An environment perception-based smart watch multi-source data fusion method and system

Abstract

The application relates to the technical field of intelligent sensors and big data processing, in particular to a smart watch multi-source data fusion method and system based on environment sensing. The method comprises the following steps: obtaining an original data set through a smart watch; generating a data frame sequence according to a preprocessing result of the original data set, and generating a fusion feature vector based on the data frame sequence; inputting each fusion feature vector into a preset health mode classification model, and outputting an initial recognition result; generating a health risk level and an environmental influence factor according to the initial recognition result, and outputting user feedback information; extracting features such as a heart rate variation interval, a body temperature change rate and an air quality fluctuation cycle, constructing a fusion feature vector, combining a pre-trained health mode classification system, analyzing a linkage relationship between environmental changes and physiological reactions, and accurately identifying a stress mode caused by environmental pollution; and finally outputting personalized health prompt information based on a risk score mechanism and a historical mode library.

Description

Technical Field

[0001] This application relates to the field of intelligent sensor or big data processing technology, and in particular to a method and system for multi-source data fusion of smartwatches based on environmental perception. Background Technology

[0002] In today's era of rapid information and intelligent development, smart wearable devices, as important tools for personal health management, have undeniable value in research and application. Especially in the field of environmentally-based health monitoring, smartwatches, through smart chips integrating multi-source data, provide users with personalized health recommendations, becoming a key direction for improving quality of life and preventing disease. However, research and practice in this field still face many challenges and urgently need to overcome the limitations of existing technologies to meet users' needs for precise health management.

[0003] First, the smartwatch's chip needs to acquire real-time data about the user's environment through geolocation information, such as air quality and pollen concentration. However, data from different sources varies greatly in format, update frequency, and accuracy, making the integration of this data and ensuring its consistency a major challenge. Second, this integrated data needs to be deeply correlated with the user's physiological indicators, such as heart rate and body temperature, to identify the specific impact of environmental factors on individuals. This analysis often lacks targeted methods, making accurate judgments difficult. For example, in an area with poor air quality, a user might experience increased breathing rate, but current technology struggles to accurately distinguish whether this is caused by environmental pollution or the user's own physical activity. This ambiguity directly affects the reliability of health recommendations. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for multi-source data fusion in smartwatches based on environmental perception, in order to solve the above-mentioned technical problems and accurately identify the impact of environmental factors on individual health.

[0005] In some embodiments of this application, a multi-source data fusion method for smartwatches based on environmental awareness is provided, including: Obtain the raw data set using a smartwatch; A data frame sequence is generated based on the preprocessing results of the original dataset, and a fused feature vector is generated based on the data frame sequence. Each fused feature vector is input into a preset health pattern classification model, and the initial recognition result is output. Based on the initial identification results, a health risk level and environmental impact factors are generated, and user feedback information is output. The original dataset includes: initial physiological data stream and initial environmental data stream.

[0006] In some embodiments of this application, obtaining the original data set includes: The system collects the user's initial physiological data stream and geographic coordinates via a smartwatch. The initial physiological data stream includes: Real-time heart rate and body temperature data are collected and an initial physiological data stream is generated. Generate coordinate evaluation values ​​based on geographic coordinates; The area of ​​the data collection region is set based on the coordinate evaluation value, and the target area is set based on the geographic coordinates and the area of ​​the data collection region. The air quality index and pollen concentration index of the target area are obtained according to the preset environmental data interface; Generate the initial environment data stream; The original dataset is established based on the initial physiological data stream and the initial environmental data stream.

[0007] In some embodiments of this application, generating the data frame sequence includes: The original dataset is processed based on a timestamp alignment model; Generate time-aligned physiological data frames and initial environmental data frames; A judgment timeline is established based on physiological data frames, and the judgment timeline includes multiple time points; Select the target time points sequentially on the timeline; Missing values ​​in the environmental data frame at the target time point; If the missing value is greater than the preset missing value threshold, generate an interpolation instruction for the target time point; Sequentially determine whether interpolation instructions are generated at each time point, and generate an environment frame subsequence based on all interpolation instructions and the initial environment data frame; Generate physiological frame subsequences based on physiological data frames; A data frame sequence is established based on physiological and environmental frame sequences.

[0008] In some embodiments of this application, the generation of fused feature vectors based on data frame sequences includes: Heart rate and body temperature sequences are extracted from physiological frame subsequences based on a sliding time window. Calculate the standard deviation of adjacent RR intervals for the heart rate sequence and generate heart rate variability intervals; Calculate the slope of the linear regression of the body temperature sequence within the window and generate the rate of change of body temperature. Time series of PM2.5 concentration and pollen concentration were generated based on environmental frame subsequences; Air quality fluctuation cycles are generated based on the processing results of PM2.5 concentration time series. The anchoring time window is selected based on the air quality fluctuation cycle; Obtain the maximum value point of the pollen concentration time series within the anchored time window; Calculate the density distribution at the maximum point to generate the peak pollen concentration distribution characteristics; The heart rate variability interval, body temperature change rate, air quality fluctuation period, and pollen concentration peak distribution characteristics are aligned and merged according to time windows to construct a multi-dimensional fusion feature vector.

[0009] In some embodiments of this application, the output of the initial identification result includes: A structured dataset is generated based on the preprocessing results of the multidimensional fused feature vectors; Feature vectors of heart rate variation intervals and air quality mutation points are extracted from the structured dataset to construct a feature correlation matrix; Based on the correlation matrix of features, the correspondence between heart rate variability intervals and air quality mutation points is generated to determine whether it is a physiological stress pattern; If the judgment result is a physiological stress pattern, the initial recognition result will be output.

