Intelligent fatigue state assessment and monitoring system and device based on multimodal sensing

Abstract

The present invention discloses an intelligent fatigue state assessment and monitoring system and device based on multimodal sensing, which relates to the field of electrical digital data processing technology. The intelligent fatigue state assessment and monitoring system based on multimodal sensing includes: an abnormal state estimation module, an abnormal state cumulative analysis module and a fatigue degree identification module. The present invention estimates the abnormal state through an abnormal state identification model, and at the same time performs an impact degree analysis to optimize the state estimation error. Then, a cumulative analysis is performed through a trend prediction algorithm to optimize the cumulative analysis error. Finally, the fatigue degree is identified through a hierarchical early warning mechanism to determine whether there are different levels of early warning and feedback requirements. This achieves the effect of improving the accuracy of intelligent fatigue state estimation and monitoring, and solves the problem in the prior art that the intelligent fatigue state estimation and monitoring accuracy is low due to the asynchronous update time of the corresponding monitoring parameters in the intelligent fatigue state estimation and identification process.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic digital data processing, and in particular to an intelligent fatigue state assessment and monitoring system and device based on multimodal sensing. Background Art

[0002] Wearable devices are playing an increasingly important role in fatigue monitoring. These devices typically integrate multimodal sensors, such as PPG (Photoplethysmography) heart rate sensors that accurately monitor heart rate changes, EDA (Electrodermal Activity) skin conductance sensors that capture electrical activity in the skin, reflecting stress levels, accelerometers that record physical activity, and temperature sensors that detect body temperature fluctuations. These sensors work together to continuously collect multi-source physiological data in real time, providing rich and comprehensive information for fatigue monitoring.

[0003] However, in actual applications, it can also be combined with other monitoring devices to further improve the capture of eye movement trajectories and blinking movements. For example, eye trackers provide high-precision eye data. In conjunction with the accelerometer in the wearable device, the number of blinks can be more accurately obtained, and then the degree of fatigue can be analyzed. For short-range communication, Bluetooth technology is a common choice. It allows wearable devices to quickly establish a connection with devices such as eye trackers and achieve stable data transmission. For example, a smart watch is connected to an eye tracker via Bluetooth. The eye data collected by the eye tracker can be synchronized to the watch in real time, and then further processed by the watch or uploaded to the cloud.

[0004] If the devices are far apart, Wi-Fi communication is more suitable. For example, in hospitals or large office spaces, multiple monitoring devices can be connected to a unified network via Wi-Fi, and the data can be aggregated to a central server for analysis and processing.

[0005] The existing system primarily consists of a data acquisition module, a feature extraction module, a fusion module, and a fatigue monitoring and early warning module. The data acquisition module uses wearable devices and other monitoring equipment to collect multi-source physiological data in real time. The feature extraction module extracts fatigue-related features from this data. The fusion module fuses these multimodal features to provide a more comprehensive assessment of fatigue status. The fatigue monitoring and early warning module determines fatigue status based on the fused features and issues an alarm. The system enables inter-device communication via Bluetooth or Wi-Fi, ensuring stable data transmission.

[0006] For example, the Chinese invention patent publication number CN118733999A discloses a multimodal intelligent cockpit driver fatigue monitoring system and model construction method, which includes: collecting multimodal data of the driver; extracting fatigue-related features from the multimodal data; fusing the multimodal features to obtain a driver fatigue status assessment; judging the driver's fatigue status based on the fused features and issuing an alarm at the same time.

[0007] For example, the fatigue recognition method and fatigue recognition device based on three-lead forehead EEG signals disclosed in the Chinese invention patent with publication number CN119719747A include: obtaining the three-lead forehead EEG signals of production safety personnel in real time; inputting the three-lead forehead EEG signals into an EEG fatigue recognition model, analyzing the three-lead forehead EEG signals using the EEG fatigue recognition model, and outputting a fatigue status label value; when the fatigue status label value is greater than the fatigue warning value, issuing a prompt message to remind the corresponding production safety personnel to suspend work.

[0008] In the prior art, during the data collection phase, noise interference further amplifies the impact of this asynchrony. Different sensors are subject to varying degrees of environmental noise interference, and inconsistent update times prevent the impact of noise on various parameters from being considered on a consistent basis, resulting in uneven data quality. Furthermore, over time, the cumulative error caused by this asynchrony increases. During feature extraction and fusion, since the parameters are not synchronized at the same time, the fused features cannot accurately reflect the true fatigue state, leading to biased fatigue state assessment. Furthermore, this inaccuracy directly impacts the practical application of fatigue monitoring systems. In scenarios requiring real-time warnings, asynchrony in parameter updates (such as data sampling frequency and sliding window length) can lead to inaccurate assessment results. This prevents the collected data from accurately corresponding to the same moment, resulting in biased assessment results. This makes it difficult to trigger warnings before high-intensity workers enter a dangerous fatigue state. This can lead to operator errors caused by excessive fatigue, potentially delaying warnings. Consequently, the asynchrony in the update times of the corresponding monitoring parameters during intelligent fatigue state estimation and recognition can lead to low accuracy in intelligent fatigue state estimation and monitoring. Summary of the Invention

[0009] In order to solve the technical problems in the prior art, the present invention provides an intelligent fatigue state assessment and monitoring system and device based on multimodal sensing. The technical solution is as follows:

[0010] On the one hand, an intelligent fatigue state assessment and monitoring system based on multimodal sensing is provided, which includes: an abnormal state estimation module, an abnormal state accumulation analysis module and a fatigue degree identification module; wherein, the abnormal state estimation module is used to obtain the state monitoring data of the target object through the multimodal sensor in the wearable device, and input it into the constructed abnormal state identification model to perform abnormal state estimation, and at the same time, based on the acquired state estimation influence data, the influence degree of the abnormal state estimation process is analyzed to determine whether to optimize the state estimation error based on the acquired heart rate signal spectrum, so as to improve the accuracy of abnormal state estimation under noise interference, and the state estimation influence data includes the maximum tolerance time Long switching delay and data sampling frequency switching delay; the abnormal state cumulative analysis module is used to obtain the abnormal state data output by the abnormal state recognition model and perform cumulative analysis when the abnormal state estimation process is qualified, and at the same time perform cumulative analysis error optimization judgment to improve the matching degree between the sliding window parameters and the actual change characteristics of the physiological signal; the fatigue level recognition module is used to identify the fatigue level of the preset recognition period based on the results of the cumulative analysis when the cumulative analysis is qualified, so as to determine whether different levels of warning and feedback requirements are required based on the duration of the abnormal state change corresponding to the fatigue level recognition process, so as to improve the timeliness of the triggering of different levels of warnings corresponding to different recognition periods.

