Mental health assessment method and equipment based on optical fiber sensor and storage medium

Abstract

The invention relates to a psychological health assessment method and equipment based on an optical fiber sensor and a storage medium. The method comprises the following steps: acquiring an original physiological signal of a person to be detected through an optical fiber sensor, and performing data preprocessing on the original physiological signal to obtain a preprocessed signal; constructing a transfer learning model, and obtaining a public physiological signal data set to train the transfer learning model to obtain an evaluation model; performing physiological signal feature extraction on the preprocessed signals, performing series fusion on the extracted features of the physiological signals to form a multi-modal feature vector, and inputting the multi-modal feature vector into the evaluation model to obtain a classification probability output result; and generating a mental health assessment report according to the classification probability output result. A complete, efficient and accurate quantitative evaluation system for anxiety depression, emotion stress, fatigue level and pressure sensitivity can be constructed, and high-precision and real-time monitoring and evaluation of the mental health state are achieved.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and in particular, to a mental health assessment method, device, and storage medium based on fiber optic sensors. Background Art

[0002] With the accelerating pace of modern social life, mental health problems have become increasingly prominent. Data from the World Health Organization shows that hundreds of millions of people around the world are troubled by mental illnesses such as anxiety and depression every year. The current mainstream mental health assessment methods have significant limitations: The traditional assessment method is highly subjective: The assessment methods based on questionnaire scales (such as the Hamilton Anxiety Scale and the Self-Rating Depression Scale) and doctor interviews rely on the subjective statements of the tested individuals and the clinical experience judgment of doctors, and are easily affected by individual cognitive biases, emotional fluctuations, and doctor's subjective factors, resulting in insufficient accuracy and reliability of the assessment results. For example, some tested individuals may deliberately conceal their true emotions due to psychological defense mechanisms, or may not be able to accurately describe their mental states due to limited expression ability.

[0003] The electronic sensor assessment has defects: The physiological signal acquisition technology based on electronic sensors, such as electrocardiogram electrodes and galvanic skin sensors, although it can obtain some physiological information, has problems such as being easily affected by electromagnetic interference, poor wearing comfort, and insufficient signal stability for long-term monitoring. In actual application scenarios, the data collected by electronic sensors in complex electromagnetic environments (such as hospital equipment rooms and industrial sites) is prone to noise interference, affecting the subsequent analysis results.

[0004] The single-modal assessment information is one-sided: Existing physiological signal assessments mostly use single-modal data (such as only analyzing heart rate or galvanic skin signals), which is difficult to comprehensively reflect the complexity of mental health states. Mental health is the result of the combined action of multiple physiological and psychological factors. Single-modal data cannot capture the correlation information between multiple factors, resulting in one-sided assessment results.

[0005] Lack of efficient algorithms and quantification systems: Although some studies have tried multi-modal fusion, they lack systematic algorithm support and quantification assessment systems. Existing algorithms have problems such as low computational efficiency and weak model generalization ability when processing large-scale physiological signal data, and have not formed a unified mental health quantification assessment index, making it difficult to meet the application requirements of multiple scenarios such as clinical diagnosis and health management.

[0006] Although fiber Bragg grating sensors have advantages such as anti-electromagnetic interference, high sensitivity, and distributed measurement in physiological signal acquisition, and multi-modal fusion and transfer learning technologies have shown good application potential in other fields, there is currently no mature solution that deeply combines the three to construct a mental health quantification assessment system. Summary of the Invention

[0007] Based on this, a mental health assessment method, device, and storage medium based on fiber optic sensors are provided to solve the technical problem of the current lack of a complete quantitative assessment system to accurately assess mental health.