[0010] In some embodiments of this application, the generation of health risk levels and environmental impact factors includes: Obtain the association strength index based on the initial identification results; Based on the correlation strength index, key data sets related to health risks are obtained, and health risk values ​​are generated based on the key data sets. If the health risk value exceeds the preset health risk threshold, a level one processing instruction is generated. The exception mode library is invoked according to the first-level processing instructions; The abnormal pattern library includes multiple historical stress features; Generate similarity values ​​for key data sets and various historical stress features; Health risk levels and environmental impact factors are generated based on all similarity values.

[0011] In some embodiments of this application, a multi-source data fusion system for smartwatches based on environmental awareness is provided, including: Data unit, used to acquire raw data sets via smartwatch; The central control unit includes: The first processing module is used to generate a data frame sequence based on the preprocessing results of the original data set; The second processing module is used to generate a fused feature vector based on the data frame sequence. The third processing module is used to input the various fused feature vectors into the preset health pattern classification model and output the initial recognition results; The fourth processing module is used to generate health risk levels and environmental impact factors based on the initial identification results, and output user feedback information; The data unit is also used for: The system collects the user's initial physiological data stream and geographic coordinates via a smartwatch. The initial physiological data stream includes: Real-time heart rate and body temperature data are collected and an initial physiological data stream is generated. Generate coordinate evaluation values ​​based on geographic coordinates; The area of ​​the data collection region is set based on the coordinate evaluation value, and the target area is set based on the geographic coordinates and the area of ​​the data collection region. The air quality index and pollen concentration index of the target area are obtained according to the preset environmental data interface; Generate the initial environment data stream; The original dataset is established based on the initial physiological data stream and the initial environmental data stream.

[0012] In some embodiments of this application, the first processing module is further configured to: The original dataset is processed based on a timestamp alignment model; Generate time-aligned physiological data frames and initial environmental data frames; A judgment timeline is established based on physiological data frames, and the judgment timeline includes multiple time points; Select the target time points sequentially on the timeline; Missing values ​​in the environmental data frame at the target time point; If the missing value is greater than the preset missing value threshold, generate an interpolation instruction for the target time point; Sequentially determine whether interpolation instructions are generated at each time point, and generate an environment frame subsequence based on all interpolation instructions and the initial environment data frame; Generate physiological frame subsequences based on physiological data frames; A data frame sequence is established based on physiological and environmental frame sequences.

[0013] In some embodiments of this application, the second processing module is further configured to: Heart rate and body temperature sequences are extracted from physiological frame subsequences based on a sliding time window. Calculate the standard deviation of adjacent RR intervals for the heart rate sequence and generate heart rate variability intervals; Calculate the slope of the linear regression of the body temperature sequence within the window and generate the rate of change of body temperature. Time series of PM2.5 concentration and pollen concentration were generated based on environmental frame subsequences; Air quality fluctuation cycles are generated based on the processing results of PM2.5 concentration time series. The anchoring time window is selected based on the air quality fluctuation cycle; Obtain the maximum value point of the pollen concentration time series within the anchored time window; Calculate the density distribution at the maximum point to generate the peak pollen concentration distribution characteristics; The heart rate variability interval, body temperature change rate, air quality fluctuation period, and pollen concentration peak distribution characteristics are aligned and merged according to time windows to construct a multi-dimensional fusion feature vector.

[0014] In some embodiments of this application, the third processing module is further configured to: A structured dataset is generated based on the preprocessing results of the multidimensional fused feature vectors; Feature vectors of heart rate variation intervals and air quality mutation points are extracted from the structured dataset to construct a feature correlation matrix; Based on the correlation matrix of features, the correspondence between heart rate variability intervals and air quality mutation points is generated to determine whether it is a physiological stress pattern; If the judgment result is a physiological stress pattern, the initial recognition result will be output.

[0015] Compared with existing technologies, the multi-source data fusion method and system for smartwatches based on environmental perception disclosed in this application have the following advantages: The time granularity of physiological and environmental data streams is unified through timestamp alignment to ensure data synchronization. For missing environmental data, linear interpolation is used to generate continuous data sequences. By extracting features such as heart rate variability intervals, body temperature change rates, and air quality fluctuation cycles, a fusion feature vector is constructed. Combined with a pre-trained health pattern classification system, the correlation between environmental changes and physiological responses is analyzed to accurately identify stress patterns induced by environmental pollution. Finally, based on a risk scoring mechanism and a historical pattern database, personalized health alerts are output. This achieves a seamless integration of environmental perception and physiological monitoring, providing users with real-time and accurate health management support. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a multi-source data fusion method for smartwatches based on environmental perception, as described in a preferred embodiment of this application. Detailed Implementation

[0017] To enable those skilled in the art to better understand the technical solution, the present invention will be described in detail below with reference to embodiments. The description in this part is only exemplary and explanatory, and should not be used to limit the scope of protection of the present invention in any way.

[0018] like Figure 1 As shown in the preferred embodiment of this application, a multi-source data fusion method for smartwatches based on environmental perception includes: S101: Acquire raw data sets via smartwatch; S102: Generate a data frame sequence based on the preprocessing results of the original dataset, and generate a fusion feature vector based on the data frame sequence; S103: Input each fused feature vector into the preset health pattern classification model and output the initial recognition result; S104: Generate health risk levels and environmental impact factors based on the initial identification results, and output user feedback information; The original dataset includes: initial physiological data stream and initial environmental data stream.

[0019] Specifically, the smartwatch is equipped with smart sensors that can continuously collect the user's heart rate and body temperature data.

[0020] Specifically, by establishing an environmental data interface, the air quality index and pollen concentration index of a specific area can be obtained.

[0021] Specifically, obtaining the original data set includes: The system collects the user's initial physiological data stream and geographic coordinates via a smartwatch. The initial physiological data stream includes: Real-time heart rate and body temperature data are collected and an initial physiological data stream is generated. Generate coordinate evaluation values ​​based on geographic coordinates; The area of ​​the data collection region is set based on the coordinate evaluation value, and the target area is set based on the geographic coordinates and the area of ​​the data collection region. The air quality index and pollen concentration index of the target area are obtained according to the preset environmental data interface; Generate the initial environment data stream; The original dataset is established based on the initial physiological data stream and the initial environmental data stream.

