figure 4 As shown, the feature extraction structure includes:

[0051] The branch feature extraction structure includes: three branch structures 411, 412 and 413 corresponding to different convolution kernels, which are used to output corresponding branch feature data according to the input coarse feature extraction data; different convolution kernel sizes are set to extract features at different scales.

[0052] Each branch structure includes: a three-layer connected ConvBNRelu_x structure, which is used to perform three residual calculations on the input coarse feature extraction data; the residual structure is used to prevent the gradient disappearance problem after the network becomes deeper; a global average pooling layer, connected to the ConvBNRelu_x structure, is used to average the data after three residual processing and reduce the dimension of the feature output;

[0053] The fusion structure 42 is connected to the branch feature extraction structure and is used to fuse the branch feature data output by the three branch structures and output fine feature extraction data.

[0054] Optionally, the sizes of the convolution kernels used in the three branch structures are 3, 5, and 7 respectively.

[0055] Optional, such as Figure 5 As shown, the residual structure used by each ConvBNRelu_x structure includes: two interconnected structures having a convolutional layer, a BatchNormalization layer, and a relu function layer connected in sequence, and a residual structure layer of a skip connection that directly adds the input to the output.

[0056] Optionally, the activation function used by the data validity judgment output structure is a sigmoid function; and the activation function used by the pulse rate output structure is a Relu function.

[0057] Optionally, the data validity determination output structure includes: a two-layer connected Dense structure; and / or the pulse rate output structure includes: a two-layer connected Dense structure.

[0058] Optionally, the pulse detection framework model is trained using a prepared training data set, so that the computer automatically calculates the parameters of the model so that the output result is as close as possible to the marked result, and the Adam optimizer is used as the optimizer for model training to train the model;

[0059] Preferably, for setting the model training parameters, the Adam optimizer can be used as the optimizer for model training for 300 epochs. The initial learning rate is set to 0.0001, and is reduced to 0.1 times after 200 epochs.

[0060] Optionally, the pulse detection framework model further includes: a loss calculation layer, connected to the pulse detection result output structure, for performing loss weighted calculation on the output pulse detection result;

[0061] The loss calculation layer includes: a first loss weighted calculation module, which is used to perform loss weighted calculation on the effective pulse result or the invalid pulse result by using a binary cross entropy loss function; and a second loss weighted calculation module, which is used to perform loss weighted calculation on the effective pulse rate or the invalid pulse rate by using a Mask-MSE loss function.

[0062] Optionally, the MSE loss function adopts a Mask-MSE loss function, and the Mask-MSE loss function includes:

[0063]

[0064] The binary cross entropy loss function is:

[0065]

[0066] Considering that the pulse rate value marked as 0 in the invalid data will cause a large disturbance to the loss value during the training process, the classification label is used as a mask to further calculate the loss value during the calculation, so that the pulse rate output of the invalid signal does not affect the model training process. That is, the parameter α is used to control the relationship between the classification loss and the pulse rate loss; that is, the model loss calculation formula of the loss weighted calculation layer is:

[0067] loss = αloss (cls) +(1-α)loss (hr) ; (4)

[0068] Optionally, the pulse detection model is deployed to the application end, that is, the trained pulse detection model is deployed to the application end; although the training of the artificial neural network model needs to be performed on a GPU device, the application reasoning process is simpler than the training process, and the requirements for the equipment are not as high as the training process. The calculation amount of the model parameters designed this time can be inferred on a general CPU device, and can also have good operating efficiency on the application end.

[0069] Therefore, the pulse detection model is relatively lightweight and can be deployed on mobile devices. It does not require a high-configuration cloud server or network support, and its usage scenarios are not restricted by network or other reasons.

[0070] Optionally, the pulse detection model deployed to the application end is subjected to model reasoning, that is, the pulse signal acquisition device of the application end can intercept the latest acquired data fragment when the element signal is acquired and the data length meets the model input length, and perform Z-Score standardization on the data in the same way as the data set is processed, and then send it into the model for reasoning, and obtain the classification result and the pulse rate result at the two output nodes of the model respectively. First, it should be determined whether the result on the classification node is a valid pulse signal. If it is a valid pulse signal, continue to read the result on the pulse rate node. Otherwise, the output value of the pulse rate node should be ignored, and its value has no actual meaning.

[0071] In order to better explain the pulse data detection method, the following specific embodiments are provided for illustration.

[0072] Embodiment 1: a kind of pulse condition data detection method, Figure 6 It is a schematic flow chart of the pulse data detection method.

