A method for constructing a vital sign signal regression model based on a fuzzy set

Abstract

The application discloses a kind of based on fuzzy set's vital sign signal regression model construction method, belong to vital sign signal detection technical field, the construction method includes: it first defines multiple fuzzy sets, then the label corresponding to each sample is mapped to each fuzzy set respective fuzzy class to obtain the membership degree of each fuzzy class, and then the membership degree vector of each sample is obtained, to calculate the fuzzy class conditional probability distance of source domain to target domain;Finally, with the fuzzy class conditional probability distance of source domain to target domain minimum as target or basis, iteratively update the initial regression model for vital sign signal regression, until the initial regression model after iteration reaches preset condition, finally obtain vital sign signal regression model.Such design, consider the information of known source domain sample and regression fuzzy class information, can better utilize the information of known label, auxiliary construction better generalization vital sign signal regression model.

Description

Technical Field

[0001] This invention belongs to the field of vital sign signal detection technology, and more specifically, relates to a method for constructing a vital sign signal regression model based on fuzzy sets. Background Technology

[0002] With the rapid development of machine learning technology, many mature classification algorithms have been developed for vital signs signals. These include electromyography (EMG), electrooculography (EOG), electrocardiography (ECG), functional near-infrared spectroscopy (FIR), and electroencephalography (EEG).

[0003] Taking EEG signals acquired by brain-computer interfaces as an example, due to individual differences, different users exhibit different EEG signals for the same task, which greatly affects the portability of algorithms. Transfer learning uses data or knowledge from the source domain (old users) to assist learning in the target domain (new users), which can mitigate the impact of individual differences and significantly reduce the need for labeled data in the target domain.

[0004] Transfer learning algorithms can be broadly categorized into traditional transfer learning based on conventional machine learning and deep transfer learning algorithms based on neural networks. Traditional transfer learning algorithms primarily achieve transfer by modeling the distributional differences between the source and target domains and constructing an optimization problem. For example, the maximum mean distance is used to measure the difference in feature marginal probability distributions between domains. Building upon this, class information is introduced, and class-conditional probabilities are used to estimate conditional probabilities, approximating the difference in joint probability distributions between domains to achieve transfer. In addition to inter-domain transferability, class discriminability is also considered, requiring more significant class differences in the transferred distribution. Deep transfer learning algorithms primarily model inter-domain distance through loss, effectively aligning the features extracted by the feature extractor with the distributions of the source and target domains. For example, the joint probability of the outputs of different layers is used to approximate the joint probability distribution of the original data. The original domain is divided into subdomains by category, and alignment is performed on these subdomains to further enhance transfer performance.

[0005] However, most of the algorithms mentioned above require discrete category information to model conditional probability distributions, class-conditional probability distributions, or joint probability distributions. This makes these classification transfer learning algorithms unsuitable for regression problems involving continuously labeled vital sign signals. Summary of the Invention

[0006] In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides a method for constructing a vital sign signal regression model based on fuzzy sets. The purpose is to solve the technical problem that the probability distribution constructed with discrete category information cannot be used to construct a vital sign signal regression model.

[0007] To achieve the above objectives, according to one aspect of the present invention, a method for constructing a vital sign signal regression model based on fuzzy sets is provided, comprising: S1: Determine the target domain samples of vital sign signals, the source domain samples of vital sign signals, and their corresponding regression-type source domain labels; concatenate all the source domain labels to obtain a label domain, and then... Source domain tags Divide into multiple boundary points, and use the boundary point definition. A fuzzy set; S2: For each sample and its tags mapped to Each fuzzy class is obtained from the fuzzy class of each fuzzy set. membership degree , and make each fuzzy class membership degree Combined to obtain the sample Membership vector = ; k is the sequence number of the sample; if the sample If the sample is from the source domain, then the label is the true label; if the sample... If the sample is from the target domain, then the label is a pseudo-label; S3: Using the formula Calculate the fuzzy class-conditional probability distance from the source domain to the target domain. ; Identify the source domain. Identify the target domain. The number of samples in the source domain. The number of samples in the target domain. For the i-th source domain sample Membership degree of the corresponding c-th fuzzy class The normalized value; For the j-th target domain sample Membership degree of the corresponding c-th fuzzy class The normalized value, It is represented by distance; S4: The fuzzy class conditional probability distance from the source domain to the target domain. Minimize the target by designing an optimization function or loss function, and iteratively update the initial regression model for vital sign signal regression. During the iterative update process, use the iterated initial regression model to generate pseudo-labels for the target domain samples in the next iteration, and return to S2. The process continues until the iterated initial regression model meets the preset conditions, at which point the vital sign signal regression model is considered to be obtained.

