A data processing method and system based on a monitoring bracelet

Abstract

The application discloses a kind of based on the data processing method and system of monitoring bracelet, comprising: through the chip module built-in in monitoring bracelet, the physiological monitoring data of monitoring object is acquired;The abnormal interval of the physiological monitoring data is identified and classified;According to all the classification of abnormal interval, and the feature of abnormal interval under each classification, the adaptive feature of the monitoring object is constructed;Through the adaptive feature of the monitoring object, parameter mapping is carried out, and the parameter of data supplement model is generated;After data cleaning is carried out to each abnormal interval, data supplement model is used after parameter assembly, and data supplement is carried out.Can reflect the feature difference of individual under different abnormal state, realize individualized data analysis;More reasonable supplementary data is generated for different abnormal characteristics;Ensure that the generated data is consistent with normal data in numerical characteristics and dynamic trend, effectively improve data continuity and physiological rationality.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of monitoring data processing, in particular to a data processing method and system based on a monitoring bracelet. BACKGROUND

[0002] With the development of wearable device technology, monitoring bracelets, as a convenient physiological data collection tool, are widely used in health management, sports tracking, and medical assistance. Existing monitoring bracelets usually obtain physiological signals such as heart rate, blood oxygen saturation, and skin temperature of the wearer in real time through built-in chip modules. However, in actual application, due to poor contact between the bracelet and the skin, loose wearing, device falling off, or external environmental interference, there are distortion, missing, or abnormal fluctuation phenomena in the collected physiological monitoring data. These abnormalities not only affect the accuracy of subsequent health assessment, but also may mask potential health risks, reducing data credibility and application effect.

[0003] Most of the existing technologies set fixed thresholds or use simple filtering algorithms to exclude or roughly compensate abnormal data, lack in-depth analysis and classification of abnormal mechanisms, and are difficult to adaptively optimize data compensation strategies according to different abnormal characteristics. In addition, traditional data compensation methods are usually based on interpolation or regression estimation, which is difficult to ensure the consistency of the supplemented data and the original waveform in dynamic characteristics and physiological rationality.

[0004] In view of the above problems, there is an urgent need for a processing method that can accurately identify abnormal intervals in physiological data, classify abnormal types, dynamically generate compensation parameters according to individual characteristics, and realize high-quality data compensation through intelligent models. SUMMARY

[0005] In view of the above problems, the present application is proposed.

[0006] To solve the above technical problems, the present application provides the following technical scheme: a data processing method based on a monitoring bracelet, comprising:

[0007] acquiring physiological monitoring data of a monitoring object through a chip module built in the monitoring bracelet;

[0008] identifying abnormal intervals and classifying abnormal intervals of the physiological monitoring data;

[0009] constructing adaptive characteristics of the monitoring object according to all abnormal interval classifications and the characteristics of the abnormal intervals under each classification;

[0010] mapping parameters through the adaptive characteristics of the monitoring object to generate parameters of a data compensation model;

[0011] After clearing the data for each abnormal interval, the data is supplemented using the parameter-assembled data supplementation model.

[0012] As a preferred embodiment of the data processing method based on the monitoring wristband described in this invention, the physiological monitoring data includes, but is not limited to, heart rate, blood oxygen saturation, and skin temperature.

[0013] Different physiological monitoring data are processed separately.

[0014] As a preferred embodiment of the data processing method based on the monitoring wristband described in this invention, the abnormal interval includes the abnormal part of the waveform after the physiological monitoring data is plotted as a continuous time series.

[0015] The waveform data is input into the trained neural network, and through the identification of abnormal intervals, the start timestamp, end timestamp, interval classification result, and interval waveform features of each abnormal interval are output.

[0016] The interval classification results include normal intervals and abnormal intervals; the abnormal intervals include, but are not limited to, poor contact of the wristband, wristband falling off, and data distortion.

[0017] The interval waveform characteristics include the average value, variance, maximum value, minimum value, peak rate, and baseline drift rate of the data within the interval.

[0018] As a preferred embodiment of the data processing method based on the monitoring wristband described in this invention, wherein: within each category of abnormal intervals, the seven interval waveform features are respectively used as seven dimensions;

[0019] Each abnormal interval of the waveform is located in a 7-dimensional data space, resulting in a location set for all abnormal intervals: H = {h} J h L h S}; where h J h represents the set of coordinates that classify the abnormal interval as a wristband contact malfunction. L The abnormal interval is classified as the set of coordinates of the wristband falling off, h S The set of coordinates representing outlier intervals that are classified as data distortion;

[0020] In the coordinate set of each anomaly interval classification, the coordinates of the midpoint of the set are integrated. The mean of the discrete coordinates in the seven dimensions is calculated to obtain the mean of the seven dimensions, which is used as the coordinates of the midpoint of the set; corresponding to the preset standard attention matrix;

[0021] The adaptive features include the coordinates of the midpoint of the set under each anomaly interval classification; represented as: H = {h J0 hL0 h S0}; where h J0 h represents the coordinates of the midpoint of the set of abnormal intervals classified as wristband contact malfunctions. L0 The abnormal interval is classified as the coordinate of the midpoint of the set of wristband detachments, h. S0 This represents the coordinates of the midpoint of the set of abnormal intervals classified as data distortion.

