Fatigue detection and early warning method based on multi-modal physiological signal fusion

Abstract

The invention relates to the technical field of fatigue detection and early warning, in particular to a multi-modal physiological signal fused fatigue detection and early warning method which comprises the steps that multi-modal physiological signals of a user are obtained, and the multi-modal physiological signals comprise electroencephalogram signals, heart rate variability signals, skin conductance signals and accelerometer signals; extracting multi-scale time sequence features based on the multi-modal physiological signals; constructing a personalized fatigue model according to the multi-scale time sequence characteristics; determining a fatigue state of the user based on the personalized fatigue model; generating fatigue early warning information according to the fatigue state; fatigue early warning information is output, and the accuracy and robustness of fatigue detection are greatly improved through the multi-modal signal fusion and multi-scale feature extraction technology. By simultaneously analyzing the electroencephalogram, the heart rate, the skin conductance and the acceleration, the method can comprehensively capture various expressions of the fatigue state. And fatigue modes on different time scales can be effectively identified through multi-scale feature extraction, so that the fatigue degree can be evaluated more accurately.

Description

Technical Field

[0001] This invention relates to the field of fatigue detection and early warning technology, and more specifically, to a fatigue detection and early warning method based on the fusion of multimodal physiological signals. Background Technology

[0002] With the accelerating pace of modern work and the increasing complexity of tasks, fatigue has become a significant factor affecting work efficiency, quality of life, and even safety. Particularly in high-risk industries such as transportation, healthcare, and industrial production, the fatigue status of operators directly impacts public safety. Therefore, accurate detection and timely early warning of fatigue have become a hot research topic.

[0003] Traditional fatigue detection methods primarily rely on single physiological signals or behavioral indicators. For example, some methods assess fatigue levels by analyzing a driver's blink frequency and pupil size; others utilize energy changes in specific frequency bands of electroencephalography (EEG) to evaluate fatigue status. While these methods have proven effective in specific scenarios, they also have significant limitations. First, a single signal source is susceptible to environmental interference and individual differences, leading to unstable test results. Second, these methods often only reflect instantaneous fatigue states, failing to capture the cumulative effects and dynamic changes of fatigue.

[0004] In recent years, with the development of multi-sensor technology and artificial intelligence algorithms, researchers have begun to explore the integration of multiple physiological signals to improve the accuracy and reliability of fatigue detection. For example, some studies have combined electroencephalogram (EEG), electrocardiogram (ECG), and eye-tracking data to assess driver fatigue levels. While this multimodal approach has indeed improved detection accuracy to some extent, several problems remain to be solved.

[0005] First, existing multimodal methods often simply splice or average various signal features without fully considering the interactions and temporal dependencies between different signals. This leads to information redundancy and loss, failing to fully leverage the advantages of multimodal data. Second, most methods employ fixed model structures and parameters, making it difficult to adapt to differences between individuals and changes in the state of the same individual at different times and under different tasks. Furthermore, existing methods primarily focus on fatigue detection, with limited research on fatigue early warning and intervention measures, hindering proactive prevention and timely intervention.

[0006] Furthermore, sleep quality, a crucial factor influencing fatigue, is often overlooked in existing fatigue detection systems. Most methods focus solely on immediate work performance, neglecting the long-term impact of sleep, a vital recovery mechanism, on fatigue.

[0007] Finally, existing methods generally lack the ability to analyze the causes of fatigue. They can determine whether a user is fatigued, but cannot pinpoint the specific reasons for the fatigue, which limits the targeting and effectiveness of subsequent interventions.

[0008] In view of the above problems, there is an urgent need for a method that can comprehensively, accurately, and personally detect and warn of fatigue. This method should be able to effectively integrate multiple physiological signals, capture characteristic changes at different time scales, adapt to individual differences and dynamic changes in condition, analyze the causes of fatigue, and provide timely and effective early warnings and intervention suggestions. Summary of the Invention

[0009] This invention addresses the aforementioned technical problems by proposing a fatigue detection and early warning method based on multimodal physiological signal fusion. This method achieves effective integration of multiple physiological signals through innovative multi-scale temporal feature extraction and fusion techniques. By introducing personalized modeling and online learning mechanisms, this method can adapt to the characteristics and state changes of different users. Furthermore, this invention considers the impact of sleep quality on fatigue and introduces a causal reasoning module to analyze the causes of fatigue.