[0008] Furthermore, the step of adding the tag field to... Source domain tags Multiple dividing points can be established, including: dividing points based on the percentage of the source domain label value; or, dividing points based on the percentage of the interval length corresponding to the theoretical upper and lower bounds of the label in the vital signs signal problem.

[0009] Furthermore, the fuzzy class conditional probability distance from the source domain to the target domain... The optimization function is designed with the goal of minimizing the input, and the initial regression model for vital sign signal regression is iteratively updated, including: utilizing the fuzzy class-conditional probability distance from the source domain to the target domain. The fuzzy joint probability distance is obtained from the category edge probability, and the optimization function is set as the optimization objective. To update and iterate the initial regression model; ; ; in, The joint probability distance matrix within the source and target domains. This is a weighting factor that balances the degree of intra-class distance and inter-class distance. Joint probability distance matrix between source and target domain classes. For the target domain Sample Belongs to the Membership degree of a fuzzy class For feature transformation, Let be the norm squared over a regenerated Hilbert space.

[0010] Furthermore, the fuzzy class conditional probability distance from the source domain to the target domain... The optimization function is designed with the goal of minimizing the initial regression model for vital sign signal regression, and the model is iteratively updated, including: using the fuzzy class-conditional probability distance from the source domain to the target domain. With the goal of minimizing, the optimization function is set as follows: To update and iterate the initial regression model; ; ; in, for The selected value within the range of values; This is a weighting factor that balances the degree of intra-class distance and inter-class distance. The joint probability distance matrix of the fuzzy class conditional probabilities of the source and target domains. The fuzzy class conditional probability distance matrix between source and target domain classes. For the target domain Sample Belongs to the Membership degree of a fuzzy class For feature transformation, Let be the norm squared over a regenerated Hilbert space.

[0011] Furthermore, To utilize The optimal value found; It is a hyperparameter used to limit the upper limit of the traversal range; ; for The second largest value in, It is the i-th source domain tag.

[0012] Furthermore, the class-conditional probability distance from the source domain to the target domain is... With the goal of minimizing the initial regression model for vital sign signal regression, iterative updates are performed, including setting the loss function as follows: The initial regression model based on the neural network is then updated and iterated; wherein, express The output of the layer, The membership matrix represents the degree of the source domain labels. The membership matrix represents the pseudo-labels of the target domain. express The feature space of the product of order tensors.

[0013] Furthermore, the fuzzy class conditional probability distance from the source domain to the target domain... The optimization function is designed with the goal of minimizing the initial regression model for vital sign signal regression, and the model is iteratively updated, including: using the class-conditional probability distance from the source domain to the target domain. Based on minimum principles, a fuzzy sub-neighborhood distance loss is constructed as the loss function. , This is to update and iterate the initial regression model based on the neural network; wherein, For neural network models, For source domain For the target domain sample, Let be the norm squared over a regenerated Hilbert space.

[0014] According to another aspect of the present invention, a method for regressing vital sign signals is provided, comprising: Collect the vital signs signal of the current user, preprocess it, and input it into the vital signs signal regression model to obtain the regression result corresponding to the vital signs signal of the current user.

[0015] According to another aspect of the present invention, a vital signs signal processing system is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of a method for constructing a vital signs signal regression model or a method for regressing vital signs signals.

[0016] According to another aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of a method for constructing a vital signs signal regression model or a method for regressing vital signs signals.

[0017] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: (1) This invention provides a method for constructing a vital sign signal regression model based on fuzzy sets, which first defines A fuzzy set, then each sample and its tags mapped to Each fuzzy class is obtained from the fuzzy class of each fuzzy set. membership degree Thus, each sample is obtained. Membership vector = Secondly, utilize Calculate the fuzzy class-conditional probability distance from the source domain to the target domain. Finally, the fuzzy class conditional probability distance from the source domain to the target domain is used as the basis for the calculation. Using a minimum as the target or baseline, the initial regression model for vital sign signal regression is iteratively updated until the iteratively updated initial regression model meets preset conditions, ultimately yielding the vital sign signal regression model. This design considers information from known source domain samples and regression fuzzy class information, allowing for better utilization of known label information and assisting in constructing a vital sign signal regression model with better generalization.

