Near infrared spectrum physical and mental pressure and sleep quality monitoring system

Abstract

The invention discloses a near infrared spectrum physical and mental pressure and sleep quality monitoring system. The system comprises a probe; the data acquisition module, the data preprocessing module, the feature extraction module, the data analysis module, the user interface module and the system integration and optimization module effectively remove noise and baseline drift in signals through multi-scale filtering and dynamic baseline correction, and ensure authenticity and accuracy of monitoring data. The transfer learning mechanism enables the model to adapt to new data distribution and user differences, the generalization ability of the model is improved, the light source power and the receiver gain are automatically adjusted according to the change of the signal intensity, and it is ensured that the data accuracy is not affected by environmental changes. Through a feature extraction and selection algorithm, in combination with a deep learning model and a machine learning model, the accuracy and robustness of sleep state recognition and physical and psychological pressure evaluation are improved. And the model can be updated in real time through an online learning mechanism to adapt to the current state and environment change of the user.

Description

Technical Field

[0001] The invention relates to the technical field of biomedical engineering, and in particular to a near-infrared spectrum physical and mental stress and sleep quality monitoring system. Background Art

[0002] In today's fast-paced society, people are facing unprecedented physical and mental stress. Long-term psychological stress not only affects the individual's emotional state, but also has a negative impact on physical health, including cardiovascular disease, decreased immune system function, and sleep disorders. At the same time, high-quality sleep is essential for maintaining physical and mental health. However, modern people generally suffer from insufficient sleep or poor sleep quality, which further aggravates the deterioration of physical and mental health. Therefore, monitoring and managing physical and mental stress and sleep quality has become an important topic in the field of public health and health management.

[0003] After searching, the invention patent with Chinese patent number CN117814766A discloses a near-infrared spectrum physical and mental stress and sleep quality monitoring system, which relates to the field of human sleep monitoring technology. The system includes a preliminary test preparation unit, a signal data acquisition unit, a sleep state recognition unit, a physical and mental stress acquisition unit, a work mode formulation unit, a real-time monitoring integration unit and a monitoring modification and adjustment unit. Compared with the prior art, the invention patent with Chinese patent number CN117814766A uses near-infrared spectrum data to calculate and reflect the physiological characteristics of the tester, and combines the classification algorithm to establish a sleep state recognition model, and combines the frequency domain change technology to build an analysis model, analyzes and evaluates the physical sign status of the tester from multiple dimensions, and integrates the sleep state recognition model and the analysis model into the massage device controller. The massage device can monitor the user's stress and sleep state in real time, and adjust the massage device's working mode in real time according to the monitoring data, providing a personalized massage experience, and providing users with a more comfortable and personalized service experience.

[0004] However, the above invention may cause inaccurate data collection when the device moves or the environment changes during actual use. In addition, each person has different physiological characteristics, so the system needs to be able to adapt to the physiological characteristics of different users.

[0005] Therefore, a near-infrared spectroscopy physical and mental stress and sleep quality monitoring system is proposed. Summary of the invention

[0006] The purpose of the present invention is to solve the shortcomings of the prior art that the collected data may be inaccurate when the device moves or the environment changes. In addition, each person has different physiological characteristics, and the system needs to be able to adapt to the physiological characteristics of different users. A near-infrared spectroscopy physical and mental stress and sleep quality monitoring system is proposed.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] A near infrared spectroscopy physical and mental stress and sleep quality monitoring system, comprising:

[0009] Probes for monitoring the user's physical and mental stress and sleep quality through near-infrared spectroscopy technology;

[0010] Data acquisition module: used to ensure data accuracy and stability; through the probe, as well as intelligent power management and adaptive calibration algorithms, it ensures that high-quality data can be collected when the device moves or the environment changes;

[0011] Data preprocessing module: removes noise and interference; uses dynamic baseline correction and multi-scale filtering technology to ensure that the preprocessed data is clean and accurate;

[0012] Feature extraction module: extracts useful features from preprocessed data for subsequent classification tasks; extracts the most representative features through multi-domain feature fusion and adaptive feature selection;

[0013] Data analysis module: Input the extracted features into the hybrid learning architecture, combining deep learning and traditional machine learning; use the deep learning framework to build a convolutional neural network and perform classification through a support vector machine;

[0014] Transfer learning mechanism: Through the transfer learning mechanism, the model can adapt to the new data distribution; using domain adaptation technology to reduce errors caused by user differences;

[0015] Online learning mechanism: Through the online learning mechanism, the model can be updated in real time; using incremental learning algorithm to ensure that the model reflects the user's current status;

[0016] User interface module: provides user experience, supports multi-platform deployment, implements natural language processing and sentiment analysis, and provides personalized feedback;

[0017] System Integration and Optimization Module: Ensures the security and reliability of the system; this module enables end-to-end encryption protocols, uses zero-knowledge proof technology to protect user privacy, and implements multi-factor authentication and automated failover solutions.