[0008] On the one hand, a mental health assessment method based on fiber optic sensors is provided. The method includes: Collect the original physiological signals of the person to be detected through a fiber optic sensor, and perform data preprocessing on the original physiological signals to obtain preprocessed signals; Construct a transfer learning model, obtain a public physiological signal dataset, label multi-modal feature vectors for each physiological signal in the public physiological signal dataset, set the transfer learning model to include a source domain model and a target domain model, use the data of the public physiological signal dataset with labeled multi-modal feature vectors as the target domain model data, input it into the pre-trained source domain model, construct a transfer learning adjustment formula according to the feature extractor and the domain classifier, introduce a gradient reversal layer between the feature extractor and the domain classifier to achieve gradient reversal, reduce the difference between the source domain model and the target domain model through the domain classifier, and perform transfer learning adjustment on the transfer learning model to obtain an evaluation model; Extract the features of each physiological signal from the preprocessed signals, concatenate and fuse the features extracted from each physiological signal into a multi-modal feature vector, and input it into the evaluation model to obtain a classification probability output result; Calculate the anxiety-depression and emotional stress index, and the fatigue level and stress sensitivity index according to the classification probability output result, and generate a mental health assessment report according to the anxiety-depression and emotional stress index, and the fatigue level and stress sensitivity index.

[0009] Further, the step of collecting the original physiological signals of the person to be detected through a fiber optic sensor and performing data preprocessing on the original physiological signals to obtain preprocessed signals includes: Set the collected original physiological signals to include heart rate, respiratory rate, body temperature, and muscle tension physiological signals, and perform wavelet transform denoising, band-pass filtering, and normalization processing on the original physiological signals to obtain preprocessed signals; When using a fiber Bragg grating sensor to collect heart rate and respiratory rate signals, based on the Bragg reflection condition of the fiber Bragg grating, its reflection center wavelength satisfies the formula ; where is the wavelength offset, is the effective refractive index of the optical fiber, is the grating period, is the photoelastic coefficient, is the strain; When using a fiber optic temperature sensor to measure body temperature, the formula used is ; where is the wavelength change caused by temperature, is the temperature coefficient, is the temperature change; When using a distributed fiber optic strain sensor to collect muscle tension signals, strain signals of optical fibers at different positions are collected at a preset sampling rate.

[0010] Furthermore, the extraction of various physiological signal features from the preprocessed signal includes: When the physiological signal is a heart rate signal, time-domain features and frequency-domain features of the preprocessed heart rate signal are extracted; When extracting time-domain features of the preprocessed heart rate signal, the mean of the time-domain features is , represents the number of sampling points, is the th sampling point of the heart rate signal value; the variance of the time-domain features is ; When extracting frequency-domain features of the preprocessed heart rate signal, after converting the time-domain heart rate signal to the frequency domain using the fast Fourier transform (FFT), the low-frequency band range c1 - c2 and the high-frequency band range c2 - c3 are obtained. The low-frequency energy obtained by integrating the power spectral density within the low-frequency band is , and the high-frequency energy obtained by integrating the power spectral density within the high-frequency band is , where represents the power spectral density at a frequency of ; By calculating the ratio of the low-frequency energy to the high-frequency energy , it is used to evaluate the balance state of the autonomic nervous system; When the physiological signal is a respiratory frequency signal, features of the preprocessed respiratory frequency signal are extracted; Let the respiratory cycle duration sequence be , then the mean of the respiratory cycle duration is , the standard deviation of the respiratory cycle duration is , and the change rate of adjacent respiratory cycle durations is ; where is the number of respiratory cycles, represents the th respiratory cycle duration; When the physiological signal is a body temperature signal, features of the preprocessed body temperature signal are extracted; Let the body temperature signal sequence be , then the mean of the body temperature signal is , the fluctuation range of the body temperature signal is , is the number of sampling points of the body temperature signal, is the body temperature value of the th sampling point, is the maximum value in the body temperature signal, is the minimum value in the body temperature signal; Linear regression fitting is adopted , where , are regression coefficients, is the error term, and the linear regression method is used to analyze the changing trend of body temperature over time for the body temperature signal; When the physiological signal is a muscle tension signal, feature extraction is performed on the preprocessed muscle tension signal; Let the muscle tension signal sequence of a certain muscle group be , find the maximum value in the muscle tension signal, the mean value of the muscle tension signal is , the variance of the muscle tension signal is , is the number of sampling points of the muscle tension signal, is the th sampling point value of the muscle tension signal; Let the muscle tension signal sequences of different muscle groups be , then the correlation coefficient between the muscle tension signals is , where is the covariance, , are the standard deviations.