[0011] On the other hand, an intelligent fatigue state assessment and monitoring device based on multimodal sensing is provided, which includes: an electrocardiogram sensor, a timer and a server; the electrocardiogram sensor is used to monitor the heart rate signal spectrum; the timer is used to monitor the maximum tolerance time switching delay, data sampling frequency switching delay, actual processing time and duration; the server is used to run an abnormal state recognition model and a hierarchical early warning mechanism.

[0012] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0013] 1. Abnormal state estimation and impact analysis are performed through the abnormal state recognition model. At the same time, state estimation error judgment is performed based on the results of the impact analysis. It can accurately capture the anomalies of monitoring parameters such as maximum tolerance time and data sampling frequency at the same time node, and effectively solve the problem of abnormal state judgment deviation caused by time asynchrony of monitoring parameters; then use the trend prediction algorithm to perform cumulative analysis and optimize error judgment. Based on the synchronized monitoring parameter time series, it prevents the deviation of cumulative analysis due to time asynchrony and realizes the optimized judgment of cumulative analysis error; finally, with the help of a hierarchical early warning mechanism to identify fatigue state, it can accurately determine the early warning and feedback needs of different levels based on accurate state estimation and trend prediction, effectively solve the problem of inaccurate fatigue state identification caused by time asynchrony, and improve the reliability of intelligent fatigue state estimation and monitoring.

[0014] 2. By first obtaining the response state of the maximum tolerated duration and data sampling frequency in abnormal state estimation, the state estimation impact data related to the switching delay is obtained. After harmonic averaging, the abnormal state estimation interference value is obtained. Based on this, it is determined whether to perform state estimation error optimization. If it fails, the heart rate signal spectrum is analyzed and the filter cutoff frequency is reduced by combining the mapping relationship to reduce low-frequency noise interference. This method can effectively quantify the impact of changes in monitoring parameters on abnormal state estimation. By specifically reducing the filter cutoff frequency, it can effectively reduce low-frequency noise interference such as breathing or movement, making the abnormal state estimation process more stable and reliable. It ensures that when the monitoring parameters change over time, it can still maintain a high estimation accuracy, thereby improving the reliability of intelligent fatigue state assessment and monitoring.

[0015] 3. By first dividing the cumulative analysis period, the changing trend of the physiological abnormality cumulative index at each moment is monitored in real time. The trend line is fitted using the least squares method to obtain the slope of the abnormal state trend, which is used to determine the increase or decrease in the target subject's abnormal state cumulative intensity. The actual processing time corresponding to the abnormal state trend slope is then recorded to determine the effectiveness of the current window parameters. This method accurately captures the changing trend of the abnormal state cumulative intensity, and the abnormal state trend slope intuitively reflects the evolution rate. At the same time, by comparing the actual processing time with the set period, window parameter deviations can be promptly identified. Error optimization ensures the accuracy of the cumulative analysis in the time dimension, avoiding the impact of time asynchrony and other issues on fatigue state assessment results. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A flowchart of an intelligent fatigue state assessment and monitoring system based on multimodal sensing provided by an embodiment of the present invention;

[0017] Figure 2 A flowchart of the abnormal state estimation interference degree quantification and error optimization determination provided by an embodiment of the present invention;

[0018] Figure 3 A flowchart of the abnormal state accumulation analysis process quantification and error optimization determination provided by an embodiment of the present invention;

[0019] Figure 4 A flowchart of the hierarchical warning and feedback determination process for fatigue level identification provided by an embodiment of the present invention;

[0020] Figure 5 The homepage of the intelligent fatigue status assessment and monitoring system based on multimodal sensing provided by an embodiment of the present invention;

[0021] Figure 6 This is a fatigue state recognition interface of an intelligent fatigue state assessment and monitoring system based on multimodal sensing provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0023] The embodiment of the present invention provides an intelligent fatigue state assessment and monitoring system based on multimodal sensing, such as Figure 1 The structure diagram of the intelligent fatigue state assessment and monitoring system based on multimodal sensing is shown, which includes the following modules: an abnormal state estimation module, an abnormal state accumulation analysis module and a fatigue degree identification module.

[0024] The present invention realizes intelligent fatigue state assessment and monitoring of multimodal sensing through the collaboration of multiple modules. The abnormal state estimation module uses the multimodal sensors of the wearable device to collect state monitoring data, inputs the abnormal state recognition model for estimation, and analyzes the influence of factors such as noise, and optimizes the error to ensure the consistency of the data benchmark. After the abnormal state cumulative analysis module is qualified, it performs cumulative analysis on the abnormal state data, optimizes the cumulative error, and improves the matching degree between the sliding window and the physiological signal changes. Based on the qualified cumulative analysis results, the fatigue level recognition module identifies the fatigue level in a preset time period, determines the warning requirements at different levels, ensures that the warning is triggered in time, and realizes full-process intelligent monitoring from multimodal data collection and analysis to accurate warning.

[0025] Specifically, the abnormal state estimation module is used to obtain state monitoring data of the target object through the multimodal sensors in the wearable device and input it into the established abnormal state recognition model. The abnormal state is estimated by the abnormal state recognition model. At the same time, the influence degree of the abnormal state estimation process is analyzed based on the acquired state estimation impact data to determine whether to optimize the state estimation error. The state estimation error optimization is used to reduce low-frequency interference caused by the target object's breathing or movement. The abnormal state accumulation analysis module is used to obtain abnormal state data output by the abnormal state recognition model when the abnormal state estimation process is qualified. The acquired abnormal state data is cumulatively analyzed using a trend prediction algorithm (i.e., the physiological abnormality accumulation index under the sliding window) to evaluate the effectiveness of the current window parameter update. At the same time, the cumulative analysis error optimization is used to determine whether to improve the matching between the sliding window parameters and the actual change characteristics of the physiological signal by adjusting the window length and time step. The fatigue level recognition module is used to input the results of the cumulative analysis into the hierarchical warning mechanism when the cumulative analysis is qualified. The fatigue level recognition module performs fatigue level recognition for a preset recognition period. Based on the duration of the abnormal state change during the fatigue level recognition process, it determines whether different levels of warning and feedback requirements are required.