[0018] (2) In this scheme, the percentage-based partitioning of label values ​​is based on known source domain data, which ensures that the number of samples in each fuzzy class is similar. The fuzzy sets will be divided into more fuzzy sets in areas with higher density according to the distribution density changes of the known source domain samples. When the number of samples is large and the label distribution is relatively scattered, it can more naturally reflect the data distribution. In addition, the percentage-based partitioning of interval length is based on prior knowledge. By dividing the label intervals of the question, it is ensured that each fuzzy set covers labels of a fixed length interval, which better adapts to out-of-distribution information and improves generalization. At the same time, when the label distribution is a clustered unimodal distribution, the percentage-based partitioning of interval length can avoid the situation in the percentage-based partitioning of label values ​​where similar labels are incorrectly classified into different fuzzy classes.

[0019] (3) The optimization function is set as follows in this scheme: To update and iterate the initial regression model; ; The above method allows the joint probability distance in classification algorithms to be applied to regression problems, enabling the construction of models with better generalization in regression scenarios where the label distributions of the source and target domains are similar.

[0020] (4) The optimization function is set as follows in this scheme: To update and iterate the initial regression model; ; The above method can avoid instability caused by large differences in the distribution of labels in the source and target domains, and achieve better generalization in real-world scenarios where label distributions vary greatly.

[0021] (5) In this plan To utilize The optimal value is found; by using the above parameter selection method, the parameters that maximize the discriminative power of the fuzzy class can be automatically selected based on the data, and a regression model with better generalization can be constructed by aligning the conditional probabilities of the fuzzy class.

[0022] (6) This plan is set up The initial regression model based on the neural network is updated and iterated. Through the above method, the joint maximum mean distance loss in classification problems can be applied to regression problems, expanding the applicability of existing algorithms and constructing a joint adaptive regression model.

[0023] (7) The loss function of this scheme is set as follows: The initial regression model based on the neural network is updated and iterated. Using the above method, the sub-domain distance loss in the classification task can be applied to the regression problem to construct a regression sub-domain adaptation network. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the framework of the method for constructing a vital sign signal regression model based on fuzzy sets provided in Embodiment 1 of the present invention; Figure 2 This is an example diagram of fuzzy set and membership degree calculation provided in Embodiment 1 of the present invention; Figure 3 This is a structural diagram of a regression model provided in Embodiment 1 of the present invention; Figure 4 This is a structural diagram of another regression model provided in Embodiment 1 of the present invention; Figure 5 This is a schematic diagram of the framework of the method for constructing a vital sign signal regression model based on fuzzy sets provided in Embodiment 2 of the present invention. The schematic diagram takes a brain-computer interface regression application scenario as an example. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0026] Example 1 This embodiment provides a method for constructing a vital sign signal regression model based on fuzzy sets, including: S1-S4, as follows: Figure 1 As shown.

[0027] S1: Determine the target domain samples of vital sign signals, the source domain samples of vital sign signals, and their corresponding regression type source domain labels; concatenate all source domain labels to obtain the label domain, and then... Source domain tags Divide into multiple boundary points, using boundary point definitions. A fuzzy set.

[0028] The dividing point can be determined by the percentage of the source domain label values; or by the percentage of the interval length corresponding to the theoretical upper and lower bounds of the labels in the brain-computer interface problem. Specifically, it is assumed that there are multiple known subjects who jointly construct the source domain. And a target domain consisting of a target subject. This method first concatenates the source domain participant labels to obtain the label domain. For the labels in this domain... One regression label The dividing point is determined based on its percentage digit, and the dividing point definition is used. A fuzzy set is used to obtain linguistic variables. .by For example, using The 5th, 50th, and 95th percentile values ​​of each label , and Define a fuzzy set. When When increasing, and Keeping it unchanged, adjusting the middle percentile; an example of a language variable is as follows: Figure 2 As shown.

[0029] S2: For each sample and its tags mapped to Each fuzzy class is obtained from the fuzzy class of each fuzzy set. membership degree , and make each fuzzy class membership degree Combined to obtain a sample Membership vector = ; k is the sample number; if the sample If the sample is from the source domain, then the label is the true label; if the sample... If the sample is from the target domain, then the label is a pseudo-label.