[0018] The above technical solution further includes:

[0019] Preferably, the probe comprises:

[0020] Three light sources with different wavelengths: 650nm, 760nm, and 850nm;

[0021] An accelerometer for detecting user activity;

[0022] Housing made of flexible electronic materials;

[0023] Temperature sensor, used to compensate for measurement errors due to temperature changes.

[0024] Preferably, the data acquisition module includes:

[0025] Adaptive calibration algorithm, automatically adjusts source power ±20% and receiver gain ±10% based on signal strength changes;

[0026] Intelligent power management mechanism adjusts device power consumption according to user activity status. When the user is detected to be stationary, the sampling frequency is reduced to 1Hz to save power; when the user is detected to be moving, the full-speed sampling is restored to 10Hz;

[0027] The adaptive calibration algorithm also includes a temperature compensation function for eliminating the effects of ambient temperature changes;

[0028] The battery management system also includes charging protection and low-battery warning functions to prevent battery damage due to overcharging or over-discharging.

[0029] Preferably, the data preprocessing module includes:

[0030] Adaptive window size adjustment mechanism: when the signal changes greatly, the window size is reduced to 5 seconds to capture the change more quickly; when the signal is relatively stable, the window size is increased to 20 seconds to reduce the amount of calculation;

[0031] A quadratic polynomial fit was used to estimate the baseline trend, and this curve was subtracted from the raw signal to obtain the corrected signal;

[0032] Use db4 wavelet transform to perform multi-scale decomposition and remove high-frequency noise according to the preset threshold rule;

[0033] For low-frequency drift, empirical mode decomposition (EMD) is used to separate the intrinsic mode function (IMF) and remove the slowly changing trend from it.

[0034] Use the standard deviation of the signal to dynamically adjust the threshold of the multi-scale filter;

[0035] The median rather than the mean was used to calculate the signal level within the sliding window to reduce the impact of burst signals on the baseline estimate.

[0036] Preferably, the feature extraction module includes:

[0037] Use the principal component analysis (PCA) dimensionality reduction algorithm to retain the principal components with a cumulative contribution rate of more than 85%;

[0038] Through recursive feature elimination (RFE) and LASSO regression algorithms, unimportant features are gradually eliminated, and the final feature set is determined through LASSO regression.

[0039] Calculate the ability of each feature to distinguish between different categories to determine the weight;

[0040] Use mutual information to measure the importance of features;

[0041] Set the penalty term l ambda value to avoid overfitting;

[0042] The optimal l ambda value is selected through cross-validation to achieve the best sparsification effect;

[0043] The feature selection process is repeated multiple times and the frequency with which each feature is selected is calculated.

[0044] Preferably, the data analysis module includes:

[0045] Convolutional neural network CNN built using deep learning framework, using ResNet50 structure;

[0046] Classification is performed through support vector machine SVM, using ResNet50 as the CNN model, and the output feature vector is input into SVM for classification;

[0047] Train the base model on a public dataset and initialize the CNN model using pre-trained weights;

[0048] By adjusting the weights of the fully connected layer, setting the learning rate to 0.001 and the number of iterations to 50;

[0049] Use exponential decay to adjust the learning rate;

[0050] Use Hilbert-Huang transform HHT to calculate the instantaneous frequency and energy of IMF;

[0051] Evaluate the classification performance of the model.

[0052] Preferably, the user interface module includes:

[0053] Support multi-platform deployment;

[0054] Implement natural language processing and sentiment analysis;

[0055] Provide personalized feedback.

[0056] Preferably, the system integration and optimization module includes:

[0057] Enable end-to-end encryption protocol;

[0058] Use intrusion detection systems to monitor abnormal access behavior;

[0059] Use zero-knowledge proof technology to protect user privacy;

[0060] Requires at least two types of authentication information when a user logs in.