[0026] It's important to understand that state monitoring data is a set of quantitative characteristic parameters that reflect how the human physiological state changes over the circadian cycle. These parameters encompass multiple dimensions: the temporal dimension includes the circadian clock phase (e.g., the onset of melatonin secretion, the lowest point of core body temperature), and circadian rhythm stability (the amplitude of fluctuations in physiological indicators within a 24-hour cycle); the physiological dimension encompasses diurnal variations in cortisol levels, heart rate variability (HRV), EEG power spectrum distribution (e.g., an increased proportion of delta waves at night), and sleep architecture parameters (the proportion of time in deep sleep); and the behavioral dimension encompasses the diurnal distribution of activity levels and periodic variations in operational reaction speed. These parameters are continuously collected via wearable devices or environmental sensors and extracted using time-series analysis algorithms (e.g., cosine fitting and nonlinear mixed-effects models) to form a dynamic feature vector representing an individual's circadian state, the result of dynamic circadian rhythm analysis.

[0027] After inputting these results into the abnormal state recognition model, the model uses a multimodal feature fusion mechanism to combine real-time physiological signals (such as electrocardiograms and electroencephalograms) with behavioral data (such as head posture and operation frequency). It then utilizes deep learning architectures (such as LSTM-Attention networks) to capture patterns in the association between circadian rhythms and abnormal states. For example, in a long-distance transport dispatch scenario, if abnormally low cortisol levels and excessive delta EEG wave power (a low-frequency brain wave recorded in an EEG, typically between 0.5 and 4 Hz) are detected at 3:00 AM, the model will dynamically adjust fatigue threshold parameters based on the driver's historical fatigue event records, output a high-confidence fatigue warning, and generate a personalized report with circadian rhythm intervention recommendations (such as adjusting the shift schedule or supplementing light stimulation).

[0028] In this embodiment, the status monitoring data provides comprehensive and accurate basic data for abnormal state identification. The abnormal state estimation module uses a multimodal feature fusion mechanism, combines real-time physiological signals and behavioral data, and uses a deep learning architecture to capture the correlation pattern between circadian rhythms and abnormal states, avoiding the limitations of single parameter judgment. The abnormal state accumulation analysis module evaluates the effectiveness of window parameter updates through a trend prediction algorithm, and performs error optimization judgment to ensure that the analysis results are reliable. The fatigue level identification module accurately determines the warning needs based on the hierarchical warning mechanism. The synergistic effect of each module effectively solves the problem of inaccurate fatigue state assessment of the target object caused by the delay in the time dimension of the monitoring parameters in the existing technology.

[0029] Furthermore, the degree of influence of the abnormal state estimation process is analyzed based on the acquired state estimation impact data. The specific process is as follows: the response state corresponding to the maximum tolerance time and the data sampling frequency in the abnormal state estimation process is obtained to obtain the state estimation impact data. The state estimation impact data includes the maximum tolerance time switching delay and the data sampling frequency switching delay. The maximum tolerance time switching delay indicates that when the maximum tolerance time required for the abnormal state estimation changes, the time interval required to complete the current maximum tolerance time switching is changed. The data sampling frequency switching delay indicates that when the data sampling frequency required for the abnormal state estimation changes, the time interval required to complete the current data sampling frequency switching is changed; the acquired maximum tolerance time switching delay and the data sampling frequency switching delay are harmonically averaged to obtain the abnormal state estimation interference value. The abnormal state estimation interference value represents quantitative data of the degree of influence of the acquired state estimation impact data on the accuracy of the abnormal state estimation.

[0030] In this embodiment, an increased maximum tolerance switching delay means the system spends more time adjusting the maximum tolerance time. This causes data sampling to consume additional time to maintain the original frequency, increasing the waiting time for the maximum tolerance time adjustment and indirectly lengthening the data sampling frequency switching delay. Conversely, if the data sampling frequency switching delay increases, the system spends more time adjusting the sampling frequency, disrupting the normal maximum tolerance switching process and requiring the system to spend more time coordinating the two, resulting in an increase in the maximum tolerance switching delay. These two factors mutually constrain each other, jointly impacting the stability and accuracy of abnormal state estimation.

[0031] Existing technologies may only consider a single factor or a simple combination of factors in the abnormal state estimation process, making it difficult to accurately quantify the impact of state estimation influencing data on the accuracy of abnormal state estimation. However, this method obtains the response state corresponding to the maximum tolerance time and data sampling frequency in the abnormal state estimation process, clearly obtains the two key state estimation influencing data: maximum tolerance time switching delay and data sampling frequency switching delay, and then obtains the abnormal state estimation interference value through harmonic averaging, achieving accurate quantification of the degree of impact. This interference value can comprehensively reflect the comprehensive impact of state estimation influencing data on the accuracy of abnormal state estimation, providing an intuitive and quantitative basis for evaluating the stability and reliability of the abnormal state estimation process, helping technicians to gain a deep understanding of system characteristics, optimize the settings of maximum tolerance time and data sampling frequency in a targeted manner, and improve the accuracy of abnormal state estimation.

[0032] like Figure 2As shown, it is a flowchart of the abnormal state estimation interference degree quantification and error optimization judgment provided by the embodiment of the present application. Its specific logic is: first, it is determined whether the abnormal state estimation interference value is less than or equal to the corresponding set value. If so, cumulative analysis is performed directly, otherwise it enters the estimation error optimization process, including analyzing the heart rate spectrum, adjusting the filter cutoff frequency and the decision factor ratio, and then re-estimating. Then it is determined again whether the newly obtained interference value is greater than the set value. If so, the abnormal state estimation warning is triggered, otherwise the estimation error optimization is completed and cumulative analysis is carried out. The entire logic ensures the accuracy and reliability of the abnormal state estimation through continuous judgment and adjustment.