[0030] Specifically, the language variables obtained through S1 This can blur continuous labels into discrete categories. Taking language variables as an example, we define three fuzzy classes corresponding to... , , Three fuzzy sets, for a label of a subject in either the source or target domain. We can obtain the vectors corresponding to the three fuzzy classes: Each element represents The membership degree belongs to the corresponding fuzzy class. After obtaining the membership vector, a continuous value label can be converted into a discrete fuzzy class. Unlike traditional classification categories, here a label can belong to multiple fuzzy classes simultaneously to varying degrees. This applies to each supervised EEG sample from the source domain subjects. All are converted to membership degree form. For unsupervised samples of the target subjects, membership vectors can be generated through iteration or pseudo-labels generated by neural networks.

[0031] S3: Utilize Calculate the fuzzy class-conditional probability distance from the source domain to the target domain. ; Identify the source domain. Identify the target domain. The number of samples in the source domain. The number of samples in the target domain. For the i-th source domain sample Membership degree of the corresponding c-th fuzzy class The normalized value; For the j-th target domain sample Membership degree of the corresponding c-th fuzzy class The normalized value, Let be the cardinality of the set.

[0032] Specifically, using the EEG samples obtained from S2 We can define the fuzzy class conditional probability distance between domains. Here, we start from the existing classification transfer learning algorithm's class conditional probability distance formula and demonstrate the construction process of the regression algorithm. Assume that in a... In the classification problem of brain-computer interfaces, and The first in The subdomains composed of classes are and The class-conditional probability distance between domains defined by a certain classification transfer learning algorithm is: ; This represents the distance metric based on discrete category information in this classification transfer learning method. For the cardinality of the set, In the source domain Subdomains composed of class samples. After fuzzing, fixed-length... Each dimension of the vector represents a fuzzy class, corresponding to the discrete categories in the above formula. Since each EEG sample belongs to a virtual fuzzy class to varying degrees, membership weights need to be added for each EEG sample when calculating the class-conditional probability. The fuzzy class distance metric in regression is: ; The membership degree of the target domain is obtained by fuzzing the pseudo-labels generated by the regressor. Thus, the category-based distance metric in regression can be calculated, and this classification transfer learning method can be reconstructed for use in regression problems.

[0033] S4: The class-conditional probability distance from the source domain to the target domain. The minimum is the target or basis, and the initial regression model for the regression of vital signs signals is iteratively updated. During the iterative update process, the iterated initial regression model generates pseudo-labels for the target domain samples in the next iteration and returns to S2. The vital signs signal regression model is considered to be obtained when the iterated initial regression model meets the preset conditions.

[0034] Specifically, obtained by reducing S3 This allows for the alignment of class-conditional probabilities between source and target domain subjects. This can be achieved by calculating using different methods. Optimization functions or loss terms can be designed. This helps align EEG samples. After data alignment, the distributional differences among different subjects decrease, and the regressor trained on the source domain subjects can be directly applied to the target subjects. In traditional transfer learning algorithms, a regressor needs to be built on the aligned source domain samples to generate target domain pseudo-labels for the next iteration, and the alignment is then achieved step-by-step through steps S2-S4 in the next iteration. For example, in deep learning algorithms, the target domain pseudo-labels output by the neural network are used to calculate the membership vector to achieve alignment.

[0035] As an optional implementation, the class-conditional probability distance from the source domain to the target domain is used. With the goal of minimizing, the initial regression model used for vital sign signal regression is iteratively updated, including: setting the optimization function as... To update and iterate the initial regression model; ; ; in, The joint probability distance matrix within the source and target domains. Joint probability distance matrix between source and target domain classes. For the source domain The sample belongs to the first Membership degree of a fuzzy class For the target domain The sample belongs to the first Membership degree of a fuzzy class For a certain feature transformation (which varies depending on the algorithm). Let be the norm squared over a regenerated Hilbert space.