[0061] Preferably, the data analysis module includes:

[0062] Generate health recommendations by combining expert system knowledge base;

[0063] Use collaborative filtering technology to recommend similar users’ behavior patterns;

[0064] Continuously optimize the suggestion content through user feedback mechanism.

[0065] Preferably, the system integration and optimization module includes:

[0066] Dynamically adjust the server cluster size based on real-time load conditions;

[0067] Implement automated failover scenarios;

[0068] Use containerization and load balancing technologies;

[0069] Back up your database regularly and store multiple copies in the cloud;

[0070] Use redundant storage.

[0071] The present invention has the following beneficial effects:

[0072] 1. In the present invention, through multi-scale filtering and dynamic baseline correction, the noise and baseline drift in the signal are effectively removed, ensuring the authenticity and accuracy of the monitoring data. Graphene composite materials are used to ensure the comfort of users when wearing for a long time. The light source power and receiver gain are automatically adjusted according to the change of signal strength to ensure that the data accuracy is not affected by environmental changes. The online learning mechanism ensures that the model can be updated in real time, adapt to the user's current state and environmental changes, and improve the accuracy and reliability of the monitoring results. Through the feature extraction and selection algorithms PCA, RFE, LASSO, combined with the deep learning model ResNet50 and the traditional machine learning model SVM, the accuracy and robustness of sleep state recognition and physical and mental stress assessment are improved. It supports multi-platform deployment, and adapts to the screen size of different devices through responsive design, which improves the friendliness and ease of use of the user interface. In addition, it can provide personalized health advice based on the user's physiological characteristics and real-time monitoring results to help users improve sleep quality and reduce physical and mental stress.

[0073] 2. In the present invention, the model can adapt to new data distribution and user differences through the transfer learning mechanism, improve the generalization ability and robustness of the model, adopt end-to-end encryption protocol in the data transmission process to ensure the security of user data, and use zero-knowledge proof technology to protect privacy. The server cluster size can be dynamically adjusted according to the real-time load situation, and the integrity and reliability of the data can be improved through regular backup and redundant storage technology to ensure the stable operation of the system. The use of multiple identity authentication methods combined with biometric recognition technology enhances the security and convenience of authentication. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 This is a system architecture diagram of a near-infrared spectroscopy physical and mental stress and sleep quality monitoring system proposed by the present invention;

[0075] Figure 2 It is a structural diagram of the hybrid learning architecture in the present invention;

[0076] Figure 3 It is a structural diagram of the online learning mechanism in the present invention;

[0077] Figure 4 Schematic diagram of the structure of the transfer learning mechanism in the present invention. DETAILED DESCRIPTION

[0078] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0079] like Figure 1-Figure 4 As shown, the present invention proposes a near infrared spectrum physical and mental stress and sleep quality monitoring system, comprising:

[0080] Probes for monitoring the user's physical and mental stress and sleep quality through near-infrared spectroscopy technology;

[0081] Data acquisition module: used to ensure data accuracy and stability; through the probe, as well as intelligent power management and adaptive calibration algorithms, it ensures that high-quality data can be collected when the device moves or the environment changes;

[0082] Data preprocessing module: removes noise and interference; uses dynamic baseline correction and multi-scale filtering technology to ensure that the preprocessed data is clean and accurate;

[0083] Feature extraction module: extracts useful features from preprocessed data for subsequent classification tasks; extracts the most representative features through multi-domain feature fusion and adaptive feature selection;

[0084] Data analysis module: Input the extracted features into the hybrid learning architecture, combining deep learning and traditional machine learning; use the deep learning framework to build a convolutional neural network and perform classification through a support vector machine;

[0085] Transfer learning mechanism: Through the transfer learning mechanism, the model can adapt to the new data distribution; using domain adaptation technology to reduce errors caused by user differences;

[0086] Online learning mechanism: Through the online learning mechanism, the model can be updated in real time; using incremental learning algorithm to ensure that the model reflects the user's current status;

[0087] User interface module: provides user experience, supports multi-platform deployment, implements natural language processing and sentiment analysis, and provides personalized feedback;

[0088] System Integration and Optimization Module: Ensures the security and reliability of the system; this module enables end-to-end encryption protocols, uses zero-knowledge proof technology to protect user privacy, and implements multi-factor authentication and automated failover solutions.

[0089] In one embodiment, the probe

[0090] The NI RS probe measures no more than 3 cm x 3 cm and weighs no more than 5 g. It includes a temperature sensor to compensate for measurement errors caused by temperature changes.