[0033] What needs to be further understood is that the specific process of determining whether to perform state estimation error optimization is as follows: if the acquired abnormal state estimation interference value is not greater than the abnormal state estimation interference value set in the database, the abnormal state estimation process is determined to be qualified and cumulative analysis is performed; otherwise, the abnormal state estimation process is determined to be unqualified and state estimation error optimization is performed.

[0034] Among them, the state estimation error is optimized, specifically: the heart rate signal spectrum corresponding to the heart rate of the target object is obtained through the fast Fourier transform algorithm, and based on the mapping relationship between the obtained abnormal state estimation interference value deviation and the filter cutoff frequency reduction value set in the database, the actual filter cutoff frequency reduction value is obtained to reduce the cutoff frequency of the filter, so as to reduce the interference of the low-frequency band of noise on the abnormal state estimation. The low-frequency band of noise is obtained from the heart rate signal spectrum and is used to reflect the low-frequency noise interference state corresponding to the breathing or movement of the target object. The abnormal state estimation interference value deviation represents the difference between the acquired abnormal state estimation interference value and the abnormal state estimation interference value set in the database. The set abnormal state estimation interference value is represented by the result of summing and averaging the historical abnormal state estimation interference values ​​in the historical abnormal state estimation process in the database.

[0035] In this embodiment, within the heart rate signal spectrum, the low-frequency noise band refers to that portion of the frequency range below the frequency range corresponding to the target subject's normal heart rate. Specifically, the heart rate signal itself has a typical frequency range (for example, a normal adult heart rate is approximately 60-100 beats / minute, corresponding to a frequency of 1-1.67 Hz). However, interference such as breathing (approximately 12-20 beats / minute when at rest, corresponding to a frequency of 0.2-0.33 Hz) and movement (such as body shaking and muscle tremors, which typically have lower frequencies, often in the 0-1 Hz range) have significantly lower frequencies than the heart rate signal.

[0036] Therefore, after conversion to the frequency domain through the fast Fourier transform algorithm, these low-frequency noise interferences will be concentrated in the area of ​​the spectrum where the frequency is lower than the frequency range corresponding to the normal heart rate. This area is defined as the "noise low-frequency band", which is manifested as: the frequency is lower than the main frequency band of the heart rate signal, and it specifically reflects the spectrum part of low-frequency interference such as breathing and exercise.

[0037] The judgment link in this example uses the interference value set in the database as the standard, and can quickly and accurately determine whether the abnormal state estimation process is qualified. If it is qualified, cumulative analysis will provide data support for subsequent research; if it is unqualified, optimization will be triggered in time. When optimizing the state estimation error, the fast Fourier transform is used to obtain the heart rate signal spectrum, accurately locate the low-frequency band of noise, and then based on the mapping relationship between the abnormal state estimation interference value deviation and the filter cutoff frequency reduction value, the filter cutoff frequency is reasonably reduced, effectively weakening low-frequency noise interference such as breathing or movement, and improving the accuracy and reliability of abnormal heart rate state estimation.

[0038] What needs to be further understood is that the state estimation error optimization also includes: the decision factor optimization of the abnormal state identification model, specifically: adjusting the decision factor ratio in the abnormal state identification model based on the obtained decision factor adjustment amplitude through the adaptive weight allocation algorithm, the decision factor adjustment amplitude represents the summed average result of the actual decision factor obtained by mapping and the current decision factor, and the actual decision factor is obtained by the mapping relationship between the obtained abnormal state estimation interference value deviation and the decision factor ratio set in the database; after the state estimation error is optimized, the abnormal state identification model is prompted to re-estimate the abnormal state and re-obtain the abnormal state estimation interference value. If the re-obtained abnormal state estimation interference value is greater than the set abnormal state estimation interference value, an abnormal state estimation warning is issued, otherwise the state estimation error optimization is completed and the abnormal state data is obtained for cumulative analysis.

[0039] In this embodiment, if the actual decision factor is greater than the current decision factor, its proportion in the abnormal state identification model is increased by the decision factor adjustment margin; if the actual decision factor is less than the current decision factor, its proportion is reduced by the adjustment margin. An adaptive weight allocation algorithm dynamically adjusts the decision factor proportions based on this adjustment margin. This adjustment method allows for dynamic model optimization based on real-time interference conditions. Increasing the proportion of effective decision factors improves the ability to capture key features, while reducing the proportion of ineffective factors reduces interference, enabling the model to more accurately identify abnormal states, enhancing adaptability and robustness, and improving the accuracy and reliability of anomaly detection.

[0040] It should be noted that abnormal state data refers to "true abnormal data" filtered through preprocessing and verification mechanisms, that is, data that excludes noise and misjudgments. For example, in physiological signal monitoring (such as electrocardiogram and blood oxygen), algorithms (such as threshold methods and machine learning classifiers) are first used to preliminarily identify abnormal events (such as arrhythmias and sudden drops in blood oxygen). Secondary verification (such as combining clinical annotation and multi-sensor cross-validation) is then used to eliminate "false anomalies" caused by device interference and individual physiological fluctuations (such as brief exercise). The final data retained is "estimated qualified abnormal state data," which is typically time series data, including the timestamp of the abnormality and a quantitative value of the abnormality level (such as the deviation of the abnormal heart rate).

[0041] It's important to understand that in long-distance driving monitoring scenarios, the interference effect of circadian rhythms on abnormal state identification is particularly significant. For example, between 2 and 4 a.m., when the body reaches its cortisol trough and melatonin secretion peaks, physiological signal baselines undergo systematic shifts. For example, EEG alpha wave power naturally increases, and heart rate variability decreases. If the decision-making mechanism of the abnormal state identification model overly relies on circadian rhythm factors (for example, with an initial weight of 40%), these normal rhythm fluctuations may be misinterpreted as signs of fatigue, leading to a high false alarm rate.