[0036] For example, the Joint Probability Distribution Adaptation (JPDA) algorithm reconstructs probabilities. JPDA enhances both domain transferability and class discriminativeness by reducing the discriminative joint probability maximum mean discrepancy (DJP-MMD) between the source and target domains. This assumes the existence of... The categories, DJP-MMD, are defined as follows: ; For the hyperparameters that balance transferability and discriminability, where: ; ; For a regenerated Hilbert space, Let be the transformation matrix to be learned. Because it uses discrete category information, the original JPDA algorithm cannot be directly applied to regression problems.

[0037] In this scheme, JPDA can be reconstructed: for a given source domain label and target domain pseudo tags (Iterative generation), first in Establish on A fuzzy class is obtained, and the source domain membership vector is obtained after fuzzification. and target domain After generalization algorithm, in regression... and It can be rewritten as: ; ; Therefore, DJP-MMD can be calculated in regression problems, and JPDA can be successfully applied to regression.

[0038] As an optional implementation, the class-conditional probability distance from the source domain to the target domain is used. With the goal of minimizing, the initial regression model used for vital sign signal regression is iteratively updated, including: setting the optimization function as... To update and iterate the initial regression model; ; ; for The selected value; The joint probability distance matrix of the fuzzy class conditional probabilities of the source and target domains. The fuzzy class conditional probability distance matrix between source and target domain classes. To use the formula The optimal value found; It is a hyperparameter used to limit the upper limit of the traversal range; ; for The second largest value in, It is the i-th source domain tag.

[0039] It should be noted that, in addition to the existing reconstruction of classification transfer learning, a new transfer algorithm can be constructed based on the method of this invention: Fuzzy Set Selection and SubdomainDistribution Adapation for Regression (FSS-SDAR). The algorithm details are as follows: In the generalization of the JPDAR algorithm, The parameters are selected manually. FSS-SDAR uses the concept of fuzzy class confusion, enabling the algorithm to automatically select the number of fuzzy sets with the minimum class confusion. Fuzzy set selection mainly includes the following steps: First, regarding interval partitioning, unlike previous percentage-based methods of defining fuzzy sets, FSS-SDAR uses the length of the interval to define the fuzzy set. FSS-SDAR divides the label range into several intervals and uses the endpoints of these intervals to define the fuzzy set. Furthermore, when the label has theoretical upper and lower limits (e.g., in brain-computer interface problems, fatigue levels are between [0,1]), FSS uses the theoretical interval of the label to define the range of the fuzzy set. Now, , and These are set to 5%, 50%, and 95% of the interval length, instead of the percentile values ​​of the labels. Language variables can be defined using these interval endpoints. Since the fuzzy sets are defined based on the theoretical range of the labels, this method can capture information outside the source domain distribution, thereby improving the model's generalization ability. Furthermore, the number of fuzzy class samples can be inconsistent under non-percentage settings, effectively mitigating the problem of the sample size gradually decreasing in each fuzzy set as the number of samples increases.

[0040] Then, using S1, we obtain... FSS-SDAR selects an appropriate number of fuzzy sets To minimize fuzzy category confusion and reduce the probability of a sample falling between two fuzzy sets, FSS-SDAR uses... Measure to select For sample labels and specific Let its membership vector be .remember for The second largest value in the range. Then... ,in Indicates in Under this setting, the difference between the largest and second largest membership values ​​of the label. It can be used to measure the distance between a sample and the peak of a fuzzy category. The larger the value, the closer the label is to the peak of the fuzzy category corresponding to the maximum membership degree, indicating a lower degree of confusion between fuzzy categories. Before alignment, FSS-SDAR... Perform the traversal: ,in These are hyperparameters used to limit the upper limit of the traversal range. FSS-SDAR calculates the samples in each source domain at different... Select Average Then select the one that makes the average The largest ,Right now ; Therefore, the selected This increases the maximum membership degree of most samples, thus mitigating the impact of class confusion on transfer learning performance. It is worth noting that... It is calculated based on a single source domain. When multiple source domains exist, the fuzzy category may change depending on the source domain.

[0041] Furthermore, unlike JPDAR, FSS-SDAR does not explicitly align the joint probability distribution and label distribution of the two domains, but only aligns the marginal probability distribution (conditional probability distribution of fuzzy classes) of each fuzzy class in the two domains, thereby mitigating the negative transfer caused by large differences in label distribution. Using S2... It can define the distance between the source and target domains in FSS-SDAR, and its form is similar to DJP-MMD, but... and Modified to: ; ; Substituting the sample, the empirical estimate is: ; .