[0022] As a preferred embodiment of the data processing method based on the monitoring wristband described in this invention, the parameter mapping includes: in the seven-dimensional space under each abnormal interval classification, a preset standard coordinate point h0 is used to represent the preset standard value under each feature;

[0023] Let h x0 h J0 h L0 h S0 Any coordinate in the middle;

[0024] h x0 Compare the coordinates of h0 with h0 in seven dimensions, and calculate h in each dimension using the coordinates of h0 as the reference. x0 The percentage increase or decrease of the coordinates;

[0025] In the standard attention matrix, the parameters of the attention matrix are adjusted according to the increase / decrease ratio and the mapping function to obtain h. x0 The corresponding attention matrix.

[0026] As a preferred embodiment of the data processing method based on the monitoring wristband described in this invention, the adjustment of the parameters of the attention matrix includes fitting the increase or decrease ratio of the coordinate data of h0 in each dimension with the change of the attention matrix parameters through the training set to obtain the mapping function;

[0027] After adjusting the parameters of the attention matrix in each dimension, the adjusted parameters of the attention matrix are normalized to obtain the final attention matrix.

[0028] The data supplementation model is an adversarial network.

[0029] As a preferred embodiment of the data processing method based on the monitoring wristband described in this invention, the data supplementation includes supplementing data for each abnormal interval using a trained adversarial network.

[0030] The input to the adversarial network is: the complete waveform, the start and end points of the abnormal interval, the adjusted attention matrix, and the classification of the abnormal interval;

[0031] Output layer: Generates a supplementary waveform sequence with the same length as the original abnormal segment;

[0032] The head of the generator of the adversarial network is a classifier, which is responsible for classifying the abnormal interval into corresponding levels, and each level is connected with an independent data channel;

[0033] For the waveform in the normal interval, the fixed time window length is used for segmentation, and the 7-dimensional selection space coordinates are constructed for the data in each window;

[0034] For the abnormal interval p, the time window in the previous normal interval is q, and the time window in the next normal interval is h;

[0035] Connect p and h in the 7-dimensional selection space, and take the midpoint of the two as the coordinate position of p in the 7-dimensional selection space; calculate the relative distance of p to each center distance divided by the space radius of the cluster center represented by the cluster; pg represents the distance of p to the cluster center g, and the g corresponding to the minimum value of pg is taken as the cluster center of p; wherein each cluster center represents a level, and in each channel adaptation module, adaptation is performed in the corresponding cluster;

[0036] The adaptation process of the adaptation module includes: presetting a standard coordinate h 0+ different from h0

[0037] In the cluster family corresponding to the adaptation module, A normal intervals are randomly selected; the difference between the 7-dimensional features of each interval and h 0+ is calculated respectively, so that each adaptation module adapts to the data of the corresponding level, and when the adaptation is completed, the physiological monitoring data is supplemented.

[0038] A data processing system based on a monitoring bracelet, wherein: a collection unit acquires physiological monitoring data of a monitoring object through a chip module built in the monitoring bracelet;

[0039] A classification unit identifies abnormal intervals and classifies abnormal intervals for the physiological monitoring data;

[0040] An analysis unit constructs adaptive features of the monitoring object according to all abnormal interval classifications and the features of the abnormal intervals under each classification;

[0041] An adjustment unit generates parameters of a data supplement model through the adaptive features of the monitoring object;

[0042] A supplement unit uses the data supplement model after parameter assembly to supplement data after data cleaning for each abnormal interval.

[0043] A computer device comprises a memory and a processor; the memory stores a computer program, wherein the processor implements the steps of the method of any one of the present application when executing the computer program.

[0044] A computer readable storage medium stores a computer program, wherein the computer program implements the steps of the method of any one of the present application when executed by a processor.

[0045] The present application has the following beneficial effects: The data processing method based on a monitoring bracelet provided by the present application can accurately identify abnormal intervals in physiological monitoring data on complete waveform data, and perform fine-grained classification according to abnormal types, thereby improving the accuracy and reliability of abnormal detection. By extracting multi-dimensional waveform features of each abnormal interval, an adaptive feature space of a monitoring object is constructed, which can reflect the feature differences of individuals in different abnormal states, and realize personalized data analysis. By comparing the adaptive features with standard features, combining the increase / decrease ratio and the mapping function, and dynamically adjusting the attention matrix, the data supplement model is guided to generate more reasonable supplementary data for different abnormal characteristics. The data is supplemented by using an adversarial network, so as to ensure that the generated data is consistent with normal data in numerical characteristics and dynamic trends, thereby effectively improving the data continuity and physiological rationality. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor.

[0047] Figure 1 A data processing method based on a monitoring bracelet is provided for the first embodiment of the present application. DETAILED DESCRIPTION

[0048] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the protection scope of the present application.