[0042] Furthermore, the acquired abnormal state data is cumulatively analyzed using a trend prediction algorithm. The specific process is as follows: the cumulative analysis period is divided (e.g., 10 minutes), and the changing trend of the physiological abnormality cumulative index at each cumulative analysis moment (t1, t2...t10) within the divided cumulative analysis period is monitored in real time. The changing trend line is fitted using the least squares method to obtain a fitted trend line. The abnormal state trend slope is obtained from the fitted trend line to reflect the evolution rate of the abnormal state of the target object. The physiological abnormality cumulative index is used to quantify the weighted cumulative intensity of the abnormal state of the target object corresponding to the cumulative analysis period. If the acquired abnormal state trend slope is greater than 0, it indicates that the target object's abnormal state cumulative intensity has increased (e.g., the heart rate is 30 beats / minute faster at time t4). If the acquired abnormal state trend slope is less than 0, it indicates that the target object's abnormal state cumulative intensity has decreased.

[0043] In this embodiment, when the slope of the acquired abnormal state trend equals 0, it indicates that within the divided cumulative analysis period, the target subject's physiological abnormality cumulative index at each cumulative analysis time (t1, t2, ..., t10) exhibits a horizontal trend. In other words, the cumulative intensity of the target subject's abnormal state has not shown a trend of continuous increase or decrease. Specifically, a fitted trend line with a slope of 0 is a horizontal line, indicating that no systematic upward or downward trend has formed within the cumulative analysis period. In this case, the target subject's abnormal state is in a dynamic equilibrium, with its cumulative intensity neither increasing nor decreasing, but rather maintaining a stable state.

[0044] The trend prediction algorithm involved in this example typically uses a time series analysis method, which can be combined with a sliding window mechanism to dynamically analyze physiological signals. The physiological abnormality cumulative index is calculated using the data within the window to capture the abnormal state trend changes of the target object. Time series models (such as ARIMA / LSTM) are then used to predict future physiological abnormality risks. The physiological abnormality cumulative index is a weighted sum of various abnormal state indicators of the target object during the cumulative analysis period, quantifying the weighted cumulative intensity of its abnormal state.

[0045] Compared with existing technologies, this example can comprehensively and meticulously capture the dynamic evolution of the abnormal state of the target object over time by dividing the specific cumulative analysis period and monitoring the changes in the cumulative index of physiological abnormalities in real time, thereby avoiding misjudgments due to instantaneous data fluctuations. Secondly, the least squares method is used to fit the changing trend line and obtain the abnormal state trend slope, reflecting the abnormal state evolution rate in an intuitive and quantitative manner, providing a strong basis for accurately judging the development trend of the abnormal state. When the abnormal state trend slope is greater than 0, the increase in the cumulative intensity of the abnormal state can be detected in time, and intervention measures can be taken in advance; when the abnormal state trend slope is less than 0, it can be confirmed that the intervention measures are effective or the abnormality has resolved itself, thereby improving the accuracy and foresight of abnormal state monitoring.

[0046] like Figure 3 As shown, it is a flowchart of the quantification and error optimization judgment of the abnormal state cumulative analysis process provided by the embodiment of the present application. Its specific logic is: first, it is determined whether the actual time period is less than or equal to the set time period. If it is satisfied, the current window parameters are determined to be valid; if not, the cumulative analysis error is optimized, and the cumulative analysis is repeated after adjusting the window length and time step. Then, it is determined again whether the new actual time period is greater than the set time period. If so, the cumulative analysis warning is triggered. Otherwise, the cumulative analysis error optimization is completed and the fatigue level identification is carried out. This logic is intended to ensure the accuracy of the cumulative analysis, timely warn of abnormalities and reasonably identify the fatigue level.

[0047] What needs to be further understood is that the specific process of cumulative analysis error optimization judgment is as follows: record the actual processing time of the abnormal state trend slope, and compare and analyze it with the processing time set in the database: if the actual processing time obtained by the record is not greater than the processing time set in the database, it is determined that the current window parameters (such as window length, time step, etc.) are valid; if the actual processing time obtained by the record is greater than the processing time set in the database, it is determined that there is a deviation in the current window parameters and cumulative analysis error optimization is performed.

[0048] Among them, the cumulative analysis error is optimized, and the specific process is as follows: the window length adjustment value and time step adjustment value obtained by mapping the actual processing time deviation in the database through the trend prediction algorithm are used to adjust the current window length and time step respectively, so as to reduce the signal strength interference while improving the timeliness of capturing the physiological state of the target object. The actual processing time deviation represents the difference between the actual processing time obtained and the set processing time. The set processing time is represented by the sum and average of the acquisition time periods of the historical abnormal state trend slope in the historical cumulative analysis process in the database; after the cumulative analysis error is optimized, the trend prediction algorithm is used to re-perform the cumulative analysis, and the actual processing time of the corresponding abnormal state trend slope is re-obtained. If the re-obtained actual processing time is greater than the set processing time, a cumulative analysis warning is issued, otherwise the cumulative analysis error optimization is completed and the fatigue level is identified.

[0049] In this embodiment, actual processing time refers to the actual time spent calculating the abnormal state trend slope for the corresponding sliding window after the cumulative analysis passes. Specifically, it is the total time from capturing the physiological abnormality cumulative index data in the sliding window, through data collation and trend line fitting, to the final calculation of the abnormal state trend slope. It directly reflects the efficiency of the system in processing the data within the sliding window and generating the trend slope. Comparing it with the processing time set in the database can determine whether the window parameters are suitable for the current data processing needs.

[0050] The mapped window length adjustment value and time step adjustment value are essentially used to quantify the specific adjustment range of the current window length and time step. Among them, the window length adjustment value is a specific value mapped from the database based on the actual processing time deviation and used to correct the current window length, which directly reflects the adjustment range of the window length; the time step adjustment value is a specific value mapped from the database based on the actual processing time deviation and used to correct the current sampling interval, which reflects the adjustment range of the sampling frequency.