[0010] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: Fatigue detection and early warning methods based on multimodal physiological signal fusion include: The acquisition steps include: Acquire the user's multimodal physiological signals, including electroencephalogram (EEG) signals, heart rate variability signals, skin conductance signals, and accelerometer signals; The processing steps include: Multi-scale temporal features are extracted based on multimodal physiological signals; Based on multi-scale temporal characteristics, a personalized fatigue model is constructed; Based on a personalized fatigue model, determine the user's fatigue state; Output steps, including: Based on fatigue status, generate fatigue early warning information; Output fatigue warning information.

[0011] Preferably, the extraction of multi-scale temporal features specifically includes: The multimodal physiological signals are processed using a multi-scale temporal feature extraction network; The parameters of the multi-scale temporal feature extraction network are optimized based on the meta-learning framework; The extracted features are weighted and fused using an attention mechanism.

[0012] Preferably, the multi-scale temporal feature extraction network includes: CNN layers are used to encode EEG signals into multichannel features; Multiple BiLSTM layers are used to capture the temporal dependencies of EEG sequences; A multi-scale structure is used to combine multi-scale feature matrices from the outputs of different layers of BiLSTM.

[0013] Preferably, the construction of the personalized fatigue model specifically includes: Obtain user's individual characteristics, task characteristics, and data characteristics; Based on the individual characteristics, task characteristics, and data characteristics, the multi-scale temporal features are individually selected and fused; Based on the results of the personalized selection and fusion, a personalized fatigue model for the user is generated.

[0014] Preferably, the personalized fatigue model includes: A multi-scale feature selection module is used to select different temporal feature channels based on an attention mechanism; A multi-scale feature fusion module is used to fuse temporal features and frequency domain features; The personalized fatigue model module is used to output the final fatigue state.

[0015] As a preferred option, it also includes: Based on the causal reasoning module, the key factors leading to user fatigue are analyzed. Based on the key factors mentioned above, personalized intervention recommendations are generated.

[0016] As a preferred option, it also includes: The parameters of the personalized fatigue model are updated in real time through an online learning mechanism; Based on user feedback, the performance of the personalized fatigue model was optimized.

[0017] Preferably, the generation of fatigue warning information specifically includes: The warning level is determined based on the fatigue state and the risk level of the current task; Based on the aforementioned warning level, corresponding warning information and intervention measures are generated.

[0018] Preferably, determining the warning level includes: Calculate the distance from the fatigue state to the center point of the fatigue state; The warning level is determined based on the comparison between the distance and the preset threshold.

[0019] As a preferred option, it also includes: Obtain user's sleep data; Based on the sleep state data, the parameters of the personalized fatigue model are adjusted.

[0020] The method of the present invention has the following significant technical effects: First, multimodal signal fusion and multi-scale feature extraction techniques significantly improve the accuracy and robustness of fatigue detection. By simultaneously analyzing multiple physiological signals such as EEG, heart rate, skin conductance, and acceleration, this method can comprehensively capture various manifestations of fatigue. Multi-scale feature extraction can effectively identify fatigue patterns at different time scales, thereby more accurately assessing the degree of fatigue.

[0021] Secondly, personalized modeling and online learning mechanisms enable this method to adapt to individual differences and dynamic changes in the states of different users. This not only improves the accuracy of detection but also allows the system to continuously optimize with use, providing users with increasingly precise services.

[0022] Furthermore, this method can not only detect fatigue but also provide early warnings. By analyzing the changing trends of fatigue status and the risk level of the current task, the system can issue timely warnings, effectively preventing fatigue-related accidents.

[0023] Furthermore, this invention incorporates sleep quality assessment, taking long-term factors into account, which makes fatigue assessment more comprehensive and accurate. By adjusting model parameters to adapt to different sleep conditions, the system can better predict and manage daytime fatigue.

[0024] Finally, the introduction of the causal reasoning module enables this method not only to detect fatigue but also to analyze the specific causes of fatigue. This provides a clear direction for subsequent intervention measures and greatly improves the effectiveness of fatigue management.