[0049] Embodiment 1, refer to Figure 1 For an embodiment of the present application, a data processing method based on a monitoring bracelet is provided, comprising:

[0050] S1: Obtain physiological monitoring data of the monitoring object by monitoring the chip module built in the bracelet.

[0051] The chip module is integrated with at least one physiological signal sensor, which can specifically include a heart rate sensor (such as a photoplethysmography PPG sensor), an oxygen saturation sensor (such as a dual-wavelength near-infrared sensor), and a skin temperature sensor (such as a thermistor or thermocouple sensor).

[0052] The chip module outputs the above-mentioned various physiological signals in the form of digital data stream at a set sampling frequency (for example, once per second or higher frequency).

[0053] Different physiological monitoring data are pre-processed according to their characteristics, including but not limited to: smoothing filtering of heart rate signals to remove high-frequency noise; artifact removal of oxygen saturation signals to enhance numerical stability; baseline correction of skin temperature signals to eliminate external environmental influences.

[0054] The pre-processed physiological data are standardized in a unified format to form a continuous time sequence as the basis data for subsequent anomaly recognition and supplementary processing.

[0055] S2: Identify and classify the abnormal interval of the physiological monitoring data.

[0056] The abnormal interval includes the abnormal part in the waveform after the physiological monitoring data are plotted into a continuous time sequence. The waveform data are input into the trained neural network, and the start timestamp, end timestamp, interval classification result, and interval waveform feature of each abnormal interval are output through the identification of the abnormal interval.

[0057] First, by plotting the physiological monitoring data into a continuous time sequence, the dynamic change characteristics of the data can be fully retained, and the continuity, stability, and mutation of the physiological signals over time can be intuitively reflected, avoiding local misjudgment that may be caused by single-point sampling. Second, using the trained neural network to analyze the complete waveform can accurately identify the abnormal interval in the signal based on the overall pattern of the time sequence, which has higher adaptability and recognition accuracy than traditional rule detection methods (such as threshold method and sliding mean method). Through the classification of the abnormal interval, the abnormal types can be subdivided into different situations such as poor bracelet contact, bracelet falling off, and data distortion, so that the subsequent data supplement and correction can be targeted to develop differentiated processing strategies according to the abnormal reasons, thereby improving the rationality and effectiveness of data supplement.

[0058] In addition, by extracting waveform features including mean value, variance, maximum value, minimum value, peak rate and baseline drift rate in each abnormal interval, a high-dimensional feature expression can be constructed for each abnormal section, which not only describes the overall statistical characteristics of the abnormal section, but also provides a quantitative basis for subsequent adaptive feature modeling, parameter mapping and supplementary model guidance.

[0059] In the embodiment, the neural network is a deep learning model for time series data anomaly detection, and the main structure is: an input layer receives complete physiological monitoring data waveform sequences; a feature extraction layer extracts local waveform features through a one-dimensional convolutional neural network (1D-CNN) to capture the change trend, mutation point and local stability in the time series; a time series modeling layer further learns the long-time dependence relationship of the sequence data using a recurrent neural network (such as LSTM, Long Short-Term Memory); a classification layer maps each time point or time period to normal or abnormal, and further subdivides into specific abnormal types such as poor hand ring contact, hand ring falling off, and data distortion; and an output layer outputs the start timestamp, end timestamp, interval classification result and interval waveform feature of each abnormal interval.

[0060] Through the above structure, the recognition accuracy can be ensured while considering the running efficiency of the model in the resource-limited environment such as the hand ring. In other implementable embodiments, the neural network can also be replaced by the following structures, including but not limited to: Transformer structure, time series convolution network and graph convolution network (GCN) combined with time series modeling, etc.

[0061] S3: According to all abnormal interval classifications and the features of the abnormal intervals in each classification, an adaptive feature of the monitoring object is constructed.

[0062] In each category of abnormal interval, seven interval waveform features are taken as seven dimensions respectively.

[0063] Each abnormal interval of the waveform is positioned in a 7-dimensional data space to obtain a positioning set of all abnormal intervals: H = {h J , h L , h S}; wherein h J represents the coordinate set of the abnormal interval classification as poor hand ring contact, h L represents the coordinate set of the abnormal interval classification as hand ring falling off, and h S represents the coordinate set of the abnormal interval classification as data distortion.

[0064] In the coordinate set of each abnormal interval classification, the coordinates of the set midpoint are integrated, the mean values of the discrete coordinates of the seven dimensions are calculated respectively, and the mean values of the seven dimensions are obtained (if a coordinate is 0 in the seven dimensions, the coordinate point is removed) as the coordinates of the set midpoint.

[0065] The adaptive features include the coordinates of the set midpoint under each abnormal interval classification, denoted as: H = {h J0 , h L0 , h S0}; wherein h J0 represents the coordinates of the set midpoint of the abnormal interval classification of poor contact of the hand ring, h L0 represents the coordinates of the set midpoint of the abnormal interval classification of the hand ring falling off, and h S0 represents the coordinates of the set midpoint of the abnormal interval classification of data distortion.