[0011] Furthermore, the concatenated fusion of the features extracted from each physiological signal into a multi-modal feature vector includes: The features extracted from heart rate, respiratory rate, body temperature, and muscle tension signals are respectively composed into corresponding feature vectors, and then concatenated into a multi-modal feature vector in sequence.

[0012] Furthermore, for the construction of the transfer learning model, obtaining a public physiological signal dataset, annotating multi-modal feature vectors for each physiological signal in the public physiological signal dataset, setting the transfer learning model to include a source domain model and a target domain model, using the data of the public physiological signal dataset annotated with multi-modal feature vectors as the target domain model data, inputting it into the pre-trained source domain model, constructing a transfer learning adjustment formula according to the feature extractor and the domain classifier, and introducing a gradient reversal layer between the feature extractor and the domain classifier to achieve gradient reversal includes: Set the transfer learning model to include a source domain model and a target domain model , the source domain model The multi-modal feature vectors formed by the heart rate, respiratory rate, body temperature, and muscle tension signals contained in are respectively Then, the multi-modal feature vectors formed by the heart rate, respiratory rate, body temperature, and muscle tension signals contained in the target domain model are respectively Then

[0013] When or is satisfied, a cross-domain mapping function is constructed to improve the performance of the transfer learning model, where is the loss function of the target domain model, are the parameters of the target domain model, is the shared feature, are the parameters of the source domain model, is the regularization term, and the norm constraint [[ID=3l]] and are used to measure the distance between is the balance factor; The Transformer model is selected as the source domain model in the transfer learning model. The calculation formula of the source domain model based on the self-attention mechanism is where , , , Q, K, and V are the query matrix, key matrix, and value matrix respectively, is the key vector dimension, is the scaling factor; The expression ability of the source domain model is enhanced through the multi-head attention mechanism. The calculation formula of the multi-head attention mechanism is where: , , , , , , , are learnable weight matrices, is the number of heads, is the model dimension; During the training process of the source domain model, the cross-entropy loss function is used as the optimization objective. The cross-entropy loss function is where is the number of batch samples, is the number of classes, is the sample belonging to the class true label, is a model prediction sample belongs to the category probability; The transfer learning adjustment formula constructed according to the feature extractor and the domain classifier is , where G is the feature extractor, is the domain classifier; A gradient reversal layer GRL is introduced between the feature extractor and the domain classifier to achieve gradient reversal. The formula is , , is the gradient reversal coefficient, The initial value is 0.1 and linearly increases to 1 with the number of training steps.

[0014] Furthermore, when obtaining the evaluation model by performing transfer learning adjustment on the transfer learning model, it further includes: Updating the target domain model parameters and source domain model parameters of the transfer learning model, where the Adam optimizer is used for parameter update, and the parameter update formula is as follows: ; where, , are the first-order and second-order moment estimates, is the gradient at time , are the decay coefficients, is the learning rate, is a minimum value to prevent the denominator from being 0, is the model parameters at time is the model parameters at time