[0022] Specifically, the smartwatch continuously collects the user's heart rate and body temperature data through its built-in sensors to build an initial physiological data stream. Based on the geographic coordinates determined by the smartwatch, the terrain parameters corresponding to the current geographic coordinates are obtained, and the corresponding terrain complexity (i.e., the elevation difference within a unit range) is generated. The higher the terrain complexity, the higher the probability of being in a mountainous area (the larger the elevation difference within a unit range). The unit range can be set based on historical parameters.

[0023] Specifically, the higher the terrain complexity, the greater the possibility of environmental differences (i.e., significant differences in environmental parameters between two areas) within a short distance, and the greater the corresponding coordinate evaluation value. The mapping relationship between the two can be set based on historical parameters.

[0024] Specifically, the larger the coordinate evaluation value, the smaller the corresponding collection area (the mapping relationship between the two can be set according to historical parameters), thereby updating and accurately collecting the environmental parameters around the user. On the other hand, the smaller the coordinate evaluation value, the flatter the current terrain, and the larger the corresponding collection area. This reduces the collection difficulty while ensuring the accuracy of environmental data (the smaller the collection environment, the higher the required data collection accuracy, and the greater the corresponding collection difficulty).

[0025] Specifically, the target area to be collected is determined based on geographical coordinates and the set area of ​​the collection area. The air quality index and pollen concentration index of the corresponding area are extracted from the preset environmental data interface to form an initial environmental data stream.

[0026] Specifically, by simultaneously collecting heart rate variability (HRV) and air quality data, denoising, aligning, and standardizing the data, a unified structured clean data is output. A basic pattern library (resting, exercise, environmental stress, etc.) is established, HRV and air quality mutation features are extracted, a cross-modal feature correlation matrix is ​​constructed, and a health pattern classification model is trained. This health pattern classification model can make pattern judgments. For example, by calculating the HRV interval and air quality change rate in real time, if "air quality rises rapidly and HRV drops significantly" is detected, it is determined to be a physiological stress pattern caused by environmental pollution.

[0027] For example, the photoplethysmography (PPG) sensor built into the smartwatch collects the user's raw heart rate signal once per second. Motion artifacts are removed by a moving average filtering algorithm. For instance, averaging 10 consecutive sample values ​​(such as 72, 75, 71, 80, 73, 74, 72, 76, 70, 74) yields a current effective heart rate of 73.7 beats per minute. Meanwhile, the temperature sensor collects the wrist skin temperature every 30 seconds, and after calibration and compensation, outputs 36.5 degrees Celsius. These real-time physiological data form an initial physiological data stream indexed by timestamps. Meanwhile, the current latitude and longitude coordinates (e.g., longitude 116.397, latitude 39.909) obtained by the smartwatch's GPS module are encapsulated as request parameters. These parameters are then used to call a pre-defined environmental data application interface via a wireless network. Upon receiving the coordinates, this interface assesses the terrain complexity corresponding to the coordinates, matches the corresponding region in its gridded environmental database, and returns structured data containing an air quality index (AQI) of 85 (good) and a pollen concentration of 120 grains per cubic meter. This constructs an initial environmental data stream synchronized with time. Subsequently, a data fusion service aligns and integrates this data stream using a unified time window (e.g., every minute). It encapsulates the heart rate (73.7), body temperature (36.5), AQI of 85, and pollen concentration of 120 at the same moment, along with the timestamp and geographic location information, into a single JSON data object. This results in a raw data set containing multi-dimensional information about the user's physiological state and the external environment, providing a foundation for subsequent real-time health analysis.

[0028] In a preferred embodiment of this application, generating a data frame sequence includes: The original dataset is processed based on a timestamp alignment model; Generate time-aligned physiological data frames and initial environmental data frames; A judgment timeline is established based on physiological data frames, and the judgment timeline includes multiple time points; Select the target time points sequentially on the timeline; Missing values ​​in the environmental data frame at the target time point; If the missing value is greater than the preset missing value threshold, generate an interpolation instruction for the target time point; Sequentially determine whether interpolation instructions are generated at each time point, and generate an environment frame subsequence based on all interpolation instructions and the initial environment data frame; Generate physiological frame subsequences based on physiological data frames; A data frame sequence is established based on physiological and environmental frame sequences.

[0029] Specifically, the timestamp alignment model can align all books in the original dataset, perform preliminary sorting of data streams with different update frequencies, and obtain preliminary data groups with uniform frequencies. Using a minute-level granularity standard, the physiological data streams and environmental data streams in the preliminary data groups are divided into time points, generating physiological data frames and environmental data frames based on the same time granularity, and determining the time-aligned data units.

[0030] Specifically, a time point is set for each physiological data frame, and a judgment time axis is constructed based on all time points.

[0031] Specifically, based on the timeline, the environmental data frames are examined to determine if there are missing values ​​for the Air Quality Index (AQI) and Pollen Concentration Index (PCI) at each time point. If missing values ​​are found, neighboring time points are identified. The rate of change of AQI between neighboring time points is calculated to determine abrupt changes in AQI. The difference in PCI between neighboring time points is calculated to obtain the pollen concentration gradient. Based on the identified abrupt changes in AQI and the pollen concentration gradient, linear interpolation is performed on the data for the missing time points. The missing values ​​are filled using the linear interpolation results to obtain a continuous and complete environmental data sequence.