[0066] Based on the waveform features of the abnormal interval under different abnormal classifications, adaptive features that can truly reflect the individual differences and abnormal performance rules of the monitoring object are constructed, thereby providing reliable support for subsequent data supplement and parameter mapping. By positioning each abnormal interval in a seven-dimensional feature space and integrating effective feature points under each classification, the feature centers of different abnormal types can be accurately extracted. In particular, during the mean value calculation process, all data points with a feature value of zero in all feature dimensions are removed to avoid invalid data caused by hand ring falling off, signal loss, etc. from interfering with the feature center, thereby improving the authenticity and robustness of feature modeling. In this way, individualized feature expression under abnormal classification can be achieved, so that the parameter mapping and the subsequent supplement model can be dynamically adjusted according to the actual characteristics, significantly improving the intelligence, adaptability and processing accuracy of the overall data processing system.

[0067] Since different individuals have natural differences in waveform feature performance when facing the same type of abnormality (such as poor contact, falling off, or data distortion), the embodiment dynamically constructs the adaptive feature center under abnormal classification by integrating the mean values of the effective abnormal interval features, which can fully reflect the specific performance differences of different abnormal types in different populations. This design not only enhances the adaptability of the data processing method to individual differences, but also provides a more accurate and personalized feature basis for subsequent parameter mapping and data supplement process, thereby improving the robustness and effectiveness of the overall system in a diversified application environment.

[0068] S4: Perform parameter mapping based on the adaptive features of the monitoring object to generate parameters of the data supplement model.

[0069] In the seven-dimensional space under each abnormal interval classification, a preset standard coordinate point h0 is used to represent a preset standard value under each feature; and a corresponding preset standard attention matrix.

[0070] Let h x0 h J0 h L0 h S0 Any coordinate in the array.

[0071] h x0 Compare the coordinates of h0 with h0 in seven dimensions, and calculate h in each dimension using the coordinates of h0 as the reference. x0 The percentage increase or decrease of the coordinates.

[0072] In the standard attention matrix, the parameters of the attention matrix are adjusted according to the increase / decrease ratio and the mapping function to obtain h. x0 The corresponding attention matrix (used to focus on the missing parts).

[0073] By setting preset standard coordinate points in the seven-dimensional feature space under each anomaly category and comparing the actually constructed adaptive features with the standard coordinates, the differences in each feature dimension are quantified, thereby dynamically adjusting the standard attention matrix to achieve precise attention and compensation optimization for missing data regions. By calculating the increase or decrease ratio of actual features relative to standard features and adjusting the attention parameters in combination with the mapping function, the attention intensity of each feature dimension during the supplementation process can be flexibly adjusted according to the anomaly type and individual differences. This design enables the supplementation model to dynamically select the key feature directions to focus on when facing different anomaly types and different individual performances, avoiding a one-size-fits-all supplementation strategy, improving the targeting, rationality, and physiological consistency of data supplementation, and thus significantly enhancing the overall system's adaptability, accuracy, and data integrity.

[0074] Furthermore, using the training set, the mapping function is obtained by fitting the changes in the attention matrix parameters to the ratio of increases and decreases in the coordinate data of h0 in each dimension. After adjusting the parameters of the attention matrix in each dimension, the adjusted parameters are normalized to obtain the final attention matrix.

[0075] In this embodiment, first, based on the training set, for each seven-dimensional feature space under the classification of each abnormal interval, a large amount of historical data of monitoring objects is collected, and the increase / decrease ratio between the adaptive feature and the standard feature of each sample in each feature dimension is calculated. At the same time, the effect change of each sample after adjusting the attention matrix parameter in the data supplement process is recorded. Through statistical analysis and supervised learning, in each feature dimension, the increase / decrease ratio is taken as the input, and the attention parameter change under the actual optimal compensation effect is taken as the output, fitting modeling is performed, and the mapping function of each feature dimension is obtained. The mapping function can be a linear function, a segmented function or a nonlinear function (such as a sigmoid function, a hyperbolic tangent function), which is determined according to the actual fitting accuracy. Subsequently, in actual application, for each new abnormal interval, the increase / decrease ratio obtained by comparing the adaptive feature with the standard feature will be input into the corresponding mapping function, and the adjusted attention matrix parameter will be output. In order to ensure that the attention matrix meets the requirements of weighted sum consistency, after all the dimension parameters are adjusted, the adjusted matrix is further normalized by row, so that the sum of the weights of each row is 1, thereby obtaining the final attention matrix that can be used for data supplement.