[0015] Furthermore, the extraction of each physiological signal feature from the preprocessed signal, the concatenation and fusion of the features extracted from each physiological signal into a multi-modal feature vector, and the input into the evaluation model to obtain the classification probability output result include: The evaluation model calculates the input feature vector and outputs the classification probabilities of the state indices of anxiety and depression, emotional stress, fatigue level, and stress sensitivity ; The calculation of the anxiety and depression and emotional stress indices, and the fatigue level and stress sensitivity indices according to the classification probability output result, and the generation of a mental health assessment report according to the anxiety and depression and emotional stress indices, and the fatigue level and stress sensitivity indices include: According to the classification probability of the anxiety and depression and emotional stress indices output by the evaluation model, the first comprehensive index is calculated through the following formula: ; where, respectively represent the probabilities when the anxiety and depression states are normal, mild, moderate, and severe; respectively represent the probabilities when the emotional stress states are low stress, medium stress, and high stress; According to the value of the first comprehensive index, the mental health state level is judged according to the threshold range divided by the preset level. The mental health state level includes good mental health state, mild mental stress, moderate mental stress, and severe mental stress; According to the fatigue level and stress sensitivity index classification probability output by the evaluation model, the second comprehensive index is calculated by the following formula: ; where respectively represent the probabilities when the fatigue levels are low fatigue, medium fatigue, and high fatigue, respectively represent the probabilities when the stress sensitivities are low sensitivity, medium sensitivity, and high sensitivity; According to the value of the second comprehensive index, the physical health state level is judged according to the threshold range divided by the preset level. The physical health state level includes good physical health state, mild fatigue, moderate fatigue, and high fatigue; According to the values of the first comprehensive index and the second comprehensive index calculated, the corresponding mental health state level and the physical health state level are obtained, and a mental health assessment report is generated.

[0016] Further, the mental health assessment method based on the fiber optic sensor further includes: Obtain the comfort state mark fed back by the user, update the label in the classification probability output result to form historical data annotation, and then feedback it to the transfer learning model for retraining to obtain a new evaluation model.

[0017] On the other hand, a computer device is provided, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the steps of the mental health assessment method based on the fiber optic sensor are implemented.

[0018] On the other hand, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by the processor, the steps of the mental health assessment method based on the fiber optic sensor are implemented.

[0019] The above-mentioned mental health assessment method, device, and storage medium based on the fiber optic sensor construct a complete, efficient, and accurate assessment system. After collecting the original physiological signals through the fiber Bragg grating sensor and performing preprocessing, combined with multi-modal feature fusion and transfer learning model, a quantitative assessment system for anxiety and depression, emotional stress, fatigue level, and stress sensitivity is constructed to realize high-precision and real-time monitoring and assessment of the mental health state. Description of the Drawings

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0021] Figure 1 It is an application environment diagram of a mental health assessment method based on an optical fiber sensor in an embodiment of this application; Figure 2 It is a schematic flowchart of a mental health assessment method based on an optical fiber sensor in an embodiment of this application; Figure 3 It is a schematic flowchart of the step of collecting the original physiological signals of the person to be detected through an optical fiber sensor and preprocessing the original physiological signals to obtain preprocessed signals in an embodiment of this application; Figure 4 It is a schematic flowchart of the step of extracting the characteristics of each physiological signal from the preprocessed signals and concatenating and fusing the extracted characteristics of each physiological signal into a multi-modal feature vector in an embodiment of this application; Figure 5 It is a schematic flowchart of training a transfer learning model to obtain an evaluation model in an embodiment of this application; Figure 6 It is a schematic flowchart of generating a mental health assessment report according to the classification probability output result in an embodiment of this application; Figure 7 It is an internal structure diagram of a computer device in an embodiment of this application. Specific embodiments

[0022] In order to make the purpose, technical solutions and advantages of this application clearer, the following further details this application in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain this application and are not used to limit this application.