[0032] For example, consider missing value detection and imputation for time-aligned environmental data frames. Assume we have an environmental data sequence containing time points, Air Quality Index (AQI), and Pollen Concentration Index (FCI), with data collected hourly from 00:00 to 23:00 on October 1, 2023, totaling 24 time points. In the initial data, the AQI value at 03:00 on October 1 is missing (null), while the FCI value is 50; similarly, the FCI value at 04:00 on October 1 is missing (null), while the AQI value is 80. The system automatically scans the data frames and detects missing values ​​for the AQI at 03:00 and the FCI at 04:00, triggering the interpolation imputation mechanism. For AQI missing value imputation, the system extracts data from nearby time points. The AQI at 02:00 is 60, and the AQI at 04:00 is 80. The slope of the abrupt change is calculated as k1 = (80-60) / (4-2) = 10, indicating that the AQI increases by 10 per hour. Therefore, the linear interpolation calculation for the AQI at 03:00 is 60 + 10 * 1 = 70. After imputation, the AQI at 03:00 is updated to 70. Next, for FCI missing value imputation, the system extracts data from nearby time points. The FCI at 03:00 is 50, and the FCI at 05:00 is 70. The gradient is calculated as k2 = (70-50) / (5-3) = 10, indicating that the FCI increases by 10 per hour. Therefore, the linear interpolation calculation for the FCI at 04:00 is 50 + 10 * 1 = 60. After imputation, the FCI at 04:00 is updated to 60. During the data infilling process, the system also incorporates business logic to verify whether the interpolation results are within a reasonable range. Assuming the reasonable range for AQI is 0 to 300 and for FCI is 0 to 200, the calculated results of 70 and 60 are both within this range, confirming the infilling is effective. If the results exceed the range, the system will call a backup algorithm, such as averaging nearby values, to ensure data reasonableness. Finally, the system generates a continuous and complete data sequence with AQI and FCI values ​​for all 24 time points, for example, (70, 50) at 03:00 and (80, 60) at 04:00. This data is stored in the database for subsequent analysis, ensuring the integrity of environmental data and supporting trend analysis of air quality and pollen concentration.

[0033] Specifically, linear interpolation is used to fill in missing data, obtaining a complete and aligned set of data frames (physiological frame subsequences and environmental frame subsequences). Based on this set of data frames, a synchronization foundation is constructed. Time points of the physiological and environmental data frames are matched one by one to determine if any time deviation exists. If the deviation exceeds a preset threshold, time correction is performed to obtain the synchronized data alignment result. From the synchronized data alignment result, key data points for subsequent analysis are extracted, generating a structured data table and determining the final data format for analysis. After obtaining the structured data table, consistency verification is performed on the physiological and environmental data frames. If the consistency check fails, the abnormal data points are marked, resulting in the verified final data set, i.e., the data frame sequence.

[0034] Specifically, a data storage structure is constructed based on the data frame sequence, and a hierarchical storage method is used to save physiological frame subsequences and environmental frame subsequences. The integrity of the storage is determined, and a synchronized data foundation that can be called is generated.

[0035] For example, we first unify the different update frequencies of physiological and environmental data streams to a minute-level time granularity using a timestamp alignment method. Specifically, assuming the physiological data stream updates heart rate data every 10 seconds, for instance, between 10:00:00 and 10:01:00 on October 1, 2023, there are six heart rate values: 75, 78, 80, 77, 79, and 76 beats per minute. Meanwhile, the environmental data stream updates temperature data every 30 seconds, with two temperature values ​​of 23.5°C and 23.8°C within the same minute. We then use a time window... The aggregation algorithm takes the average value of physiological data within a minute, i.e., (75+78+80+77+79+76) / 6=77.5 beats / minute, as the heart rate frame value for that minute. Environmental data is estimated at minute-level data points using a linear interpolation algorithm. If the temperature is 23.5°C at 30 seconds and 23.8°C at the next 30 seconds, the interpolated temperature at the end of the minute can be approximated as 23.65°C, serving as the environmental frame value for that minute. This method generates time-aligned physiological and environmental data frames, ensuring that each type of data has a corresponding value at each minute. Next, to ensure a one-to-one correspondence of data points in subsequent analysis, a time synchronization verification mechanism is designed. A timestamp comparison algorithm checks whether the time points of the two data frames are consistent. If a discrepancy is found, for example, the physiological data frame is at 10:00:00 while the environmental data frame is at 10:00:01, the environmental data frame is adjusted to 10:00:00 using nearest neighbor matching of timestamps, and the discrepancy is logged for later traceability. Finally, after determining the data foundation for synchronization, the system automatically generates a time-aligned dataset. For example, within the 5 minutes from 10:00:00 to 10:05:00 on October 1, 2023, there are heart rate frame values ​​such as 77.5, 78.2, 79.0, 77.8, 78.5 and temperature frame values ​​such as 23.65, 23.72, 23.80, 23.75, 23.70 per minute, forming a complete correspondence. This provides data support for subsequent analyses such as modeling the correlation between heart rate and temperature. The analysis process can calculate the correlation between the two sequences using the Pearson correlation coefficient. For example, a calculation result of 0.85 indicates that the two are highly correlated, thus verifying the synchronization effect.

[0036] In a preferred embodiment of this application, generating a fused feature vector based on a data frame sequence includes: Heart rate and body temperature sequences are extracted from physiological frame subsequences based on a sliding time window. Calculate the standard deviation of adjacent RR intervals for the heart rate sequence and generate heart rate variability intervals; Calculate the slope of the linear regression of the body temperature sequence within the window and generate the rate of change of body temperature. Time series of PM2.5 concentration and pollen concentration were generated based on environmental frame subsequences; Air quality fluctuation cycles are generated based on the processing results of PM2.5 concentration time series. The anchoring time window is selected based on the air quality fluctuation cycle; Obtain the maximum value point of the pollen concentration time series within the anchored time window; Calculate the density distribution at the maximum point to generate the peak pollen concentration distribution characteristics; The heart rate variability interval, body temperature change rate, air quality fluctuation period, and pollen concentration peak distribution characteristics are aligned and merged according to time windows to construct a multi-dimensional fusion feature vector.

[0037] Specifically, a sliding window method is used to extract heart rate and body temperature sequences from physiological data.