[0073] The method comprises:

[0074] Step 1: Process the dataset;

[0075] The main structure of the technical solution is carried out by using an artificial neural network training model. Its training process cannot be separated from the support of the data set, and a successful neural network model cannot be separated from a good data set. Therefore, the correct pulse data and incorrect non-pulse data obtained in the past are marked separately, and the correct pulse data marking classification label is 1, and the incorrect data marking classification label is 0, which is used to train the results of classification. At the same time, in order to output the pulse rate in the model, the pulse rate results in the data fragments should also be marked, and there is no pulse rate result in the non-pulse data, and its value is set to 0 by default when marking the data. For effective pulse data, due to the different amplitude heights that may be presented under different pulse pressures, in order to ignore the influence of pulse pressure and effective peak-to-peak value, and also in order to make the trained model have better robustness, the data is subjected to Z-Score standardization. The calculation formula is:

[0076]

[0077] In the formula, mean(x) represents the mean of the input data x sequence, and std(x) represents the standard deviation of the input data x sequence.

[0078] Step 2: Model building: Build a pulse detection framework model, such as Figure 7 Shown is a schematic diagram of the structure of the pulse detection framework model;

[0079] Among them, the pulse detection framework model includes:

[0080] Two Conv+Dropout+BN structures are used for rough feature extraction. The Conv+Dropout+BN structure contains a convolution layer + a Dropout layer + a BatchNormalization layer. The data is mapped to the feature space through the convolution layer. The Dropout layer is used to remove the training unit of the neural network from the network with a certain probability, so that the activation value of its neuron is suspended. This can make the model more generalizable. The probability value of its removal is set to 0.4. BatchNormalization normalizes the output value of the upper layer to solve the impact of input data transmission offset and increase.

[0081] The ConvBNRelu_x structure has three branches in parallel, and each ConvBNRelu_x structure uses a residual structure. Each branch has three levels, and the sizes of the convolution kernels in the branches are 3, 5, and 7 respectively. Different convolution kernel sizes are set to extract features at different scales. The residual structure is used to prevent the gradient from disappearing after the network becomes deeper. After completing the three residual processing, the global average pooling layer GlobalAvePool is used to average the output elements and reduce the dimension of the feature output.

[0082] The fusion structure fuses the branch feature data output by the three branch structures.

[0083] The pulse detection result output structure has two tasks: one is to judge whether the data is valid pulse data as a classification result, and the other is that the output node of the pulse rate recognition task needs to output a specific pulse rate value; that is, it includes:

[0084] The data effective discrimination output structure includes: a two-layer connected Dense structure, which is used to output a valid pulse result or an invalid pulse result based on the input pulse fine feature extraction data; the pulse rate output structure includes: a two-layer connected Dense structure, which is used to output a valid pulse rate or an invalid pulse rate based on the input pulse fine feature extraction data; its pulse detection result output structure uses the Sigmoid function as the activation function, and its result can be considered as the probability value of whether it is pulse data. The pulse rate output structure uses the Relu function as the activation function.

[0085] The loss weighted calculation layer, in which the output layer of the pulse signal classification result uses binary cross entropy as the loss function, and the pulse rate recognition result uses Mask-MSE to calculate the error value.

[0086] Step 3: Train the network model;

[0087] The model was trained using the prepared data set, so that the computer automatically calculated the parameters of the model to make the output result as close to the labeled result as possible. In this model training, the Adam optimizer was used as the optimizer for model training for 300 epochs. The initial learning rate was set to 0.0001, which was reduced to 0.1 times after 200 epochs. The output layer of the pulse signal classification result used binary cross entropy as the loss function, and the pulse rate recognition result used Mask-MSE to calculate the error value.

[0088] 4. Deploy the model to the application end;

[0089] Deploy the trained network model to the application end; although the training of the artificial neural network model needs to be performed on a GPU device, the application reasoning process is simpler than the training process, and the requirements for the device are not as high as the training process. The calculation amount of the model parameters designed this time can be inferred on a general CPU device, and it also has good operating efficiency on the mobile end;

[0090] 5. Model reasoning;

[0091] When the pulse signal acquisition device at the application end obtains the element signal and the data length meets the model input length, it can intercept the latest data fragment, and perform Z-Score standardization on the data in the same way as processing the data set, and then send it to the model for reasoning, and obtain the classification result and pulse rate result at the two output nodes of the model respectively. First, it should be determined whether the result on the classification node is a valid pulse signal. If it is a valid pulse signal, continue to read the result on the pulse rate node. Otherwise, the output value of the pulse rate node should be ignored, and its value has no actual meaning.

[0092] 6. Pulse data detection;

[0093] The acquired pulse data to be detected is input into the constructed model and the pulse detection result corresponding to the data is obtained.