[0025] In summary, the fatigue detection and early warning method based on multimodal physiological signal fusion provided by this invention achieves accurate detection, timely early warning, and effective intervention of user fatigue status through multi-dimensional data acquisition, intelligent feature extraction and fusion, personalized modeling and early warning, and continuous online learning and optimization. This comprehensive, accurate, and personalized fatigue management solution has significant practical implications for improving work efficiency, reducing accident risks, and enhancing quality of life. Attached Figure Description

[0026] Figure 1 This is an overall flowchart of the method of the present invention.

[0027] Figure 2 This is a flowchart of the extraction of multi-scale temporal features according to the present invention.

[0028] Figure 3 This is a flowchart illustrating the construction of a personalized fatigue model according to the present invention.

[0029] Figure 4 This is a flowchart illustrating the generation of fatigue warning information according to the present invention. Detailed Implementation

[0030] like Figure 1-4 As shown, this invention provides a fatigue detection and early warning method based on multimodal physiological signal fusion. This method achieves accurate detection and timely early warning of user fatigue by fusing multiple physiological signals. The specific embodiments of this invention will be described in detail below.

[0031] First, the method of the present invention includes an acquisition step, a processing step, and an output step. In the acquisition step, the method acquires the user's multimodal physiological signals, including electroencephalogram (EEG) signals, heart rate variability (HRV) signals, skin conductance signals, and accelerometer signals. Preferably, portable devices such as EEG headsets and smartwatches can be used to acquire these physiological signals. For example, EEG signals can be acquired through EEG headsets at a sampling frequency of 1250 Hz, then preprocessed using a Butterworth filter with a cutoff frequency set to 0.5-45 Hz, and finally downsampled to 450 Hz.

[0032] In the processing steps, the method of this invention first extracts multi-scale temporal features based on the acquired multimodal physiological signals. This step is one of the core innovations of this invention. Through multi-scale feature extraction, changes in physiological signals at different time scales can be captured, thereby more comprehensively reflecting the user's fatigue state. Specifically, this invention employs a multi-scale temporal feature extraction network to process multimodal physiological signals. The network structure includes CNN layers, multiple BiLSTM layers, and a multi-scale structure.

[0033] CNN layers are primarily used to encode EEG signals into multi-channel features. For example, multiple convolutional kernels of different sizes (e.g., 3x3, 5x5, 7x7) can be used to extract spatial features at different scales. BiLSTM layers are used to capture the temporal dependencies of EEG sequences. By setting different numbers of hidden units (e.g., 64, 128, 256), temporal features of different granularities can be obtained. The multi-scale structure is used to combine the multi-scale feature matrices output from different BiLSTM layers to obtain a comprehensive feature representation.

[0034] It is worth noting that this invention also introduces a meta-learning framework to optimize the parameters of the multi-scale temporal feature extraction network. This method can effectively address the problems of large individual differences in physiological signals and limited labeled data. Specifically, model-independent meta-learning algorithms (such as MAML) can be used to quickly adapt to data from new users. For example, the inner loop learning rate can be set to 0.01, the outer loop learning rate to 0.001, and the task batch size to 5.

[0035] Furthermore, this invention also employs an attention mechanism to perform weighted fusion of the extracted features. A self-attention mechanism can be used here, the mathematical expression of which is as follows: in, , , These represent the query, key, and value matrices, respectively. This represents the dimension of the key vector. In this way, the model can adaptively focus on the importance of different features.

[0036] After obtaining multi-scale temporal features, the method of this invention constructs a personalized fatigue model based on these features. This step fully considers individual differences among users, thereby improving the accuracy of fatigue detection. Specifically, firstly, the method acquires the user's individual characteristics (such as age, gender, and occupation), task characteristics (such as task type and difficulty), and data characteristics (such as signal quality and acquisition time). Then, based on these characteristics, the multi-scale temporal features are personalized for selection and fusion.

[0037] For example, a multilayer perceptron can be used to learn the importance weights of features: in, , , These represent individual characteristics, task characteristics, and data characteristics, respectively. These are learnable parameters. After obtaining the weights w, weighted fusion of multi-scale temporal features can be performed.

[0038] Finally, based on a personalized fatigue model, this method determines the user's fatigue state. A multi-classification model, such as a softmax classifier, can be used to classify the user's fatigue state into multiple levels (e.g., mild fatigue, moderate fatigue, severe fatigue).