[0023] To solve the above problems, in the embodiments of the present invention, a mental health assessment method based on fiber optic sensors is creatively proposed. For the first time, fiber optic sensing technology is deeply combined with multi-modal data fusion and transfer learning algorithms and applied to the quantitative assessment of mental health, achieving dual innovations in signal acquisition methods and assessment algorithms, filling the gaps in related technologies. Multi-modal feature fusion comprehensively integrates various physiological signal information, overcoming the one-sidedness of single-modal assessment; transfer learning optimizes the model performance, reduces data dependence, and significantly improves the assessment accuracy compared with traditional methods. The generated quantitative assessment index has good generality and interpretability, and can be widely applied to scenarios such as clinical psychological diagnosis assistance, mental health screening of enterprise employees, psychological state monitoring of students, and psychological assessment of special occupational groups, providing a standardized and objective mental health assessment solution for different fields. This assessment method can realize the early screening and real-time monitoring of mental health status, help reduce the treatment cost of mental diseases, and improve the efficiency of social mental health management; at the same time, it provides personalized mental health guidance for individuals, promotes the improvement of the public's mental health level, and has significant economic and social benefits. This assessment method has good scalability, can increase or adjust the types of physiological signal acquisition, optimize algorithm parameters, and improve the model structure according to actual needs, and is suitable for the mental health assessment needs of different populations and different scenarios, laying a foundation for subsequent technology upgrading and application expansion.

[0024] Embodiment 1 In this embodiment, indices such as anxiety and depression, emotional stress, fatigue level, and stress sensitivity are mainly used as mental health assessment criteria. Currently, everyone's comfort state is different, and everyone's anxiety and depression, emotional stress, fatigue level, and stress sensitivity are different. The present invention uses fiber Bragg grating sensors to obtain the vital sign data of personnel, mark the comfort state, and evaluate the real-time mental health assessment through a transfer learning model.

[0025] The fiber Bragg grating sensor is used to monitor data such as the heart rate, respiratory rate, temperature, and muscle group state of the human body, and the mental health of personnel is evaluated through transfer learning using an existing data set.

[0026] (I) Physiological signal acquisition and preprocessing Signal acquisition part: Heart rate and respiratory rate signal acquisition: Fiber Bragg grating (FBG) sensors are used for acquisition. The working principle of the FBG sensor is based on the Bragg reflection condition of the fiber Bragg grating, and its reflection center wavelength satisfies the formula . Among them, is the wavelength offset, which is a key parameter reflecting the change of physiological signals; is the effective refractive index of the optical fiber, which is determined by the material and structure of the optical fiber; is the grating period, which is an inherent property of the grating; is the elastic-optical coefficient, which is related to the optical properties of the fiber material; In this invention, strain is generated by chest movement due to heartbeat and breathing. When the heart beats and the body breathes, the rise and fall of the chest induces mechanical strain in the attached FBG sensor, which in turn causes slight changes in the grating period and effective refractive index, ultimately shifting the reflection center wavelength. A high-precision fiber optic signal demodulator monitors and measures this wavelength shift in real time, converting this physical change into a processable electrical signal, thereby enabling the acquisition of heart rate and respiratory rate signals.

[0027] Body temperature signal acquisition: Fiber optic temperature sensor is used to measure body temperature, and the formula is: .in, The wavelength change caused by temperature is a quantitative reflection of body temperature signal; is the temperature coefficient, which is an inherent parameter of the optical fiber temperature sensor and characterizes the proportional relationship between wavelength change and temperature change; The temperature change is the change in human body temperature. Fiber optic temperature sensors utilize the thermo-optical effect of optical fiber materials. When human body temperature changes, the fiber's refractive index shifts with the temperature, causing the wavelength of the light transmitted through the fiber to drift. By measuring this wavelength shift and combining it with a known temperature coefficient, the human body temperature can be accurately calculated, enabling the acquisition of temperature signals.

[0028] Muscle tension signal acquisition: Distributed fiber optic strain sensors are used to collect muscle tension signals, sampling at a 50Hz rate from different locations on the optical fiber. These sensors exploit the strain-light effect of optical fibers. When muscles contract or relax, they stretch or compress the optical fiber attached to or near the muscle surface, causing strain. This strain changes the properties of the light transmitted through the fiber. By detecting these changes, signals related to muscle tension can be obtained. A 50Hz sampling rate allows for real-time and accurate capture of changes in muscle tension at different times, providing a rich data foundation for subsequent feature extraction and mental health assessment.