[0076] By fitting the mapping relationship between the increase / decrease ratio and the attention matrix parameter change in the training set, the supplement effect deviation problem caused by direct proportional linear scaling is avoided, so that the attention adjustment can more accurately reflect the importance of each feature dimension in data supplement under abnormal state. By using the mapping function obtained by fitting, the increase / decrease amplitude of attention can be flexibly controlled according to the change law of different feature dimensions, and the consistency of the supplemented data in dynamic trend and physiological rationality is improved. Through normalization processing, the numerical specification of the attention matrix is ensured, the attention weight distortion caused by adjustment is avoided, and the stability and robustness of the model in abnormal section compensation are further enhanced. Overall, this design enables the supplement model to achieve more accurate and personalized missing data repair, and improves the data processing capability of the system under complex abnormal conditions.

[0077] S5: After data cleaning for each abnormal interval, the data supplement model after parameter assembly is used for data supplement.

[0078] The data supplement model is an adversarial network. The trained adversarial network is used to supplement data for each abnormal interval respectively. The input of the adversarial network is: complete waveform, start and end point of abnormal interval, adjusted attention matrix, classification of abnormal interval.

[0079] Output layer: generate a supplemented waveform sequence consistent with the length of the original abnormal section.

[0080] For the detected abnormal interval, the intelligent supplement of missing or distorted physiological monitoring data is realized through the trained adversarial network model, and the integrity and continuity of the data are improved. By taking the complete waveform data, the start and end positions of the abnormal interval, the adjusted attention matrix, and the classification information of the abnormal interval as inputs, the context dynamic features before and after the abnormal occurrence, the specific position range of the abnormal section, and the feature focus corresponding to the abnormal type can be provided for the supplement model, so that the generator can generate a supplementary waveform that conforms to the physiological logic and naturally connects with the original data dynamic trend based on the real waveform trend and abnormal characteristics. The output layer generates a supplementary waveform sequence with the same length as the original abnormal section, ensuring the continuity of the data on the time axis and avoiding data jumps or logical breaks caused by mismatched supplement length. Overall, by introducing the adversarial training mechanism, the present application can effectively improve the performance of the supplemented data in terms of numerical rationality, dynamic trend continuity, and physiological feature consistency, significantly enhancing the credibility of the monitoring data and the robustness of the system.

[0081] The above-mentioned adversarial network is a multi-channel network, and the head of the network is used to determine the level to which each abnormal interval is adapted. Each level is configured with an adaptation module (there are multiple adaptation modules, and the more adaptation modules there are, the stronger the adaptation ability and the more accurate the data supplement result), so that each abnormal interval is input into the corresponding channel.

[0082] Further, through the design of multiple adaptation modules, each waveform feature can be completed in a targeted manner. Through the hierarchical random interception and multi-channel adaptation mechanism proposed in the present application, the feature rules of the normal interval of the same level can be fully extracted before the data in the abnormal interval is supplemented, so that the supplemented data not only conforms to the overall state of the monitoring object in terms of numerical distribution, but also is highly consistent with the actual physiological characteristics in terms of local dynamic change trend. Overall, the present design effectively improves the naturalness, coherence and physiological rationality of the supplemented data, and significantly enhances the adaptability, stability and supplement accuracy of the system under diversified abnormal conditions.

[0083] Specifically, the head of the generator of the adversarial network is a classifier, which is responsible for classifying the abnormal interval into a corresponding level, and each level is connected with an independent data channel. The data in the abnormal interval will be supplemented according to the input channel. The discriminator is a general discriminator, and its structure is a one-dimensional convolutional neural network.

[0084] The waveforms in the normal interval are divided according to a fixed time window length, and a 7-dimensional data space coordinate (for distinction from the foregoing, referred to as "7-dimensional selection space") is constructed for the data in each window.

[0085] For the abnormal interval p, set the time window in the previous normal interval as q and the time window in the next normal interval as h.

[0086] Connect p and h in the 7-dimensional selection space, take the midpoint of the two as the coordinate position of p in the 7-dimensional selection space; calculate the relative distance of p to each center distance divided by the space radius of the cluster center represented by the cluster; pg represents the distance of p to the cluster center g, take the minimum value of pg corresponding to g as the cluster center of p; wherein each cluster center represents a level, and in each adaptive module, adaptation is performed in the corresponding cluster.

[0087] It is to be noted that the present scheme is actually completed by two attention strategies. Before the data of the abnormal interval is supplemented by using the trained adversarial network model, in each adaptive module, first, the feature information of the continuous normal interval is extracted from the collected physiological monitoring data. The normal interval refers to the waveform section determined as normal after abnormal identification classification. For each normal section, seven waveform features are extracted, including data mean, variance, maximum value, minimum value, peak rate and baseline drift rate, and the features of multiple normal sections are aggregated and analyzed to form the global physiological state feature description of the monitoring object. Subsequently, the global features extracted from the normal section are input into the adaptive module reserved inside the trained adversarial network to dynamically adjust the attention mechanism inside the generator. The adversarial network with the parameter configuration of attention is used for subsequent analysis.