[0039] In the output step, the method of the present invention generates fatigue warning information based on the determined fatigue state and outputs it. The warning information may include fatigue level, recommended rest time, precautions, etc. For example, for mild fatigue, the user may be advised to take a short rest of 5-10 minutes; for moderate fatigue, the user may be advised to rest for 15-30 minutes and engage in some relaxation activities; for severe fatigue, the user may need to be advised to stop their current work and rest fully.

[0040] Through the above steps, the method of this invention achieves real-time monitoring and early warning of user fatigue status. Compared with traditional methods, this invention has the following advantages: First, the fusion of multimodal physiological signals provides more comprehensive fatigue status information; second, multi-scale temporal feature extraction can capture changes in physiological signals at different time scales; third, the personalized fatigue model considers individual differences among users, improving detection accuracy; and finally, the real-time early warning mechanism can promptly remind users and effectively prevent fatigue-related accidents. In a preferred embodiment of this invention, the process of constructing a personalized fatigue model is further refined. Specifically, this process includes acquiring the user's individual characteristics, task characteristics, and data characteristics; performing personalized selection and fusion of multi-scale temporal features based on these characteristics; and finally generating a personalized fatigue model for the user based on the results of personalized selection and fusion.

[0041] Preferably, individual characteristics may include the user's age, gender, occupation, and past medical history. For example, for a 35-year-old male programmer, his individual characteristic vector might be represented as [35,1,3,0], where 1 represents male, 3 represents the IT industry, and 0 represents no relevant medical history. Task characteristics may include the type, difficulty, and duration of the current task. For example, a high-difficulty programming task lasting 4 hours might be represented as [2,4,4], where 2 represents the programming task, 4 represents high difficulty, and 4 represents 4 hours. Data characteristics may include the time of signal acquisition, environmental conditions, and device status. For example, [14,2,1] might represent 2 PM, an indoor environment, and good device status.

[0042] The method of this invention performs personalized selection and fusion of previously extracted multi-scale temporal features based on these characteristics. The core of this step lies in adaptively adjusting the importance of different features according to the user's specific circumstances. For example, for older users, more emphasis may be placed on changes in low-frequency EEG signals; while for high-intensity tasks, greater attention may be paid to fluctuations in heart rate variability.

[0043] In one embodiment of the invention, personalized selection and fusion can be achieved through an attention mechanism. Specifically, a multilayer perceptron can be used to learn the importance weights of features: in, , , These represent individual characteristics, task characteristics, and data characteristics, respectively. These are learnable parameters. In this way, the model can automatically adjust its focus on different temporal features based on different combinations of features.

[0044] After obtaining the weights w, weighted fusion of multi-scale temporal features can be performed: in, Represents the i-th time series feature. This represents the corresponding weight, where n is the total number of features.

[0045] In another preferred embodiment of the invention, the personalized fatigue model includes a multi-scale feature selection module, a multi-scale feature fusion module, and a personalized fatigue model module. The multi-scale feature selection module selects different temporal feature channels based on an attention mechanism. This attention mechanism can be the self-attention mechanism mentioned earlier, or other variations, such as a multi-head attention mechanism. The multi-scale feature fusion module is responsible for fusing temporal and frequency domain features. For example, a convolutional neural network can be used to extract joint features from the temporal and frequency domains. Finally, the personalized fatigue model module outputs the final fatigue state based on the fused features.

[0046] Preferably, the method of the present invention further includes analyzing key factors leading to user fatigue based on a causal reasoning module, and generating personalized intervention suggestions based on these key factors. The introduction of the causal reasoning module enables the method not only to detect fatigue but also to deeply analyze the causes of fatigue, thereby providing more targeted intervention measures.

[0047] In one embodiment of the present invention, causal reasoning can be implemented based on a Bayesian network. First, a Bayesian network containing various possible influencing factors (such as working hours, sleep quality, environmental noise, etc.) is constructed. Then, based on observed data, the probability distribution of each node in the network is updated. Finally, by calculating conditional probabilities, the factors that have the greatest impact on the current fatigue state are identified.

[0048] For example, suppose we observe that a user's fatigue state is "severe fatigue," a Bayesian network might give the following conditional probability: P({Working hours}>8h|{Severe fatigue})=0.8; P({sleep quality}={poor}|{severe fatigue})=0.7; P({Ambient noise}={High}|{Severe fatigue})=0.3; Based on these results, this method can determine that excessive working hours and poor sleep quality are the main causes of current severe fatigue, and thus generate corresponding intervention suggestions, such as "It is recommended to rest immediately and ensure sleep quality tonight".