[0094] The model of this embodiment can analyze whether it is valid data while the pulse detection device is detecting, and can exit the invalid collection process in advance, greatly shortening the invalid collection time, which has a positive effect on improving the user experience of pulse detection products. In addition, the model is relatively light and can be deployed on mobile devices, without the need to use a high-configuration cloud server, without network support, and the use scenario is not limited by network and other reasons. After data training, the model proposed by the present invention has a classification accuracy of more than 97%, an average absolute pulse rate error of 0.18bpm, and samples with a pulse rate error exceeding ±3bpm account for only 6.5 per 10,000.

[0095] Similar to the principle of the above embodiment, the present invention provides a pulse data detection system.

[0096] The following provides specific embodiments in conjunction with the accompanying drawings:

[0097] like Figure 8 A structural schematic diagram of a pulse data detection system in an embodiment of the present invention is shown.

[0098] The system comprises:

[0099] Data acquisition module 81, used for acquiring pulse data to be detected;

[0100] The pulse detection module 82 is connected to the data acquisition module 81 and is used to obtain a pulse detection result corresponding to the data according to the pulse data based on the deployed pulse detection model; wherein the pulse detection result includes: a pulse discrimination result and a pulse rate monitoring result; and wherein the pulse discrimination result includes: a valid pulse result or an invalid pulse result;

[0101] Wherein, the construction method of the pulse detection model includes:

[0102] Construct pulse detection framework model;

[0103] The pulse detection framework model is trained using a training data set to obtain a pulse detection model; and wherein the training data set includes: a valid training data set and an invalid training data set.

[0104] It should be understood that Figure 8 The division of each module in the system embodiment is only a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or physically separated. And these modules can be all implemented in the form of software calling through processing elements; or all implemented in the form of hardware; or some modules can be implemented in the form of software calling through processing elements, and some modules can be implemented in the form of hardware;

[0105] For example, each module may be one or more integrated circuits configured to implement the above method, such as: one or more application specific integrated circuits (ASIC), or one or more digital signal processors (DSP), or one or more field programmable gate arrays (FPGA). For another example, when a module above is implemented in the form of a processing element scheduling program code, the processing element may be a general-purpose processor, such as a central processing unit (CPU) or other processor that can call program code. For another example, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0106] It should be noted that since the structure and implementation principle of the pulse data detection system and the pulse detection model have been described in the aforementioned embodiments, they will not be repeated here.

[0107] In summary, the pulse data detection method and system of the present application obtain the pulse discrimination result and pulse rate monitoring result corresponding to the pulse data to be detected through the constructed pulse detection model. The scheme of the present invention can analyze whether it is valid data while detecting the pulse, and predict the value of the pulse rate when it is judged to be a correct pulse signal; it can also exit the invalid collection process in advance, greatly shorten the invalid collection time, and play a positive role in improving the user experience of pulse detection products.

[0108] The above embodiments are merely illustrative of the principles and effects of the present application and are not intended to limit the present application. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present application. Therefore, all equivalent modifications or changes made by a person of ordinary skill in the art without departing from the spirit and technical ideas disclosed in the present application shall still be covered by the claims of the present application.

Claims

1. A pulse data detection method, characterized in that, include: Acquire pulse data to be detected; Based on the deployed pulse detection model, a pulse detection result corresponding to the data is obtained according to the pulse data; wherein the pulse detection result includes: a pulse discrimination result and a pulse rate monitoring result; and wherein the pulse discrimination result includes: a valid pulse result or an invalid pulse result; Wherein, the construction method of the pulse detection model includes: Construct pulse detection framework model; The pulse detection framework model is trained using a training data set to obtain a pulse detection model; wherein the training data set includes: a valid training data set and an invalid training data set; The pulse data is fixed-length data, and the method of obtaining the pulse data includes: cutting the collected original pulse data to meet the input length of the pulse detection model, and performing Z-Score standardization processing on the obtained data to obtain the pulse data; Described pulse condition detection framework model comprises: A feature rough extraction structure is used to perform rough feature extraction on the input pulse data to output rough pulse feature extraction data; the feature rough extraction structure includes: two connected feature rough sub-extraction structures; each feature rough sub-extraction structure includes: a convolution layer, a Dropout layer connected to the convolution layer, and a BatchNormalization layer connected to the Dropout layer; A fine feature extraction structure is connected to the coarse feature extraction structure and is used to perform fine feature extraction on the coarse pulse feature extraction data to output the fine pulse feature extraction data; the fine feature extraction structure includes: a branch feature extraction structure, including: three branch structures corresponding to different convolution kernels, which are used to output corresponding branch feature data according to the input coarse feature extraction data; wherein each branch structure includes: a three-layer connected ConvBNRelu_x structure, which is used to perform three residual calculations on the input coarse feature extraction data; a global average pooling layer, which is used to perform average processing on the data processed by the three residuals; a fusion structure, which is connected to the branch feature extraction structure and is used to fuse the branch feature data output by the three branch structures to output fine feature extraction data; The pulse detection result output structure is connected to the feature fine extraction structure and is used to obtain the pulse detection result corresponding to the pulse data based on the input fine feature extraction data; the pulse detection result output structure includes: a data valid discrimination output structure and a pulse rate output structure; the data valid discrimination output structure includes: a two-layer connected Dense structure, which is used to output a valid pulse result or an invalid pulse result based on the input pulse fine feature extraction data; the pulse rate output structure includes: a two-layer connected Dense structure, which is used to output a pulse rate monitoring result based on the input pulse fine feature extraction data; The loss weighted calculation layer is connected to the pulse detection result output structure, and includes: a first loss weighted calculation module and a second loss weighted calculation module; the first loss weighted calculation module is used to use a binary cross entropy loss function to perform loss weighted calculation on the pulse discrimination result; the second loss weighted calculation module is used to use an MSE loss function to perform loss weighted calculation on the pulse rate monitoring result.