[0051] Because the actual processing time obtained is longer than the set processing time, indicating a slower data rate, the mapped window length adjustment value is positive. Increasing the current window length by this value covers a longer period of data and captures a more comprehensive trend. Similarly, the time step adjustment value is also positive, increasing the time step by this value reduces computational effort. Compared to existing technologies, this method dynamically adapts to data changes, precisely adjusts window parameters, and reduces errors. Through multiple analyses and assessments, data reliability is ensured, improving the accuracy of abnormal state trend analysis, providing a solid foundation for fatigue level identification and effectively avoiding misjudgments.

[0052] like Figure 4As shown, the hierarchical warning and feedback determination flow chart of the fatigue level identification process provided by the embodiment of the present application has the following design logic: First, continuously monitor the changes in the slope of the abnormal state trend and the duration. Then determine whether the trend slope and duration are both 0. If so, it indicates that there is no warning and feedback requirement. Otherwise, further determine whether the slope or duration is 0. If so, it means that there is an error in the identification and manual intervention is required. Otherwise, different levels of warnings are performed based on the range of the slope, which are divided into level 1, level 2, level 3, and level 4 warnings. This logic accurately identifies abnormal states and takes corresponding measures through step-by-step judgment.

[0053] What needs to be further understood is that the specific process of determining whether there are different levels of warning and feedback needs is as follows: real-time monitoring of the changes in the abnormal state trend slope during the fatigue level identification process, and statistics on the duration of the abnormal state trend slope during the fatigue level identification process, that is, the total duration of the sliding window collecting the abnormal state trend slope in the non-zero state, helps to combine the amplitude of the abnormal state trend slope change and divide the warning levels accordingly to ensure that different levels of fatigue status can be fed back in a timely manner, thereby improving the targeted monitoring response; based on the obtained abnormal state trend slope and duration, different levels of warning and feedback needs are determined: if the obtained abnormal state trend slope and duration are both 0, it is determined that there are no different levels of warning and feedback needs; if one of the obtained abnormal state trend slope and duration is 0, it is determined that the fatigue level identification is incorrect and the preset personnel are prompted to intervene; if the obtained abnormal state trend slope and duration are both not 0, it is determined that there are different levels of warning and feedback needs.

[0054] Among them, if the obtained abnormal state trend slope and duration are both not 0, it is determined that there are different levels of warning and feedback needs, specifically: if the obtained abnormal state trend slope is within the first abnormal state trend slope range (usually set between 0 and 0.3, including the case where it is equal to 0.3), a first-level warning (mild fatigue) is triggered and a first-level reminder measure is taken; if the obtained abnormal state trend slope is within the second abnormal state trend slope range (usually set between 0.3 and 0.6, including the case where it is equal to 0.6), a second-level warning (moderate fatigue) is triggered and a second-level reminder measure is taken; if the obtained abnormal state trend slope is within the third abnormal state trend slope range (usually set between 0.6 and 1, excluding the case where it is equal to 1), a third-level warning (high fatigue) is triggered and a third-level reminder measure is taken; if the obtained abnormal state trend slope is within the fourth abnormal state trend slope range (usually set to be greater than or equal to 1), a fourth-level warning (extreme fatigue) is triggered and a fourth-level reminder measure is taken.

[0055] What needs to be further understood is that if one of the obtained abnormal state trend slope and duration is 0, it indicates that there is significant asymmetry in the data states of the two, and an effective fatigue state assessment logical closed loop cannot be formed. Specifically: when the abnormal state trend slope is 0 but the duration is not 0, it means that the physiological abnormality accumulation index has no fluctuation during the monitoring period, and the existence of the duration makes the system still record data. This contradiction may be caused by sensor signal drift or excessive filtering of the data cleaning algorithm, resulting in the trend characteristics being incorrectly erased; on the contrary, when the duration is 0 and the abnormal state trend slope is not 0, it reflects that the system has output the trend calculation result before completing the effective monitoring cycle (such as the period being truncated due to sampling interruption). At this time, the abnormal state trend slope only represents the local instantaneous change rather than the real fatigue development trend. If either value is 0, it indicates that there are defects in data integrity and analysis timeliness, which cannot support the hierarchical early warning mechanism's continuous judgment of the fatigue degree from "quantitative change to qualitative change". Therefore, it is necessary to prioritize the identification of fatigue degree errors and trigger correction processes such as data re-collection, algorithm backtracking or manual review to ensure that the abnormal state trend slope and duration strictly coexist in the time dimension, providing a reliable basis for subsequent early warning classification.

[0056] For example, in scenarios where doctors work for long periods of time, the above-mentioned hierarchical warning feedback mechanism can also be used to ensure surgical safety. For example, during a consultation with a sudden increase in patients, or during a multi-doctor collaborative operation, under the premise that the surgical process is controllable, appropriate rest can be taken according to the reminder of the wearable device. Specifically: if the slope of the abnormal state trend obtained is between 0 and 0.3, the system will pop up a gentle prompt corresponding to the first-level warning through the operating room display, such as "Doctor, the operation has been going on for a while. Proper adjustment of the rhythm can improve the accuracy of the operation." At the same time, a gentle prompt tone will be issued to remind the doctor to pay attention to his or her own status; if the slope of the abnormal state trend obtained is between 0.3 and 0.6, a gentle prompt corresponding to the second-level warning will pop up on the display, such as "Doctor, you are moderately fatigued. It is recommended to take a short break to restore your energy." The operating room audio will play a more rapid prompt tone and send a message to the nurse station outside the operating room. The nurse can enter and remind the doctor in time.

[0057] If the slope of the abnormal trend is between 0.6 and 1, a mild warning corresponding to the third level warning will pop up on the display screen. This means the warning lights in the operating room will flash, an emergency alert text message will be sent to the doctor's mobile phone, and the surgical team leader will be notified, who will then arrange for other doctors to assist or take over the operation. If the slope of the abnormal trend is greater than or equal to 1, a fourth level warning (extreme fatigue) will be triggered. The display will also pop up a mild warning corresponding to the fourth level warning, which means the operation rights of some high-risk surgical equipment will be restricted, and a surgical suspension plan will be automatically planned. Hospital management will be notified to ensure that the doctor can stop the operation in time and get enough rest, avoiding surgical errors caused by fatigue and ensuring patient safety and surgical quality.