[0091] The probe includes three different wavelengths of light sources (650nm, 760nm, 850nm), using three LEDs as light sources to enhance signal contrast and improve monitoring accuracy. Each wavelength of light source has an independent driving circuit to ensure the consistency and stability of the light source brightness.

[0092] The probe shell uses a 100μm thick graphene composite material to improve the fit of the device and ensure the accuracy and consistency of data collection. The graphene composite material has good thermal conductivity and tensile resistance.

[0093] Temperature sensor: The probe has a built-in temperature sensor to compensate for measurement errors caused by temperature changes

[0094] In one embodiment, the data acquisition module

[0095] Intelligent Power Management:

[0096] Adjust device power consumption according to user activity status (static / dynamic). When the user is detected to be stationary, the sampling frequency is reduced to 1Hz to save power; when the user is detected to be moving, the full-speed sampling is restored to 10Hz. Specifically, the accelerometer detects the user's activity status and adjusts the sampling frequency accordingly to extend the device's working time and improve the user experience. The battery management system also includes charging protection and low-battery warning functions to prevent battery damage caused by overcharging or over-discharging.

[0097]

[0098] Where a is the user activity intensity detected by the accelerometer, a threshold is the set threshold.

[0099] When the system detects abnormal signal strength (such as signal weakening exceeding the threshold of 5%), the automatic calibration process is triggered. Calibration compensates for signal loss by adjusting the light source power (±20%) and receiver gain (±10%). Specifically, when the signal strength is below the preset threshold, the light source power is automatically increased and the receiver gain is adjusted to ensure that a stable signal can be obtained in various situations and compensate for individual differences in hardware design. The calibration algorithm also includes a temperature compensation function to eliminate the impact of ambient temperature changes.

[0100] P new =P old +ΔP

[0101] G new =G old +ΔG

[0102] Where P represents the light source power, G represents the receiver gain, and ΔP and ΔG represent the changes in power and gain, respectively.

[0103] In one embodiment, the data preprocessing module

[0104] Dynamic baseline correction:

[0105] The mean value of the signal is monitored in real time. When the baseline drift is detected to exceed the set threshold (e.g. 5%), the baseline correction algorithm is started. The correction is achieved by estimating and removing the trend component in the baseline signal. Specifically, the average value of the signal is calculated using a sliding window size of 10 seconds, and the correction is started when the mean value changes by more than the threshold, which improves the accuracy of subsequent feature extraction.

[0106] Baseline(t)=median({x(tn),x(t-n+1),…,x(t)})

[0107] Where n is the window size and x is the original signal.

[0108] When the signal changes greatly, the window size is reduced to 5 seconds to capture the changes more quickly; when the signal is relatively stable, the window size is increased to 20 seconds to reduce the amount of calculation. A quadratic polynomial fit is used to estimate the baseline trend, and this curve is subtracted from the original signal to obtain the corrected signal.

[0109] y(t)=at 2 +bt+c

[0110] Among them, a, b, and c are the polynomial coefficients obtained by solving the least squares method.

[0111] The median rather than the mean was used to calculate the signal level within the sliding window to reduce the impact of burst signals on the baseline estimate.

[0112] Multiscale filtering:

[0113] For each set of sampled data, multi-scale decomposition is performed using db4 wavelet transform. High-frequency noise is removed according to preset threshold rules (such as hard threshold or soft threshold). For low-frequency drift, empirical mode decomposition (EMD) is used to separate the intrinsic mode function (IMF) and remove slowly changing trends from it. Specifically, db4 wavelet is used for decomposition, and the threshold is set to 1% of the maximum value of the signal to remove noise, retain useful signals, and ensure the accuracy of subsequent analysis.

[0114]

[0115] Among them, C j,k is the j-th layer wavelet coefficient, ψ j,k (n) is the wavelet basis function, and T is the threshold.

[0116] The standard deviation of the signal is used to dynamically adjust the threshold of the multi-scale filter. The Hilbert-Huang transform (HHT) is used to calculate the instantaneous frequency and energy of the IMF to better capture the time-frequency characteristics of the signal.

[0117]

[0118] Among them, IMF i (t) is the ith intrinsic mode function, and r(t) is the residual signal.