[0038] Specifically, the processing result of PM2.5 concentration time series refers to performing a fast Fourier transform on the PM2.5 concentration series, identifying the period corresponding to the dominant frequency (i.e., the heart rate variation interval), and determining the time window corresponding to the air quality fluctuation period as the anchor time window.

[0039] For example, assuming heart rate data is acquired from physiological data frames, sampled 60 times per minute within a 5-minute time window, the heart rate variability (HRV) is calculated. Using the standard deviation method, the standard deviation of 300 heart rate values ​​within 5 minutes is calculated, yielding an HRV value of 5.2 bpm, representing the amplitude of heart rate fluctuation. Simultaneously, body temperature data is sampled once per minute, with 5 temperature values ​​collected within 5 minutes. The rate of change in body temperature is calculated, and the slope is fitted using linear regression, yielding a rate of change of 0.1°C / minute, reflecting the upward trend in body temperature. Next, environmental data is processed. The Air Quality Index (AQI) is sampled hourly, with 24 values ​​collected over 24 hours. Fourier transform analysis of the fluctuation period reveals a main cycle of 6 hours, with an AQI fluctuation amplitude of ±15 units, indicating regular changes in air quality. Pollen concentration data is sampled hourly, with the peak occurring in the 10th hour of the 24-hour period at a concentration of 120 pollen grains / m³. Fitting the peak distribution using a Gaussian distribution yields a peak influence range of ±2 hours. Then, dynamic change values ​​for the above features are calculated within a 5-minute time window. The change value for heart rate variation is 0.3 bpm / window, the change value for body temperature change rate is 0.02°C / minute / window, the change value for air quality fluctuation is interpolated to the 5-minute window according to the period, with a change value of ±2 units / window, and the change value for pollen concentration influence is interpolated according to the peak distribution, with a change value of 10 grains / cubic meter / window. Finally, these dynamic change values ​​are constructed into a fused feature vector in the form of [0.3, 0.02, 2, 10], corresponding to the features of heart rate, body temperature, air quality, and pollen concentration, respectively. Through normalization processing (such as the Min-Max method, mapping each dimension value to the [0, 1] interval), the final vector [0.6, 0.4, 0.5, 0.8] is obtained, which is used for subsequent analysis or model input. The above process is automatically calculated by the algorithm, the data processing logic is rigorous, and the feature extraction and fusion form a complete chain, ensuring the multidimensionality and representativeness of the feature vector.

[0040] In a preferred embodiment of this application, the output of the initial identification result includes: A structured dataset is generated based on the preprocessing results of the multidimensional fused feature vectors; Feature vectors of heart rate variation intervals and air quality mutation points are extracted from the structured dataset to construct a feature correlation matrix; Based on the correlation matrix of features, the correspondence between heart rate variability intervals and air quality mutation points is generated to determine whether it is a physiological stress pattern; If the judgment result is a physiological stress pattern, the initial recognition result will be output.

[0041] Specifically, the multi-dimensional fused feature vectors are subjected to noise reduction and standardization to generate a unified structured dataset, resulting in clean data that can be used for subsequent analysis. For this clean data, a pre-trained health pattern classification system is used to extract feature vectors from heart rate variability intervals and air quality mutation points, constructing a feature correlation matrix to determine the linkage pattern between the two.

[0042] Specifically, if an increase in the rate of change in air quality and a significant decrease in the range of heart rate variability are detected, the pattern is determined to be a physiological stress pattern caused by environmental pollution based on the pattern classification boundary.

[0043] For example, the heart rate variability data and air quality index data are first standardized through data preprocessing. Assuming that the original heart rate variability data has a standard deviation of 2.5 to 5.0 per minute and the air quality index ranges from 50 to 200, the Z-score standardization method is used to convert the data into a distribution with a mean of 0 and a standard deviation of 1. Then, the two are merged into a 10-dimensional feature vector, where the first 5 dimensions represent the heart rate variability features and the last 5 dimensions represent the air quality features.

[0044] Next, the correlation between heart rate variability and air quality mutation points is calculated based on the feature correlation matrix. It is assumed that the correlation coefficient between the heart rate variability interval and the air quality index change rate obtained by the Pearson correlation coefficient algorithm is -0.75, indicating that the two are significantly negatively correlated. The diagonal elements in the matrix are set to 1, and the off-diagonal elements are the correlation coefficients, which are used for subsequent analysis.

[0045] Furthermore, utilizing the pattern classification boundary, the feature vectors are classified using a support vector machine algorithm. A linear kernel function is set as the classification boundary. The training dataset contains 5000 samples, where samples with an air quality index change rate greater than 10% and a heart rate variability interval decrease exceeding 20% ​​are labeled as physiological stress patterns, achieving a classification accuracy of 85%. If the air quality index change rate rises to 15% while the heart rate variability interval decreases to 30% of its original value, the system automatically identifies it as a physiological stress pattern caused by environmental pollution, outputs a preliminary pattern recognition result of "high-risk stress state," and records relevant feature values ​​and timestamps to the database, forming a data traceability chain.

[0046] Specifically, this involves generating health risk levels and environmental impact factors, including: Obtain the association strength index based on the initial identification results; Based on the correlation strength index, key data sets related to health risks are obtained, and health risk values ​​are generated based on the key data sets. If the health risk value exceeds the preset health risk threshold, a level one processing instruction is generated. The exception mode library is invoked according to the first-level processing instructions; The abnormal pattern library includes multiple historical stress features; Generate similarity values ​​for key data sets and various historical stress features; Health risk levels and environmental impact factors are generated based on all similarity values.

[0047] Specifically, the similarity between data in the key data set and historical stress characteristics is compared, and a corresponding similarity value is set based on the similarity. The higher the similarity, the larger the corresponding similarity value. All similarity values ​​are then generated to determine whether a high-risk pattern exists.