[0058] In this embodiment, by monitoring the slope and duration of abnormal state trends during long doctor shifts, doctors' fatigue levels can be accurately assessed. Based on the severity of these conditions, corresponding warning levels are triggered and targeted measures are implemented: Level 1 provides gentle reminders to alert doctors to initial fatigue; Level 2 reinforces the warning through multiple reminders; Level 3 implements mandatory intervention to avoid surgical risks; and Level 4 initiates emergency treatment. This hierarchical and diverse warning feedback mechanism can effectively prevent medical accidents caused by fatigue among medical staff.

[0059] An embodiment of the present application provides an intelligent fatigue state assessment and monitoring device based on multimodal sensing, which may include but is not limited to: an electrocardiogram (ECG) sensor, an electroencephalogram (EEG) sensor, a timer, a data center storage device, and a server; wherein the ECG sensor is used to monitor the heart rate, heart rate variability, and heart rate signal spectrum of the target object; the EEG sensor is used to monitor the EEG alpha wave power and EEG delta wave power of the target object; the timer is used to monitor the maximum tolerance time switching delay, the data sampling frequency switching delay, the actual processing time, and the duration; the data center storage device is used to store data during the intelligent fatigue state assessment and monitoring process, including but not limited to the maximum tolerance time switching delay, the data sampling frequency switching delay, and historical data in the database, including but not limited to the set abnormal state estimated interference value and the set processing time; the server is used to run the abnormal state recognition model, the trend prediction algorithm, and the hierarchical early warning mechanism.

[0060] It is necessary to add that, if Figure 5 As shown, this is the homepage of the interface of the intelligent fatigue status assessment and monitoring system based on multimodal sensing provided by the embodiment of the present application. The left side is a navigation bar, covering functional options such as abnormal status assessment, cumulative analysis, status identification, fatigue monitoring, data analysis and system settings; the top of the homepage displays the current user information, and the bottom presents key data such as wearable device status, heart rate, EEG rhythm and head posture in the form of cards, with status descriptions and suggestions next to each indicator; the middle part is the fatigue analysis curve, which shows the trend of fatigue time changes in the past 30 minutes; the bottom is equipped with optimization prompts and early warning prompts areas to provide timely feedback on abnormalities and provide processing suggestions. The system status bar displays the operating status and data synchronization status, which makes it convenient for users to grasp the system dynamics in real time and effectively helps medical staff to accurately monitor and intervene in the user's fatigue status.

[0061] Among them, in terms of fatigue monitoring, it may include continuous tracking of changes in user fatigue levels over time and special monitoring of fatigue status in specific scenarios; data analysis will conduct in-depth mining of various types of collected physiological, behavioral and other data, such as analyzing the relationship between fatigue and various indicators; system settings allow users to personalize monitoring parameters, reminder methods, data synchronization frequency, etc. according to their own needs to improve the user experience and monitoring effect.

[0062] like Figure 6 As shown, the fatigue state identification interface of the intelligent fatigue state assessment and monitoring system based on multimodal sensing provided by the embodiment of the present application is shown. The main part of the interface presents various types of data in multiple sections: the physiological signal section displays key indicators such as heart rate variability and brain wave proportion; the behavioral characteristics section records behavioral data such as the number of blinks and head deviation angle; the environmental data section displays environmental information such as temperature and noise; in addition, it also presents the physiological data, connection status and fatigue analysis results of the wearable device; the current fatigue level and warning level are clearly given below, and targeted warning prompts are provided. By integrating multi-source data, this interface provides strong support for medical staff or users themselves to fully and intuitively understand the fatigue state, facilitating timely response measures.

[0063] The above embodiments can be implemented in whole or in part via software, hardware (e.g., circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions or computer programs. When loaded or executed on a computer, the processes or functions described in the embodiments of this application are fully or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired means (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server or data center that contains a collection of one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0064] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interface, indirect coupling or communication connection of the device or unit, which can be electrical, mechanical or other forms.

[0065] If the functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or the portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage media include various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical disks.

[0066] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. Intelligent fatigue status assessment and monitoring system based on multimodal sensing, characterized by: include: Abnormal state estimation module, abnormal state accumulation analysis module and fatigue level identification module; The abnormal state estimation module is used to obtain state monitoring data of the target object through the multimodal sensor in the wearable device, and input it into the established abnormal state recognition model to perform abnormal state estimation. At the same time, the abnormal state estimation process is analyzed for its impact based on the acquired state estimation impact data to determine whether to optimize the state estimation error based on the acquired heart rate signal spectrum to improve the accuracy of abnormal state estimation under noise interference. The state estimation impact data includes the maximum tolerated time switching delay and the data sampling frequency switching delay. The abnormal state accumulation analysis module is used to obtain the abnormal state data output by the abnormal state recognition model and perform cumulative analysis when the abnormal state estimation process is qualified, and at the same time perform cumulative analysis error optimization judgment to improve the matching degree between the sliding window parameters and the actual change characteristics of the physiological signal; The fatigue level identification module is used to perform fatigue level identification for a preset identification period based on the results of the cumulative analysis when the cumulative analysis is qualified, so as to determine whether to perform different levels of warning and feedback requirements based on the duration of the abnormal state change corresponding to the fatigue level identification process, thereby improving the timeliness of triggering different levels of warnings corresponding to different identification periods.

2. The intelligent fatigue state assessment and monitoring system based on multimodal sensing according to claim 1, characterized in that: The influence degree analysis of the abnormal state estimation process based on the acquired state estimation influence data is performed as follows: Obtaining the response status of the maximum tolerance duration and the data sampling frequency during the abnormal state estimation process to obtain the maximum tolerance duration switching delay and the data sampling frequency switching delay, wherein the maximum tolerance duration switching delay represents the time interval required to complete the current maximum tolerance duration switching when the maximum tolerance duration required for abnormal state estimation changes, and the data sampling frequency switching delay represents the time interval required to complete the current data sampling frequency switching when the data sampling frequency required for abnormal state estimation changes; The acquired maximum tolerance duration switching delay and the data sampling frequency switching delay are harmonically averaged to obtain an abnormal state estimation interference value, which represents quantitative data on the degree of influence of the acquired state estimation influence data on the accuracy of abnormal state estimation.