[0042] Ultimately, similar to JPDAR, it can be achieved by minimizing... The optimization objective of FSS-SDAR is obtained.

[0043] As an optional implementation, the class-conditional probability distance from the source domain to the target domain is used. With the goal of minimizing the initial regression model for vital sign signal regression, iterative updates are performed, including setting the loss function as follows: This is used to update and iterate the initial regression model based on the neural network; whereby, express The output of the layer, The membership matrix represents the degree of the source domain labels. The membership matrix represents the pseudo-labels of the target domain. express The feature space of the product of order tensors.

[0044] Among them, such as Figure 3 As shown, the Joint Adaptation Network (JAN) approximates the alignment of the joint probability distribution between the source and target domains by aligning the outputs of different network layers. JAN assumes that the label differences between the source and target domains mainly exist in the neural network layers closer to the output, while the feature differences mainly exist in the neural network layers closer to the input. It also assumes that the network has a total of... Layer, number The output of the layer is ,but This can serve as an effective alternative to the joint probability distribution. Based on this assumption, JAN defines the Joint Maximum Mean Discrepancy (JMMD), and transfer is achieved by minimizing JMMD. In practical applications of JAN, one-hot vectors of class labels are needed. However, in regression problems, labels are continuous, making JAN unsuitable for direct application. By establishing fuzzy sets and fuzzification, JMMD can be modeled in regression problems as follows: .

[0045] As an optional implementation, the class-conditional probability distance from the source domain to the target domain is used. With the goal of minimizing the initial regression model for vital sign signal regression, iterative updates are performed, including setting the loss function as follows: This is used to update and iterate the initial regression model based on the neural network; whereby, This is a neural network model.

[0046] Among them, such as Figure 4 As shown, Deep Subdomain Adaptation Networks (DSANs) achieve local alignment, rather than global alignment, by dividing the domain into several subdomains and computing losses on those subdomains. DSANs defines Local Maximum Mean Discrepancy (LMMD) to measure the discrepancy between subdomains. When subdomains are divided by class, minimizing LMMD can be seen as alignment with class-conditional probabilities. Because class-conditional probabilities are difficult to compute in regression problems, DSANs cannot be directly used in regression problems. Definition As a mapping to a regenerated Hilbert space, after fuzzification, LMMD can be defined in regression as: ; ; For a having Layer 1 network, denoted as the first The deep network activations generated by the layers are respectively the source domain and the target domain. and Subsequently, the first Layered LMMD can be restated as: ; ; Therefore, DSAN can be applied to regression problems.

[0047] Example 2 This embodiment provides a method for regressing vital sign signals, such as... Figure 5As shown, the process includes: collecting the current user's vital signs signals, preprocessing them, and inputting them into the vital signs signal regression model to obtain the regression results corresponding to the current user's vital signs signals.

[0048] Example 3 This embodiment provides a vital signs signal processing system, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of a method for constructing a vital signs signal regression model or a method for regressing vital signs signals.

[0049] Example 4 This embodiment provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of a method for constructing a vital sign signal regression model or a method for regressing vital sign signals.

[0050] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for constructing a vital sign signal regression model based on fuzzy sets, characterized in that, include: S1: Determine the target domain samples of vital signs signals, the source domain samples of vital signs signals, and their corresponding source domain labels for regression types; Concatenate all the source domain tags to obtain a tag domain, and then concatenate the tags in the tag domain. Source domain tags Divide into multiple boundary points, and use the boundary point definition. A fuzzy set; the vital signs signals are: electromyography signals, electrooculography signals, electrocardiogram signals, functional near-infrared spectroscopy signals, or electroencephalogram signals; S2: For each sample and its tags mapped to Each fuzzy class is obtained from the fuzzy class of each fuzzy set. membership degree , and make each fuzzy class membership degree Combined to obtain the sample Membership vector = ; k is the sequence number of the sample; if the sample If the sample is from the source domain, then the label is the true label; if the sample... If the sample is from the target domain, then the label is a pseudo-label; S3: Using the formula Calculate the fuzzy class-conditional probability distance from the source domain to the target domain. ; Identify the source domain. Identify the target domain. The number of samples in the source domain. The number of samples in the target domain. For the i-th source domain sample Membership degree of the corresponding c-th fuzzy class The normalized value; For the j-th target domain sample Membership degree of the corresponding c-th fuzzy class The normalized value, It is represented by distance; S4: The fuzzy class conditional probability distance from the source domain to the target domain. Minimize the design of the optimization function or loss function, and iteratively update the initial regression model used for vital sign signal regression. During the iterative update process, the pseudo-labels of the target domain samples in the next iteration are generated using the iterated initial regression model, and S2 is returned; until the iterated initial regression model reaches the preset conditions, it is considered that the vital sign signal regression model has been obtained. The tag field Source domain tags Multiple dividing points are established, including dividing points based on the percentage of the source domain label value; Alternatively, based on the percentage of the interval length corresponding to the upper and lower bounds of the label in the vital signs signal problem.