[0088] Before the data of the abnormal section is formally supplemented, the feature information of the monitoring object in the normal state is fully utilized to dynamically adapt the attention mechanism inside the adversarial network (attention and adaptation of each adaptive module), thereby realizing personalized guidance of the supplement generation process. Due to the natural differences in physiological baseline states of different individuals and the fluctuations over time, direct data supplementation with fixed attention parameters is easy to cause the generated data to deviate from the actual state of the individual. By introducing the self-adaptive adjustment mechanism based on the features of the normal interval, the generator can more accurately capture the feature tendency of the current individual during the supplementation process, ensuring that the supplemented data is highly consistent with the individual normal state in terms of numerical trend, fluctuation form and physiological reasonableness, thereby improving the continuity, naturalness and credibility of the supplemented data, and enhancing the stability and adaptability of the overall system in practical application.

[0089] The adaptive process of the adaptive module includes: presetting a standard coordinate h 0+ In the adaptive module, A normal intervals are randomly selected from the corresponding cluster; the features of 7 dimensions in each interval are calculated respectively. 0+The difference between the adjusted feature and the standard feature of each interval is calculated, and the average value is obtained. According to the average value of the difference in each dimension, the adjustment factor is calculated.

[0090] Specifically, in each adaptation module, the adjustment method is similar to the method described above, and a plurality of intervals are randomly intercepted in the normal interval. Then a standard coordinate h 0+ is preset, which is different from h 0+ . The difference between the adjusted feature and the standard feature of each interval is calculated, and the average value is obtained. According to the average value of the difference in each dimension, the adjustment factor is calculated. (i) = 1 + a x tanh(b i x Af (i) ); where a > 0 represents the adjustment amplitude coefficient, which controls the maximum increase / decrease ratio; b i > 0 represents the adjustment sensitivity coefficient of the i-th feature, which controls the response speed of the offset; tanh represents the hyperbolic tangent function, which ensures the adjustment to be smooth and limited. Af (i) represents the average value of the difference in the i-th feature; r (i) represents the adjustment factor of the i-th feature. After inputting the adjustment factor of each dimension into the trained adversarial network model, the adversarial network model first needs to adapt to the wearer, and then perform the subsequent monitoring and data supplement. Each cluster center represents a level; each abnormal interval needs to be selected from the normal interval part in the same cluster family when randomly intercepting a plurality of intervals.

[0091] It should be noted that in this step, the adjustment factor calculation method used in the attention mechanism adjustment based on the normal interval feature is significantly different from the adjustment scheme for the abnormal interval. The attention adjustment of the abnormal interval is mainly based on the individualized offset between the abnormal feature and the standard feature to dynamically calculate the feature variation amplitude of each abnormal instance, emphasizing the accurate adaptation of individual abnormal performance differences to achieve individualized supplement and correction of missing data or abnormal data.

[0092] The adjustment of the normal interval is based on the common law of the physiological monitoring data in the normal state. By randomly intercepting multiple local segments in the normal interval, the features are extracted and compared with the standard features, the average difference in each feature dimension is calculated, and the smooth adjustment factor is calculated to control the change range of the attention weight in the normal physiological state. The purpose of this adjustment method is to strengthen the generator's understanding of the normal physiological feature distribution of the current wearer, maintain attention and adaptation to the main physiological law, rather than overemphasize individual deviations, so as to improve the rationality and continuity of the overall supplementary data within the normal fluctuation range.

[0093] Therefore, the attention adjustment of the present step belongs to an adaptive mechanism based on regularity guidance, which is different from the mechanism based on individualized difference compensation in the abnormal interval processing. The two are independent of each other and adapt to the characteristics of different stages in the data supplement task, respectively, to ensure the authenticity, continuity and physiological rationality of the supplementary data. That is, the first attention mechanism is to achieve user adaptation, so that the neural network focuses on the features without abnormalities, thereby achieving rapid adaptation. The second attention mechanism focuses on the "missing amount" or "excessive abnormal amount" of the abnormal part, thereby achieving data completion.

[0094] In order to further improve the accuracy and adaptability in the data supplement process of the abnormal interval, the present application increases the constraint mechanism based on grade division when randomly intercepting the normal interval for feature adaptation, to ensure that the randomly intercepted normal interval segment and the target abnormal interval are within the same grade, maintaining the consistency and characteristic continuity of the feature space. By dividing the complete normal interval into a fixed time window, a 7-dimensional selection space is constructed, and a density-based spatial clustering method (DBSCAN) is used for classification division, so that normal data with different stability features can be reasonably grouped.

[0095] On the other hand, the present embodiment also provides a data processing system based on a monitoring bracelet, which comprises:

[0096] The acquisition unit acquires physiological monitoring data of the monitoring object through the chip module built in the monitoring bracelet.

[0097] The classification unit identifies and classifies the abnormal interval of the physiological monitoring data.

[0098] The analysis unit constructs adaptive features of the monitoring object according to all abnormal interval classifications and the features of the abnormal interval under each classification.

[0099] The adjustment unit maps parameters through the adaptive features of the monitoring object to generate parameters of the data supplement model.

[0100] The supplement unit supplements data by using the parameter-assembled data supplement model after data cleaning for each abnormal interval.