[0048] Specifically, based on the preliminary classification results, a correlation strength index between environmental pollution and heart rate variability is generated by combining the feature vectors of physiological stress patterns. According to the correlation strength index, high-risk physiological stress pattern data (data features related to physiological responses and environmental triggers) are screened out, thus obtaining key data sets related to health risks. For the screened key data sets, physiological response weights and environmental trigger thresholds are fused and weighted according to preset calculation rules to determine the current risk score. If the calculated risk score exceeds the preset health risk threshold, data from the abnormal pattern library is retrieved, and the similarity between the current data and historical stress features is compared to determine if there are potential high-risk patterns. Through comparison, abnormal pattern data matching historical stress features is obtained, and combined with the current risk score, the final health risk level is determined. Corresponding environmental influencing factors are extracted from the environmental trigger data, and correlation analysis is used to obtain specific environmental impact categories. Based on the environmental impact category and health risk level, corresponding risk distribution data is generated, stored in the abnormal pattern library, and the historical stress feature dataset is updated.

[0049] For example, in the implementation of health risk assessment, based on the preliminary pattern recognition results, the system first analyzes the user's physiological data such as heart rate, blood pressure, and skin conductance response. Assuming a heart rate of 85 beats / minute, blood pressure of 130 / 85 mmHg, and skin conductance response of 0.75 microSiemens, the system calculates a weighted average by combining preset physiological response weights (heart rate 0.4, blood pressure 0.3, skin conductance response 0.3), i.e., 0.4 * 85 risk threshold + risk threshold 0.3 * 130 risk threshold + risk threshold 0.3 * 0.75 risk threshold = risk threshold 34 risk threshold + risk threshold 39 risk threshold + risk threshold 0.225 risk threshold = risk threshold 73.225, which serves as the preliminary physiological risk index. Next, the system detects the current ambient temperature as 32 degrees Celsius, humidity as 80%, and noise level as 65 decibels based on environmental trigger thresholds. These are compared to the thresholds (temperature 30 degrees Celsius, humidity 75%, noise 60 decibels), and all three exceed the limits, triggering the environmental risk factor. The environmental risk bonus is calculated as: (32-30)*0.5 = Risk threshold + (80-75)*0.2 = Risk threshold + (65-60)*0.3 = Risk threshold 1 + Risk threshold 1 + Risk threshold 1.5 = Risk threshold 3.5. Subsequently, through a risk scoring mechanism, the physiological risk index is added to the environmental risk bonus, resulting in a current health risk score of 73.225: Risk threshold + Risk threshold 3.5 = Risk threshold 76.725. A risk score range is set (0-50 for low risk, 51-80 for medium risk, and 81-100 for high risk). The current score falls into the medium risk range. Furthermore, the system references historical stress pattern features stored in the abnormal pattern library, extracting characteristic data of user anxiety responses under similar heart rate and environmental conditions in the past. For example, the probability of anxiety is 70% when the heart rate is 80-90 beats / minute and the temperature is 30-35 degrees Celsius. Using a feature matching algorithm, the similarity between the current data and historical patterns is calculated to be 85%, confirming the risk level as medium risk. The specific types of environmental influencing factors are identified as high temperature and humidity and noise interference. Finally, the system automatically records the risk level and influencing factors into the health record, forming a complete logical chain from data collection to risk assessment. This ensures the targeted nature of subsequent health interventions, such as automatically generating cooling and noise reduction suggestions, thus perfecting the business loop.

[0050] Specifically, by acquiring real-time air quality data from the environmental monitoring system and analyzing its abrupt change trends, the intensity of physiological responses related to air quality changes is calculated based on these trends to determine the potential health risk level. If the health risk level exceeds a preset threshold, corresponding reaction delay data is extracted from the database to assess the duration and severity of the potential impact on the user. A comprehensive analysis of the reaction delay data and physiological response intensity generates health alerts matching the health risk level, providing specific protective recommendations. By comparing these recommendations with the user's historical data, personalized needs are analyzed to obtain customized health alert information (i.e., user feedback information).

[0051] For example, the system first collects real-time data on PM2.5, ozone, and other pollutant concentrations using environmental monitoring sensors. Then, it uses time-series prediction algorithms, such as the ARIMA model, to predict the Air Quality Index (AQI) for the next hour. When the AQI is predicted to surge from 80 (good) to 150 (moderate pollution) within the next 30 minutes, the system identifies this as an abrupt change in air quality. Next, the system accesses the user's health record database, combining historical data such as age (e.g., 45 years old) and history of chronic respiratory diseases (e.g., asthma), and calculates the health risk level using a pre-defined risk assessment matrix algorithm. For instance, this abrupt change might trigger a "high-risk" level for the current user. Then, the system queries a response delay database, built based on epidemiological research data. For example, for asthmatic individuals, the probability of experiencing physiological reactions such as coughing and shortness of breath within 2 to 4 hours after the AQI reaches 150 is 65%, with a predicted response intensity index of 7.2 (based on a 0-10 intensity scale). Based on the above analysis, the system generates a personalized alert: The core message includes, "Air quality is expected to drop to moderate pollution within the next half hour. Based on your health condition, there is a high probability of experiencing significant respiratory discomfort within the next 2-4 hours, with an estimated intensity of 7.2. It is recommended that you immediately close doors and windows, turn on an air purifier (preferably set to high-speed mode), and avoid outdoor activities. If mild symptoms have already appeared, you may consider using pre-stored medications." Finally, this alert is sent to the user's terminal in real time via push notification service, completing the automated feedback process.