[0049] Furthermore, the method of this invention also includes updating the parameters of the personalized fatigue model in real time through an online learning mechanism and optimizing the model performance based on user feedback. This continuous learning mechanism enables the method to constantly adapt to changes in user status and maintain long-term effectiveness.

[0050] In practical applications, a sliding time window (e.g., 1 hour) can be set to collect new data samples within each time window. These new samples are then used to incrementally update the model. The update process can employ stochastic gradient descent, with a learning rate set to a small value (e.g., 0.001) to ensure the model can smoothly adapt to the new data.

[0051] Simultaneously, this method also collects user feedback, such as user confirmation or denial of fatigue warnings. This feedback can serve as additional supervisory signals to further guide model optimization. For example, if users frequently deny the fatigue warnings given by the system, it may be necessary to adjust the model's sensitivity and increase the warning threshold.

[0052] In this way, the method of the present invention can continuously improve itself, providing users with increasingly accurate fatigue detection and early warning services. This personalized, adaptive method has stronger practicality and long-term effectiveness compared to traditional fixed-model methods. In a preferred embodiment of the present invention, the process of generating fatigue early warning information is further refined. Specifically, this process includes determining the early warning level based on the user's fatigue state and the risk level of the current task, and then generating corresponding early warning information and intervention measures based on the early warning level. This multi-level early warning mechanism can more accurately respond to different levels of fatigue, thereby providing more personalized and effective early warning services.

[0053] Preferably, this method can classify fatigue states into multiple levels, such as mild fatigue, moderate fatigue, and severe fatigue. Simultaneously, task risk levels can also be divided into low risk, medium risk, and high risk. By combining these two dimensions, a warning level matrix can be obtained. For example: 1. Mild fatigue + low-risk task = low-level warning; 2. Moderate fatigue + medium-risk task = intermediate warning level; 3. Severe fatigue + high-risk task = advanced warning; In determining the warning level, the method of this invention also considers the dynamic changes in fatigue state. Specifically, this method calculates the distance from the current fatigue state to the center point of the fatigue state, and determines the final warning level based on a comparison of this distance with a preset threshold. This method can more sensitively capture the changing trend of fatigue state, thereby achieving more timely warnings.

[0054] In one embodiment of the present invention, the distance from the fatigue state to the center point of the fatigue state can be calculated by the following formula: Where D represents distance, This represents the i-th feature value of the current fatigue state. Let represent the i-th eigenvalue of the fatigue state center point, and n be the total number of eigenvalues.

[0055] Preferably, multiple thresholds can be set to define different warning levels. For example: if This constitutes a low-level warning; if If so, it is a medium-level warning; if This constitutes a high-level warning; in, and These are predefined thresholds. The specific values ​​of these thresholds can be determined through experimental data analysis or expert experience. For example, they can be... Set it to 1.5 times the standard deviation. Set to 2.5 times the standard deviation.

[0056] Based on the determined warning level, the method of this invention generates corresponding warning information and intervention measures. For a low-level warning, it may simply be a reminder to the user to take a break. For a medium-level warning, it may suggest that the user take a short break immediately and provide some simple relaxation exercises. For a high-level warning, it may strongly advise the user to stop the current task, get plenty of rest, and seek medical help if necessary.

[0057] In another preferred embodiment of the invention, the method further includes acquiring the user's sleep state data and adjusting the parameters of a personalized fatigue model based on this data. This step allows the method to more comprehensively assess the user's fatigue state, as sleep quality has a significant impact on daytime fatigue.

[0058] Preferably, sleep data can be obtained through smartwatches or other wearable devices. This data typically includes metrics such as total sleep time, deep sleep time, REM sleep time, and sleep efficiency. For example, typical sleep data might look like this: Total sleep time: 7 hours and 30 minutes; Deep sleep time: 1 hour and 45 minutes; REM sleep duration: 1 hour 30 minutes; Sleep efficiency: 85%.

[0059] Based on this sleep data, this method can adjust the parameters of a personalized fatigue model. For example, if insufficient deep sleep time is detected in a user, the model's sensitivity to fatigue symptoms may be increased. This adjustment can be achieved by modifying the model's weights or thresholds.