[0119] In one embodiment, the feature extraction module

[0120] Multi-domain feature fusion:

[0121] Extract time domain features (such as mean, variance), frequency domain features (such as power spectrum density) and time-frequency domain features (such as STFT coefficients) from the preprocessed signal. Combine feature vectors from different fields through feature fusion algorithms (such as principal component analysis PCA and linear discriminant analysis LDA). Specifically, use PCA to reduce the dimension and combine features from different fields into one feature vector to improve the recognition accuracy of the model.

[0122] X reduced =XW

[0123] Among them, X is the feature matrix and W is the matrix composed of the first few principal components.

[0124] Different weights are given according to the importance of the features to highlight the key features. The weights can be determined by calculating the ability of each feature to distinguish between different categories (for example, using mutual information MI).

[0125] When using PCA for dimensionality reduction, retain the first few principal components (such as the first five) to ensure that most of the information is retained while reducing the dimension. The number of principal components to be retained can be determined by the Cumulative Contribut ion Rate. Generally, principal components with a cumulative contribution rate of more than 85% are selected.

[0126] F selected =RFE(X,y)

[0127] Among them, X is the feature matrix, y is the label vector, and RFE is the recursive feature elimination algorithm.

[0128] Adaptive Feature Selection:

[0129] After feature extraction is completed, the features are sorted according to their importance through methods such as recursive feature elimination (RFE) and LASSO regression, and the optimal subset is selected for model training. Specifically, the RFE method is used to gradually eliminate unimportant features, and the final feature set is determined through LASSO regression to reduce redundant information and improve model efficiency.

[0130] Mutual information is used to measure the importance of features, and RFE is combined to gradually eliminate unimportant features. Mutual information can measure the degree of dependence between two random variables and is applicable to nonlinear relationships.

[0131] Set an appropriate penalty term (such as the lambda value) to avoid overfitting. The optimal lambda value can be selected through cross-validation to achieve the best sparsification effect.

[0132] Repeat the feature selection process several times and calculate the frequency of each feature being selected. If certain features appear frequently, it means that these features are relatively stable and can be used as the final feature set.

[0133] In one embodiment, the data analysis module

[0134] Hybrid Learning Architecture:

[0135] Use a deep learning framework such as TensorFlow or PyTorch to build a convolutional neural network (CNN) with multiple convolutional layers (e.g. ResNet50) and fully connected layers.

[0136] Classification is performed through support vector machine (SVM), ResNet50 is used as the CNN model, and the output feature vector is input into SVM for classification, combining deep learning and traditional machine learning to improve the overall performance of the model.

[0137] The base model is trained on public datasets (such as the UCI ECG dataset) and the CNN model is initialized using pre-trained weights on ImageNet.

[0138] A small amount of user data is used to fine-tune the basic model by adjusting the weights of the fully connected layer and setting the learning rate to 0.001 and the number of iterations to 50.

[0139] Use exponential decay to adjust the learning rate to speed up convergence. You can use strategies such as exponential decay or cosine annealing to adjust the learning rate.

[0140] η(t)=η0·γ t / T

[0141] Among them, η0 is the initial learning rate, γ is the decay factor, and T is the cycle length.

[0142] Use indicators such as AUC (Area Under the Curve) to evaluate the classification performance of the model, and adjust training parameters or increase training data based on the evaluation results.

[0143]

[0144] Among them, L is the loss function, y i is the true label, is the predicted label.

[0145] Transfer learning mechanism: In the process of transfer learning, domain adaptation technology is used to adjust the model so that it can better adapt to the new data distribution. Domain adaptation technology helps the model transfer knowledge between different domains and improve generalization ability.

[0146] Use biometric technology to improve the reliability and convenience of authentication.

[0147] Update authentication strategies regularly to address new security challenges.

[0148] Online learning mechanism: Use incremental learning algorithms (such as online stochastic gradient descent OSGD) to adjust model parameters online. When new data arrives, use the OSGD algorithm to update model parameters and regularly evaluate model performance to ensure that the model reflects the current state of the user.

[0149] Considering the characteristics of data stream, a sliding window method is used to process the continuously arriving new data to reduce memory usage. A fixed-size window is used, and the earliest data is removed one by one as new data arrives.

[0150] When updating model parameters, momentum techniques are used to accelerate convergence. Momentum helps the algorithm to move through flat regions faster and reduce oscillations.

[0151] Adjust the learning rate by regularly evaluating the model performance. Use a validation set to evaluate the model's performance on unseen data and adjust the learning rate or optimization algorithm based on the evaluation results.