3. The intelligent fatigue state assessment and monitoring system based on multimodal sensing according to claim 2, characterized in that: The specific process of determining whether to perform state estimation error optimization based on the acquired heart rate signal spectrum is as follows: If the obtained abnormal state estimation interference value is not greater than the abnormal state estimation interference value set in the database, the abnormal state estimation process is determined to be qualified and cumulative analysis is performed; otherwise, the abnormal state estimation process is determined to be unqualified and state estimation error optimization is performed; The state estimation error optimization is specifically as follows: a heart rate signal spectrum corresponding to the heart rate of the target object is obtained through a fast Fourier transform algorithm; based on a mapping relationship between the obtained abnormal state estimation interference value deviation and a filter cutoff frequency reduction value set in a database, an actual filter cutoff frequency reduction value is obtained to reduce the filter cutoff frequency, thereby reducing the interference of a low-frequency noise band on the abnormal state estimation. The low-frequency noise band is obtained from the heart rate signal spectrum and is used to reflect the low-frequency noise interference state corresponding to the target object's breathing or movement.

4. The intelligent fatigue state assessment and monitoring system based on multimodal sensing according to claim 3, characterized in that: The state estimation error optimization also includes: decision factor optimization of the abnormal state recognition model, specifically: Adjusting the proportion of decision factors in the abnormal state recognition model based on the obtained decision factor adjustment amplitude through an adaptive weight allocation algorithm, wherein the decision factor adjustment amplitude represents the summed average result of the mapped actual decision factor and the current decision factor, wherein the actual decision factor is obtained by mapping the obtained abnormal state estimated interference value deviation to the decision factor proportion set in the database; After the state estimation error is optimized, the abnormal state identification model is prompted to re-estimate the abnormal state and re-obtain the abnormal state estimation interference value. If the re-obtained abnormal state estimation interference value is greater than the set abnormal state estimation interference value, an abnormal state estimation warning is issued. Otherwise, the state estimation error optimization is completed and the abnormal state data is obtained for cumulative analysis.

5. The intelligent fatigue state assessment and monitoring system based on multimodal sensing according to claim 1, characterized in that: The specific process of the cumulative analysis is as follows: Real-time monitoring of the changing trend of the physiological abnormality cumulative index at each cumulative analysis moment within the divided cumulative analysis period to obtain a fitted trend line, and obtaining the abnormal state trend slope from the fitted trend line to reflect the evolution rate of the abnormal state of the target object. The physiological abnormality cumulative index is used to quantify the weighted cumulative intensity of the abnormal state of the target object corresponding to the cumulative analysis period; If the obtained abnormal state trend slope is greater than 0, it indicates that the abnormal state cumulative intensity of the target object increases. If the obtained abnormal state trend slope is less than 0, it indicates that the abnormal state cumulative intensity of the target object decreases.

6. The intelligent fatigue state assessment and monitoring system based on multimodal sensing according to claim 5, characterized in that: The specific process of optimizing the cumulative analysis error is as follows: Record the actual processing time of the abnormal state trend slope and compare it with the processing time set in the database: If the actual processing time recorded is not greater than the processing time set in the database, the current window parameters are considered valid; If the actual processing time obtained from the record is longer than the processing time set in the database, it is determined that there is a deviation in the current window parameters and cumulative analysis error optimization is performed.

7. The intelligent fatigue state assessment and monitoring system based on multimodal sensing according to claim 6, characterized in that: The specific process of the cumulative analysis error optimization is as follows: Through the trend prediction algorithm, the window length adjustment value and time step adjustment value obtained by mapping the actual processing time deviation in the database are used to adjust the current window length and time step respectively; After the cumulative analysis error is optimized, the cumulative analysis is performed again through the trend prediction algorithm, and the actual processing time of the corresponding abnormal state trend slope is obtained again. If the actual processing time obtained again is greater than the set processing time, a cumulative analysis warning is issued; otherwise, the cumulative analysis error is optimized and fatigue degree identification is performed.

8. The intelligent fatigue state assessment and monitoring system based on multimodal sensing according to claim 6, characterized in that: The specific process of carrying out early warning and feedback requirements at different levels is as follows: Monitor the changes in the abnormal state trend slope during the fatigue level identification process in real time, and calculate the duration of the abnormal state trend slope during the fatigue level identification process; Determine the need for early warning and feedback at different levels based on the acquired abnormal state trend slope and duration: If the obtained abnormal state trend slope and duration are both 0, it is determined that there is no warning and feedback demand at different levels; If either the slope or duration of the abnormal state trend is 0, it is determined that the fatigue level is incorrectly identified and the preset personnel are prompted to intervene; If the obtained abnormal state trend slope and duration are not 0, it is determined that there are different levels of warning and feedback needs.

9. The intelligent fatigue state assessment and monitoring system based on multimodal sensing according to claim 8, characterized in that: If the obtained abnormal state trend slope and duration are not 0, it is determined that there are different levels of warning and feedback needs, specifically: If the acquired abnormal state trend slope is within the first abnormal state trend slope range, a first-level warning is triggered and first-level reminder measures are taken; If the acquired abnormal state trend slope is within the second abnormal state trend slope range, a second-level warning is triggered and second-level reminder measures are taken; If the obtained abnormal state trend slope is within the third abnormal state trend slope range, a third-level warning is triggered and third-level reminder measures are taken; If the acquired abnormal state trend slope is within the fourth abnormal state trend slope range, a fourth-level warning is triggered and fourth-level reminder measures are taken.

10. An intelligent fatigue state assessment and monitoring device based on multimodal sensing, applying the intelligent fatigue state assessment and monitoring system based on multimodal sensing according to any one of claims 1 to 9, characterized in that: include: ECG sensors, timers, and servers; The electrocardiogram sensor is used to monitor the heart rate signal spectrum; The timer is used to monitor the maximum tolerated switching delay, the data sampling frequency switching delay, the actual processing time and the duration; The server is used to run an abnormal state recognition model and a hierarchical early warning mechanism.

Images