[0101] If the above functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for making a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0102] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered a list of executable instructions for implementing logic functions, and can be specifically embodied in any computer-readable medium for use by an instruction execution system, apparatus, or device, such as a computer-based system, a system including a processor, or other system that can take instructions from an instruction execution system, apparatus, or device, or in conjunction with such instructions execution system, apparatus, or device. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

[0103] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection having one or more wires (electrical devices), a portable computer diskette (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium can even be paper or other suitable medium on which the program can be printed, as the program can be electronically obtained, for example, by optical scanning of the paper or other medium, followed by electronic editing, interpretation, or necessary processing, and then stored in a computer memory if necessary. Other suitable media can also be used.

[0104] It should be understood that portions of the present application can be implemented in hardware, software, firmware, or combinations thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and as in another embodiment, can be implemented using any or a combination of the following technologies, which are all well-known in the art: a discrete logic circuit having logic gates for implementing logic functions upon an application of data signals, an application specific integrated circuit having appropriate combinational logic gates, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0105] Example 2, which is an embodiment of the present application, provides a data processing method based on a monitoring bracelet. In order to verify the beneficial effects of the present application, economic benefit calculation and simulation experiments are used for scientific demonstration.

[0106] Ten healthy adult volunteers were selected as test subjects, and they wore the same type of monitoring bracelet. They performed routine life activities for 24 hours, during which time some of them intentionally set the bracelet to be loose, short-term off, and environmental interference, to simulate the occurrence of real abnormal conditions. The collected data included three physiological parameters: heart rate (bpm), blood oxygen saturation (%), and skin temperature (°C). The sampling frequency was set to 1 Hz to ensure high time resolution.

[0107] In the data processing stage, first, according to the present method, the complete data waveform collected is input into the trained neural network to identify the abnormal interval and classify it. Then, seven feature parameters (mean, variance, maximum, minimum, peak rate, baseline drift rate, and abnormal duration) are extracted under each abnormal classification to form a seven-dimensional feature space of the abnormal interval, and an adaptive feature midpoint is generated according to each classification. To make adaptive adjustments, multiple data segments in the normal interval are randomly intercepted, features are extracted and compared with the preset standard features, and adjustment factors are generated using hyperbolic tangent mapping to dynamically adjust the attention mechanism of the generator inside the adversarial network, so as to complete the rapid adaptation of individual physiological characteristics.

[0108] Subsequently, for the identified abnormal interval, data is supplemented based on the adjusted adversarial network, and the length of the generated data sequence is strictly consistent with the abnormal segment. To make a comparison, the traditional interpolation method (linear interpolation) is used as a reference standard, and the continuity index (such as root mean square error RMSE) of the supplemented data, the physiological feature consistency index (such as mean error), and the smoothness score of the waveform after abnormal correction (rated by a third-party evaluator) are recorded under each method.

[0109] Finally, through data index comparison, the superiority of the present method in terms of data quality and system adaptability is evaluated.

[0110] Specifically, in terms of continuity index (RMSE), the average error of the method is between 2.1-2.5 bpm, compared with 4.5-5.1 bpm of the traditional interpolation method, the error is reduced by more than 50%, showing the obvious advantage of the supplementary data in the continuity of the time series. In terms of mean error and variance error of the supplementary section, the mean error of the method is generally controlled in 1.0-1.3 bpm, and the variance error is between 3.2-3.8 bpm 2 , which is better than the corresponding error range (mean error 2.6-3.2 bpm, variance error 6.8-7.6 bpm 2 ) of the traditional interpolation method, indicating that the supplementary data more accurately restores the dynamic change characteristics of the physiological signal.

[0111] In terms of waveform smoothness score, the supplementary waveforms are subjectively scored by third-party evaluators, and the waveforms supplemented by the method generally get high scores of 9, while the waveforms supplemented by the traditional interpolation method only get scores of 5-6. The physiological feature consistency score also shows the obvious advantage of the method, further proving the improvement effect of the method in the naturalness and rationality of the supplementary data.

[0112] The above results fully demonstrate that, by adapting the attention mechanism guided by the characteristics of the normal interval before data supplementation, the adversarial network can quickly adapt to the current physiological state of the wearer, so as to accurately capture the key features in the process of abnormal section data supplementation, and improve the continuity, rationality and physiological feature consistency of the supplemented data.