[0152] In one embodiment, the user interface module

[0153] Multi-platform support: deploy a unified front-end framework on different operating systems (i OS, Android, Web); identify the user terminal type through the device detection API and automatically load the corresponding UI layout file to ensure that users get a consistent experience on different devices.

[0154] Use responsive design techniques to adapt to the screen size and resolution differences of different devices. Responsive design can automatically adjust the layout and style according to the screen size of the device.

[0155] Use local storage technology to cache user data and improve application performance. Local storage reduces network requests and improves user experience.

[0156] Simplify the development process by using cross-platform frameworks. These frameworks allow developers to write code once and run it on multiple platforms, reducing development costs.

[0157] Interaction design: Use natural language processing technology to parse the user's natural language input; call the corresponding back-end service based on the semantic understanding and intent recognition results; judge the user's emotional state through the sentiment analysis engine and adjust the feedback tone to make the system more humane.

[0158] Use deep learning models for semantic understanding and intent recognition to improve accuracy and robustness.

[0159] Use sentiment analysis to determine the user's emotional state and adjust the tone of feedback based on the emotion to improve user experience.

[0160] Health advice: Based on the results of data analysis, the system evaluates the user's current physical condition and generates specific health improvement suggestions based on the medical knowledge base and personalized recommendation algorithm; these suggestions are communicated to users through push notifications or in-app messages to help users improve their lifestyle and quality of life.

[0161] Use collaborative filtering technology to recommend behavior patterns of similar users, and combine the knowledge base of the expert system to generate scientific and reasonable suggestions.

[0162] Provide advice on diet, exercise and rest to ensure that users can choose appropriate advice based on their actual situation.

[0163] Continuously optimize the suggestion content through the user feedback mechanism to improve user satisfaction.

[0164] In one embodiment, the system integrates and optimizes

[0165] Security considerations: Enable end-to-end encryption protocols during data transmission; regularly check the validity of encryption keys and replace them when necessary; use intrusion detection systems to monitor abnormal access behavior and issue timely alarms to ensure the security of user data.

[0166] Use zero-knowledge proof technology to protect user privacy.

[0167] User authentication mechanism: requires users to provide at least two types of identity authentication information when logging in; uses timestamp technology to generate dynamic verification codes, which are valid for only a few minutes each time they log in; and regularly updates authentication policies to ensure account security.

[0168] Use biometric technology to improve the reliability and convenience of authentication.

[0169] Cloud service support: Dynamically adjust the server cluster size according to real-time load conditions; regularly back up the database and store multiple copies in the cloud; implement automated failover solutions to quickly switch to backup nodes when service interruptions are detected to ensure stable operation and efficient response of the system.

[0170] Use containerization technology and load balancing technology to improve system availability and response speed.

[0171] Use regular backups to prevent data loss and increase data reliability through redundant storage.

[0172] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A near infrared spectroscopy physical and mental stress and sleep quality monitoring system, characterized in that: include: Probes for monitoring the user's physical and mental stress and sleep quality through near-infrared spectroscopy technology; Data acquisition module: used to ensure the accuracy and stability of data; The probe, as well as intelligent power management and adaptive calibration algorithms, ensures that high-quality data can be collected even when the device is moving or the environment changes; Data preprocessing module: remove noise and interference; use dynamic baseline correction and multi-scale filtering technology; Feature extraction module: extracts features from preprocessed data for subsequent classification tasks; extracts features through multi-domain feature fusion and adaptive feature selection; Data analysis module: Input the extracted features into the hybrid learning architecture, combining deep learning and traditional machine learning; use the deep learning framework to build a convolutional neural network and perform classification through a support vector machine; Transfer learning mechanism: Through the transfer learning mechanism, the model can adapt to the new data distribution; using domain adaptation technology to reduce errors caused by user differences; Online learning mechanism: Through the online learning mechanism, the model can be updated in real time; using incremental learning algorithm to ensure that the model reflects the user's current status; User interface module: provides user experience, supports multi-platform deployment, implements natural language processing and sentiment analysis, and provides personalized feedback; System integration and optimization module: ensure the security and reliability of the system; The module enables end-to-end encryption protocols, uses zero-knowledge proof technology to protect user privacy, and implements multi-factor authentication and automated failover schemes.

2. A near infrared spectroscopy physical and mental stress and sleep quality monitoring system according to claim 1, characterized in that: The probe comprises: Three light sources with different wavelengths: 650nm, 760nm, and 850nm; An accelerometer for detecting user activity; Housing made of flexible electronic materials; Temperature sensor, used to compensate for measurement errors due to temperature changes.