[0060] Specifically, a sleep quality index (SQI) can be defined: SQI = {Deep sleep time + 0.5 * REM sleep time} / {Total sleep time} * Sleep efficiency Then, the parameters of the fatigue model are adjusted based on the SQI value. For example: if Keep the model parameters unchanged; if This lowers the fatigue detection threshold of the model; if The fatigue detection threshold of the model was reduced by 20%.

[0061] In this way, the method of the present invention can incorporate sleep quality, an important factor, into the scope of fatigue detection, thereby providing a more comprehensive and accurate fatigue assessment.

[0062] In summary, the fatigue detection and early warning method based on multimodal physiological signal fusion provided by this invention achieves accurate detection and timely warning of user fatigue status through multi-dimensional data acquisition, intelligent feature extraction and fusion, personalized modeling and early warning, and continuous online learning and optimization. This method not only considers the user's immediate physiological state but also incorporates long-term factors such as task characteristics, personal characteristics, and sleep quality, thus providing a comprehensive, accurate, and personalized fatigue management solution. This is of great significance for improving work efficiency, reducing accident risks, and improving quality of life.

[0063] The above description is merely a preferred embodiment of the present invention; however, the scope of protection of the present invention is not limited thereto; any equivalent substitutions or modifications made by those skilled in the art within the scope disclosed in the present invention, based on the scheme and improved concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A fatigue detection and early warning method based on multimodal physiological signal fusion, characterized in that, include: The acquisition steps include: Acquire the user's multimodal physiological signals, including electroencephalogram (EEG) signals, heart rate variability (HRV) signals, skin conductance signals, and accelerometer signals; The processing steps include: Multi-scale temporal features are extracted based on multimodal physiological signals; Based on multi-scale temporal characteristics, a personalized fatigue model is constructed; Based on a personalized fatigue model, determine the user's fatigue state; Output steps, including: Based on fatigue status, generate fatigue early warning information; Output fatigue warning information.

2. The method according to claim 1, characterized in that, The extraction of multi-scale temporal features specifically includes: The multimodal physiological signals are processed using a multi-scale temporal feature extraction network; The parameters of the multi-scale temporal feature extraction network are optimized based on the meta-learning framework; The extracted features are weighted and fused using an attention mechanism.

3. The method according to claim 2, characterized in that, The multi-scale temporal feature extraction network includes: CNN layers are used to encode EEG signals into multichannel features; Multiple BiLSTM layers are used to capture the temporal dependencies of EEG sequences; A multi-scale structure is used to combine multi-scale feature matrices from the outputs of different layers of BiLSTM.

4. The method according to claim 1, characterized in that, The construction of the personalized fatigue model specifically includes: Obtain user's individual characteristics, task characteristics, and data characteristics; Based on the individual characteristics, task characteristics, and data characteristics, the multi-scale temporal features are individually selected and fused; Based on the results of the personalized selection and fusion, a personalized fatigue model for the user is generated.

5. The method according to claim 4, characterized in that, The personalized fatigue model includes: A multi-scale feature selection module is used to select different temporal feature channels based on an attention mechanism; A multi-scale feature fusion module is used to fuse temporal features and frequency domain features; The personalized fatigue model module is used to output the final fatigue state.

6. The method according to claim 1, characterized in that, Also includes: Based on the causal reasoning module, the key factors leading to user fatigue are analyzed. Based on the key factors mentioned above, personalized intervention recommendations are generated.

7. The method according to claim 1, characterized in that, Also includes: The parameters of the personalized fatigue model are updated in real time through an online learning mechanism; Based on user feedback, the performance of the personalized fatigue model was optimized.

8. The method according to claim 1, characterized in that, The generation of fatigue early warning information specifically includes: The warning level is determined based on the fatigue state and the risk level of the current task; Based on the aforementioned warning level, corresponding warning information and intervention measures are generated.

9. The method according to claim 8, characterized in that, The determination of the early warning level includes: Calculate the distance from the fatigue state to the center point of the fatigue state; The warning level is determined based on the comparison between the distance and the preset threshold.

10. The method according to claim 1, characterized in that, Also includes: Obtain user's sleep data; Based on the sleep state data, the parameters of the personalized fatigue model are adjusted.

Images