[0113] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A data processing method based on a monitoring bracelet, characterized in that, The method comprises the following steps: Obtain physiological monitoring data of a monitoring object by monitoring a chip module built in a bracelet; Identify and classify abnormal intervals of the physiological monitoring data; Construct adaptive features of the monitoring object according to all abnormal interval classifications and the features of the abnormal intervals under each classification; Map parameters by using the adaptive features of the monitoring object to generate parameters of a data supplement model; After data cleaning of each abnormal interval, use the data supplement model after parameter assembly to supplement data; The abnormal interval includes an abnormal part in a waveform obtained by drawing the physiological monitoring data in a continuous time sequence; Input the waveform data into a trained neural network, identify the abnormal interval, and output the start timestamp, end timestamp, interval classification result, and interval waveform feature of each abnormal interval; The interval classification result includes a normal interval and an abnormal interval; The abnormal interval includes poor bracelet contact, bracelet falling off, and data distortion; The interval waveform feature includes the average value, variance, maximum value, minimum value, peak rate, and baseline drift rate of the data in the interval; In each abnormal interval of a category, seven interval waveform features are used as seven dimensions respectively; locating each abnormal interval of the waveforms in a 7-dimensional data space, resulting in a set of locations for all abnormal intervals: ; wherein, represents a set of coordinates for which the abnormal interval is classified as poor hand contact, represents a set of coordinates for which the abnormal interval is classified as hand loss, represents a set of coordinates for which the abnormal interval is classified as data distortion; In the coordinate set of each abnormal interval classification, the coordinates of the set midpoint are integrated, the discrete coordinates of the seven dimensions are calculated respectively, the mean values of the seven dimensions are obtained, and the coordinates of the set midpoint are obtained; The adaptive features include, under each abnormal interval classification, coordinates of a set midpoint; represented as: ; wherein, represents coordinates of a set midpoint for which the abnormal interval classification is poor hand ring contact, represents coordinates of a set midpoint for which the abnormal interval classification is hand ring loss, represents coordinates of a set midpoint for which the abnormal interval classification is data distortion. 2.The data processing method based on the monitoring bracelet according to claim 1, characterized in that: The physiological monitoring data includes heart rate, blood oxygen saturation, and skin temperature; Different physiological monitoring data are processed respectively. 3.The data processing method based on the monitoring bracelet according to claim 2, characterized in that: The parameter mapping includes, in a seven-dimensional space under each abnormal interval classification, preset standard coordinate points for representing preset standard values under each feature; a corresponding preset standard attention matrix; Let denotes any one of the coordinates; Will With The coordinates of the seven dimensions are compared respectively, and the increase and decrease of the coordinates are calculated in each dimension based on the coordinates of ​​ In the standard attention matrix, according to the increase / decrease ratio and the mapping function, the parameters of the attention matrix are adjusted to obtain The corresponding attention matrix. 4.The data processing method based on the monitoring bracelet according to claim 3, characterized in that: The adjusting the parameters of the attention matrix comprises, by a training set, The increasing and decreasing proportion of the coordinate data in each dimension is fitted with the change of the attention matrix parameters to obtain the mapping function. Adjust the parameters of the attention matrix in each dimension respectively, normalize the parameters of the adjusted attention matrix, and obtain the final attention matrix; The data supplement model is an adversarial network. 5.The data processing method based on the monitoring bracelet according to claim 4, characterized in that: The data supplement includes using the trained adversarial network to supplement data for each abnormal interval respectively; The input of the adversarial network is: complete waveform, start and end points of the abnormal interval, adjusted attention matrix, and classification of the abnormal interval; The output layer generates a supplementary waveform sequence with the same length as the original abnormal section; The head of the generator of the adversarial network is a classifier, which is responsible for classifying the abnormal interval into corresponding levels, and each level is connected to an independent data channel; For the waveform in the normal interval, divide it into windows according to a fixed time window length, and construct 7-dimensional selection space coordinates for the data in each window; For an abnormal interval p, set the time window in the previous normal interval as q and the time window in the next normal interval as h; Connect p and h in the 7-dimensional selection space, take the midpoint of the two as the coordinate position of p in the 7-dimensional selection space, calculate the relative distance from p to each center distance divided by the space radius of the cluster center represented by the cluster, and obtain the relative distance; pg represents the distance from p to the cluster center g, and the g corresponding to the minimum value of pg is taken as the cluster center of p; wherein each cluster center represents a level, and in the adaptation module of each channel, adaptation is performed in the corresponding cluster. The adaptation process of the adaptation module includes: presetting a standard coordinate different from the coordinate of the target image ; In the cluster family corresponding to the adaptation module, take A normal intervals; calculate the difference between the characteristics of the 7 dimensions in each interval and , so that each adaptation module adapts to the corresponding level of data, and when the adaptation is completed, the physiological monitoring data is supplemented.

6. A data processing system based on a monitoring bracelet using the method according to any one of claims 1 to 5, characterized in that: The acquisition unit acquires physiological monitoring data of the monitoring object by monitoring a chip module built in the bracelet. The classification unit identifies and classifies the abnormal interval of the physiological monitoring data. The analysis unit constructs adaptive features of the monitoring object according to all the abnormal interval classifications and the features of the abnormal interval under each classification. The adjustment unit generates parameters of the data supplement model through parameter mapping based on the adaptive features of the monitoring object. The supplement unit supplements data by using the data supplement model after data cleaning of each abnormal interval.

7. A computer device comprising: A memory and a processor; The memory stores a computer program, and the processor executes the computer program to implement the steps of the method according to any one of claims 1-5.

8. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to implement the steps of the method according to any one of claims 1-5.

Images