2. according to the pulse data detection method described in claim 1, it is characterized in that, The effective training data set includes: a plurality of effective pulse data samples with effective classification labels and the pulse rate corresponding to each sample; the invalid training data set includes: a plurality of invalid pulse data samples with invalid classification labels.

3. according to the pulse data detection method described in claim 1, it is characterized in that, The activation function used by the data effective discrimination output structure is the sigmoid function; the activation function used by the pulse rate output structure is the Relu function.

4. A pulse data detection system, characterized in that, include: A data acquisition module is used to acquire pulse data to be detected; A pulse detection module, connected to the data acquisition module, is used to obtain a pulse detection result corresponding to the data according to the pulse data based on the deployed pulse detection model; wherein the pulse detection result includes: a pulse discrimination result and a pulse rate monitoring result; and wherein the pulse discrimination result includes: a valid pulse result or an invalid pulse result; Wherein, the construction method of the pulse detection model includes: Construct pulse detection framework model; The pulse detection framework model is trained using a training data set to obtain a pulse detection model; wherein the training data set includes: a valid training data set and an invalid training data set; The pulse data is fixed-length data, and the method of obtaining the pulse data includes: cutting the collected original pulse data to meet the input length of the pulse detection model, and performing Z-Score standardization processing on the obtained data to obtain the pulse data; Described pulse condition detection framework model comprises: A feature rough extraction structure is used to perform rough feature extraction on the input pulse data to output rough pulse feature extraction data; the feature rough extraction structure includes: two connected feature rough sub-extraction structures; each feature rough sub-extraction structure includes: a convolution layer, a Dropout layer connected to the convolution layer, and a BatchNormalization layer connected to the Dropout layer; A fine feature extraction structure is connected to the coarse feature extraction structure and is used to perform fine feature extraction on the coarse pulse feature extraction data to output the fine pulse feature extraction data; the fine feature extraction structure includes: a branch feature extraction structure, including: three branch structures corresponding to different convolution kernels, which are used to output corresponding branch feature data according to the input coarse feature extraction data; wherein each branch structure includes: a three-layer connected ConvBNRelux structure, which is used to perform three residual calculations on the input coarse feature extraction data; a global average pooling layer, which is used to perform average processing on the data processed by the three residuals; a fusion structure, which is connected to the branch feature extraction structure and is used to fuse the branch feature data output by the three branch structures to output fine feature extraction data; The pulse detection result output structure is connected to the feature fine extraction structure and is used to obtain the pulse detection result corresponding to the pulse data based on the input fine feature extraction data; the pulse detection result output structure includes: a data valid discrimination output structure and a pulse rate output structure; the data valid discrimination output structure includes: a two-layer connected Dense structure, which is used to output a valid pulse result or an invalid pulse result based on the input pulse fine feature extraction data; the pulse rate output structure includes: a two-layer connected Dense structure, which is used to output a pulse rate monitoring result based on the input pulse fine feature extraction data; The loss weighted calculation layer is connected to the pulse detection result output structure, and includes: a first loss weighted calculation module and a second loss weighted calculation module; the first loss weighted calculation module is used to use a binary cross entropy loss function to perform loss weighted calculation on the pulse discrimination result; the second loss weighted calculation module is used to use an MSE loss function to perform loss weighted calculation on the pulse rate monitoring result.

Images