[0052] In another preferred embodiment of the multi-source data fusion method for smartwatches based on environmental awareness, based on any of the above preferred embodiments, this preferred embodiment provides a multi-source data fusion system for smartwatches based on environmental awareness, comprising: Data unit, used to acquire raw data sets via smartwatch; The central control unit includes: The first processing module is used to generate a data frame sequence based on the preprocessing results of the original data set; The second processing module is used to generate a fused feature vector based on the data frame sequence. The third processing module is used to input the various fused feature vectors into the preset health pattern classification model and output the initial recognition results; The fourth processing module is used to generate health risk levels and environmental impact factors based on the initial identification results, and output user feedback information; The data unit is also used for: The system collects the user's initial physiological data stream and geographic coordinates via a smartwatch. The initial physiological data stream includes: Real-time heart rate and body temperature data are collected and an initial physiological data stream is generated. Generate coordinate evaluation values ​​based on geographic coordinates; The area of ​​the data collection region is set based on the coordinate evaluation value, and the target area is set based on the geographic coordinates and the area of ​​the data collection region. The air quality index and pollen concentration index of the target area are obtained according to the preset environmental data interface; Generate the initial environment data stream; The original dataset is established based on the initial physiological data stream and the initial environmental data stream.

[0053] In a preferred embodiment of this application, the first processing module is further configured to: The original dataset is processed based on a timestamp alignment model; Generate time-aligned physiological data frames and initial environmental data frames; A judgment timeline is established based on physiological data frames, and the judgment timeline includes multiple time points; Select the target time points sequentially on the timeline; Missing values ​​in the environmental data frame at the target time point; If the missing value is greater than the preset missing value threshold, generate an interpolation instruction for the target time point; Sequentially determine whether interpolation instructions are generated at each time point, and generate an environment frame subsequence based on all interpolation instructions and the initial environment data frame; Generate physiological frame subsequences based on physiological data frames; A data frame sequence is established based on physiological and environmental frame sequences.

[0054] In a preferred embodiment of this application, the second processing module is further configured to: Heart rate and body temperature sequences are extracted from physiological frame subsequences based on a sliding time window. Calculate the standard deviation of adjacent RR intervals for the heart rate sequence and generate heart rate variability intervals; Calculate the slope of the linear regression of the body temperature sequence within the window and generate the rate of change of body temperature. Time series of PM2.5 concentration and pollen concentration were generated based on environmental frame subsequences; Air quality fluctuation cycles are generated based on the processing results of PM2.5 concentration time series. The anchoring time window is selected based on the air quality fluctuation cycle; Obtain the maximum value point of the pollen concentration time series within the anchored time window; Calculate the density distribution at the maximum point to generate the peak pollen concentration distribution characteristics; The heart rate variability interval, body temperature change rate, air quality fluctuation period, and pollen concentration peak distribution characteristics are aligned and merged according to time windows to construct a multi-dimensional fusion feature vector.

[0055] In a preferred embodiment of this application, the third processing module is further configured to: A structured dataset is generated based on the preprocessing results of the multidimensional fused feature vectors; Feature vectors of heart rate variation intervals and air quality mutation points are extracted from the structured dataset to construct a feature correlation matrix; Based on the correlation matrix of features, the correspondence between heart rate variability intervals and air quality mutation points is generated to determine whether it is a physiological stress pattern; If the judgment result is a physiological stress pattern, the initial recognition result will be output.

[0056] Based on the first concept of this application, a timestamp alignment method is used to unify the temporal granularity of physiological and environmental data streams, ensuring data synchronization. For missing environmental data, linear interpolation is used to generate continuous data sequences. By extracting features such as heart rate variability intervals, body temperature change rates, and air quality fluctuation cycles, a fused feature vector is constructed. Combined with a pre-trained health pattern classification system, the linkage between environmental changes and physiological responses is analyzed to accurately identify stress patterns induced by environmental pollution. Finally, based on a risk scoring mechanism and a historical pattern database, personalized health alerts are output. This achieves a seamless integration of environmental perception and physiological monitoring, providing users with real-time and accurate health management support.

[0057] It should be noted that, in this document, the terms "comprising," "including," and any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Specific examples have been used in this document to illustrate the principles and implementation methods of the present invention. These examples are only intended to aid in understanding the method and core ideas of the present invention. The above descriptions are merely preferred embodiments of the present invention. It should be pointed out that, due to the limitations of written expression and the objective existence of infinite specific structures, those skilled in the art can make various improvements, modifications, or variations without departing from the principles of the present invention, and can also combine the above technical features in an appropriate manner. These improvements, modifications, variations, or combinations, or the direct application of the concept and technical solution of the present invention to other situations without modification, should all be considered within the scope of protection of the present invention.

Claims

1. A multi-source data fusion method for smartwatches based on environmental perception, characterized in that, include: Obtain the raw data set using a smartwatch; A data frame sequence is generated based on the preprocessing results of the original dataset, and a fused feature vector is generated based on the data frame sequence. Each fused feature vector is input into a preset health pattern classification model, and the initial recognition result is output. Based on the initial identification results, a health risk level and environmental impact factors are generated, and user feedback information is output. The original dataset includes: initial physiological data stream and initial environmental data stream.

2. The multi-source data fusion method for smartwatches based on environmental perception as described in claim 1, characterized in that, The acquisition of the original data set includes: The system collects the user's initial physiological data stream and geographic coordinates via a smartwatch. The initial physiological data stream includes: Real-time heart rate and body temperature data are collected and an initial physiological data stream is generated. Generate coordinate evaluation values ​​based on geographic coordinates; The area of ​​the data collection region is set based on the coordinate evaluation value, and the target area is set based on the geographic coordinates and the area of ​​the data collection region. The air quality index and pollen concentration index of the target area are obtained according to the preset environmental data interface; Generate the initial environment data stream; The original dataset is established based on the initial physiological data stream and the initial environmental data stream.

3. The multi-source data fusion method for smartwatches based on environmental perception as described in claim 2, characterized in that, The generated data frame sequence includes: The original dataset is processed based on a timestamp alignment model; Generate time-aligned physiological data frames and initial environmental data frames; A judgment timeline is established based on physiological data frames, and the judgment timeline includes multiple time points; Select the target time points sequentially on the timeline; Missing values ​​in the environmental data frame at the target time point; If the missing value is greater than the preset missing value threshold, generate an interpolation instruction for the target time point; Sequentially determine whether interpolation instructions are generated at each time point, and generate an environment frame subsequence based on all interpolation instructions and the initial environment data frame; Generate physiological frame subsequences based on physiological data frames; A data frame sequence is established based on physiological and environmental frame sequences.