3. A near infrared spectroscopy physical and mental stress and sleep quality monitoring system according to claim 1, characterized in that: The data acquisition module comprises: Adaptive calibration algorithm, automatically adjusts source power ±20% and receiver gain ±10% based on signal strength changes; Intelligent power management mechanism adjusts device power consumption according to user activity status. When the user is detected to be stationary, the sampling frequency is reduced to 1Hz to save power; when the user is detected to be moving, the full-speed sampling is restored to 10Hz; The adaptive calibration algorithm also includes a temperature compensation function for eliminating the effects of ambient temperature changes; The battery management system also includes charging protection and low-battery warning functions to prevent battery damage due to overcharging or over-discharging.

4. A near infrared spectroscopy physical and mental stress and sleep quality monitoring system according to claim 1, characterized in that: The data preprocessing module comprises: Adaptive window size adjustment mechanism: when the signal changes greatly, the window size is reduced to 5 seconds to capture the change more quickly; when the signal is relatively stable, the window size is increased to 20 seconds to reduce the amount of calculation; A quadratic polynomial fit was used to estimate the baseline trend, and this curve was subtracted from the raw signal to obtain the corrected signal; Use db4 wavelet transform to perform multi-scale decomposition and remove high-frequency noise according to the preset threshold rule; For low-frequency drift, empirical mode decomposition (EMD) is used to separate the intrinsic mode function (IMF) and remove the slowly changing trend from it. Use the standard deviation of the signal to dynamically adjust the threshold of the multi-scale filter; The median rather than the mean was used to calculate the signal level within the sliding window to reduce the impact of burst signals on the baseline estimate.

5. A near infrared spectroscopy physical and mental stress and sleep quality monitoring system according to claim 1, characterized in that: The feature extraction module comprises: Use the principal component analysis (PCA) dimensionality reduction algorithm to retain the principal components with a cumulative contribution rate of more than 85%; Through recursive feature elimination (RFE) and LASSO regression algorithms, unimportant features are gradually eliminated, and the final feature set is determined through LASSO regression. Calculate the ability of each feature to distinguish between different categories to determine the weight; Use mutual information to measure the importance of features; Set the penalty term lambda value to avoid overfitting; The optimal lambda value is selected through cross-validation to achieve the best sparsification effect; The feature selection process is repeated multiple times and the frequency with which each feature is selected is calculated.

6. A near infrared spectroscopy physical and mental stress and sleep quality monitoring system according to claim 1, characterized in that: The data analysis module includes: Convolutional neural network CNN built using deep learning framework, using ResNet50 structure; Classification is performed through support vector machine SVM, using ResNet50 as the CNN model, and the output feature vector is input into SVM for classification; Train the base model on a public dataset and initialize the CNN model using pre-trained weights; By adjusting the weights of the fully connected layer, setting the learning rate to 0.001 and the number of iterations to 50; Use exponential decay to adjust the learning rate; Use Hilbert-Huang transform HHT to calculate the instantaneous frequency and energy of IMF; Evaluate the classification performance of the model.

7. A near infrared spectroscopy physical and mental stress and sleep quality monitoring system according to claim 1, characterized in that: The user interface module comprises: Support multi-platform deployment; Implement natural language processing and sentiment analysis; Provide personalized feedback.

8. The near infrared spectroscopy physical and mental stress and sleep quality monitoring system according to claim 1, characterized in that: The system integration and optimization module includes: Enable end-to-end encryption protocol; Use intrusion detection systems to monitor abnormal access behavior; Use zero-knowledge proof technology to protect user privacy; Requires at least two types of authentication information when a user logs in.

9. The near infrared spectroscopy physical and mental stress and sleep quality monitoring system according to claim 1, characterized in that: The data analysis module includes: Generate health recommendations by combining expert system knowledge base; Use collaborative filtering technology to recommend similar users’ behavior patterns; Continuously optimize the suggestion content through user feedback mechanism.

10. The near infrared spectroscopy physical and mental stress and sleep quality monitoring system according to claim 1, characterized in that: The system integration and optimization module includes: Dynamically adjust the server cluster size based on real-time load conditions; Implement automated failover scenarios; Use containerization and load balancing technologies; Back up your database regularly and store multiple copies in the cloud; Use redundant storage.

Images