2. The method for constructing a vital sign signal regression model based on fuzzy sets as described in claim 1, characterized in that, The fuzzy class conditional probability distance from the source domain to the target domain is... The optimization function is designed with the goal of minimizing the input, and the initial regression model for vital sign signal regression is iteratively updated, including: utilizing the fuzzy class-conditional probability distance from the source domain to the target domain. The fuzzy joint probability distance is obtained from the category edge probability, and the optimization function is set as the optimization objective. To update and iterate the initial regression model; ； ； in, This is the joint probability distance matrix within the source and target domains. This is a weighting factor that balances the degree of intra-class distance and inter-class distance. This is the joint probability distance matrix between classes in the source and target domains. For the target domain Sample Belongs to the Membership degree of a fuzzy class For feature transformation, Let be the norm squared over a regenerated Hilbert space.

3. The method for constructing a vital sign signal regression model based on fuzzy sets as described in claim 1, characterized in that, The fuzzy class conditional probability distance from the source domain to the target domain is... The optimization function is designed with the goal of minimizing the initial regression model for vital sign signal regression, and the model is iteratively updated, including: using the fuzzy class-conditional probability distance from the source domain to the target domain. With the goal of minimizing, the optimization function is set as follows: To update and iterate the initial regression model; ； ； in, for The selected value within the range of values; This is a weighting factor that balances the degree of intra-class distance and inter-class distance. Let the joint probability distance matrix be the conditional probability of the source domain and the target domain. This represents the fuzzy class conditional probability distance matrix between the source and target domain classes. For the target domain Sample Belongs to the Membership degree of a fuzzy class For feature transformation, Let be the norm squared over a regenerated Hilbert space.

4. The method for constructing a vital sign signal regression model based on fuzzy sets as described in claim 3, characterized in that, To utilize The optimal value found; It is a hyperparameter used to limit the upper limit of the traversal range; ; for The second largest value in, It is the i-th source domain tag.

5. The method for constructing a vital sign signal regression model based on fuzzy sets as described in claim 1, characterized in that, The class-conditional probability distance from the source domain to the target domain Based on this, construct the joint maximum mean distance loss of the regression. With the goal of minimizing this loss, the initial regression model used for vital sign signal regression is iteratively updated, including: setting... The initial regression model based on the neural network is then updated and iterated; wherein, express The output of the layer, The membership matrix represents the degree of the source domain labels. The membership matrix represents the pseudo-labels of the target domain. Represents integers The feature space of the product of order tensors.

6. The method for constructing a vital sign signal regression model based on fuzzy sets as described in claim 1, characterized in that, The fuzzy class conditional probability distance from the source domain to the target domain is... The optimization function is designed with the goal of minimizing the initial regression model for vital sign signal regression, and the model is iteratively updated, including: using the class-conditional probability distance from the source domain to the target domain. Based on minimum principles, a fuzzy sub-neighborhood distance loss is constructed as the loss function. , This is to update and iterate the initial regression model based on the neural network; wherein, For neural network models, For source domain samples, For the target domain sample, Let be the norm squared over a regenerated Hilbert space.

7. A regression method for vital sign signals, comprising: Collect the vital signs signal of the current user, preprocess it, and input it into the vital signs signal regression model constructed by the method of constructing the vital signs signal regression model based on fuzzy set as described in any one of claims 1-6, to obtain the regression result corresponding to the vital signs signal of the current user.

8. A vital signs signal processing system, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method of claim 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for constructing a vital sign signal regression model based on fuzzy sets as described in any one of claims 1 to 6, or the method for regressing vital sign signals as described in claim 7.

Images