4. The multi-source data fusion method for smartwatches based on environmental perception as described in claim 3, characterized in that, The generation of fused feature vectors based on data frame sequences includes: Heart rate and body temperature sequences are extracted from physiological frame sub-sequences based on a preset sliding time window; Calculate the standard deviation of adjacent RR intervals for the heart rate sequence and generate heart rate variability intervals; Calculate the slope of the linear regression of the body temperature sequence within the window and generate the rate of change of body temperature. Time series of PM2.5 concentration and pollen concentration were generated based on environmental frame subsequences; Air quality fluctuation cycles are generated based on the processing results of PM2.5 concentration time series. The anchoring time window is selected based on the air quality fluctuation cycle; Obtain the maximum value point of the pollen concentration time series within the anchored time window; Calculate the density distribution at the maximum point to generate the peak pollen concentration distribution characteristics; The heart rate variability interval, body temperature change rate, air quality fluctuation period, and pollen concentration peak distribution characteristics are aligned and merged according to time windows to construct a multi-dimensional fusion feature vector.

5. The multi-source data fusion method for smartwatches based on environmental perception as described in claim 4, characterized in that, The initial recognition result output includes: A structured dataset is generated based on the preprocessing results of the multidimensional fused feature vectors; Feature vectors of heart rate variation intervals and air quality mutation points are extracted from the structured dataset to construct a feature correlation matrix; Based on the correlation matrix of features, the correspondence between heart rate variability intervals and air quality mutation points is generated to determine whether it is a physiological stress pattern; If the judgment result is a physiological stress pattern, the initial recognition result will be output.

6. The multi-source data fusion method for smartwatches based on environmental perception as described in claim 5, characterized in that, The generation of health risk levels and environmental impact factors includes: Obtain the association strength index based on the initial identification results; Based on the correlation strength index, key data sets related to health risks are obtained, and health risk values ​​are generated based on the key data sets. If the health risk value exceeds the preset health risk threshold, a level one processing instruction is generated. The exception mode library is invoked according to the first-level processing instructions; The abnormal pattern library includes multiple historical stress features; Generate similarity values ​​for key data sets and various historical stress features; Health risk levels and environmental impact factors are generated based on all similarity values.

7. A multi-source data fusion system for smartwatches based on environmental perception, employing the multi-source data fusion method for smartwatches based on environmental perception as described in any one of claims 1-6, characterized in that, include: Data unit, used to acquire raw data sets via smartwatch; The central control unit includes: The first processing module is used to generate a data frame sequence based on the preprocessing results of the original data set; The second processing module is used to generate a fused feature vector based on the data frame sequence. The third processing module is used to input the various fused feature vectors into the preset health pattern classification model and output the initial recognition results; The fourth processing module is used to generate health risk levels and environmental impact factors based on the initial identification results, and output user feedback information; The data unit is also used for: The system collects the user's initial physiological data stream and geographic coordinates via a smartwatch. The initial physiological data stream includes: Real-time heart rate and body temperature data are collected and an initial physiological data stream is generated. Generate coordinate evaluation values ​​based on geographic coordinates; The area of ​​the data collection region is set based on the coordinate evaluation value, and the target area is set based on the geographic coordinates and the area of ​​the data collection region. The air quality index and pollen concentration index of the target area are obtained according to the preset environmental data interface; Generate the initial environment data stream; The original dataset is established based on the initial physiological data stream and the initial environmental data stream.

8. The multi-source data fusion system for smartwatches based on environmental perception as described in claim 7, characterized in that, The first processing module is further configured to: The original dataset is processed based on a timestamp alignment model; Generate time-aligned physiological data frames and initial environmental data frames; A judgment timeline is established based on physiological data frames, and the judgment timeline includes multiple time points; Select the target time points sequentially on the timeline; Missing values ​​in the environmental data frame at the target time point; If the missing value is greater than the preset missing value threshold, generate an interpolation instruction for the target time point; Sequentially determine whether interpolation instructions are generated at each time point, and generate an environment frame subsequence based on all interpolation instructions and the initial environment data frame; Generate physiological frame subsequences based on physiological data frames; A data frame sequence is established based on physiological and environmental frame sequences.

9. The multi-source data fusion system for smartwatches based on environmental perception as described in claim 8, characterized in that, The second processing module is also used for: Heart rate and body temperature sequences are extracted from physiological frame subsequences based on a sliding time window. Calculate the standard deviation of adjacent RR intervals for the heart rate sequence and generate heart rate variability intervals; Calculate the slope of the linear regression of the body temperature sequence within the window and generate the rate of change of body temperature. Time series of PM2.5 concentration and pollen concentration were generated based on environmental frame subsequences; Air quality fluctuation cycles are generated based on the processing results of PM2.5 concentration time series. The anchoring time window is selected based on the air quality fluctuation cycle; Obtain the maximum value point of the pollen concentration time series within the anchored time window; Calculate the density distribution at the maximum point to generate the peak pollen concentration distribution characteristics; The heart rate variability interval, body temperature change rate, air quality fluctuation period, and pollen concentration peak distribution characteristics are aligned and merged according to time windows to construct a multi-dimensional fusion feature vector.

10. The multi-source data fusion system for smartwatches based on environmental perception as described in claim 9, characterized in that, The third processing module is also used for: A structured dataset is generated based on the preprocessing results of the multidimensional fused feature vectors; Feature vectors of heart rate variation intervals and air quality mutation points are extracted from the structured dataset to construct a feature correlation matrix; Based on the correlation matrix of features, the correspondence between heart rate variability intervals and air quality mutation points is generated to determine whether it is a physiological stress pattern; If the judgment result is a physiological stress pattern, the initial recognition result will be output.

Images