[0029] Data preprocessing unit: Wavelet transform denoising: Wavelet transform is used to denoise the collected original physiological signals. The formula is: In this formula, Indicates signal In scale and translation parameters Wavelet transform coefficients under ; It is the original physiological signal; is a wavelet function. In the present invention, the db4 wavelet basis function is selected, which has good orthogonality, compact support, and certain vanishing moment characteristics, and is suitable for the processing of physiological signals; is a scale parameter used to control the stretching of the wavelet function. Different scales correspond to different frequency resolutions; is a translation parameter used to control the position of the wavelet function on the time axis. By performing 5-layer wavelet decomposition on the original signal, the signal is decomposed into different frequency sub-bands. Among them, noise is usually concentrated in the high-frequency sub-bands, while useful physiological signals are mainly distributed in the low-frequency and part of the intermediate-frequency sub-bands. Soft threshold processing is performed on the high-level detail components (i.e., high-frequency sub-bands). By setting an appropriate threshold, wavelet coefficients smaller than the threshold are set to zero or subjected to shrinkage processing, and then signal reconstruction is performed using the processed wavelet coefficients, thereby effectively removing the noise in the signal and improving the quality of the signal.

[0030] Band-pass filtering: For different physiological signals, corresponding band-pass filters are designed for filtering. Taking the heart rate signal as an example, a band-pass filter of \(0.5 - 4Hz\) is designed, and zero-phase filtering is implemented through the filtfilt function in Matlab. The frequency range corresponding to the normal human heart rate is roughly between \(0.5 - 4Hz\). Designing a band-pass filter in this frequency band can effectively retain the useful components of the heart rate signal while suppressing interference signals in other frequency bands. The filtfilt function uses zero-phase filtering technology, avoiding the phase distortion problem that may be introduced by traditional filtering methods, ensuring that the phase information of the filtered signal does not change, and thus more accurately reflecting the characteristics of the original heart rate signal.

[0031] Normalization processing: According to the formula the filtered signal is normalized to map the signal to the interval \([0,1]\). Among them, is the value of the normalized signal, is the value of the original signal, and are the minimum and maximum values of the original signal respectively. Normalization processing can eliminate the influence of different physiological signals on subsequent processing and analysis due to different dimensions and numerical ranges, making all signals in the same scale space, facilitating the extraction and fusion of multi-modal features, and also helping to improve the training efficiency and accuracy of subsequent machine learning models.

[0032] (2) Multi-modal feature extraction and fusion Feature extraction is performed on signals of multiple parameters, including heart rate, respiratory rate, body temperature, muscle tension, etc.

[0033] Heart rate signal: Time-domain feature mean:

[0034] In the formula, Indicates the number of sampling points, For the The formula is used to calculate the average heart rate signal value at each sampling point. This formula is used to calculate the average heart rate signal over a period of time, which can intuitively reflect the overall heart rate status of the subject during the measurement period. For example, for a normal adult at rest, this average value usually fluctuates between 60 and 100 beats per minute.

[0035] variance:

[0036] This formula measures the fluctuation of the heart rate signal by calculating the average of the squared difference between the signal value at each sampling point and the mean. A larger variance indicates more dramatic heart rate fluctuations, which may indicate stress (such as anxiety or tension) or cardiac dysfunction.

[0037] Frequency domain characteristics: Assume that the power spectrum density of the frequency domain signal is , low-frequency energy:

[0038] After converting the time domain heart rate signal into the frequency domain using Fast Fourier Transform (FFT), Indicates the frequency This formula integrates the power spectral density within the low-frequency range (0.04 - 0.15 Hz) to calculate the low-frequency energy, which primarily reflects the combined effects of the sympathetic and vagus nerves.

[0039] High frequency energy:

[0040] Similar to the calculation of low-frequency energy, this formula integrates the power spectral density within the high-frequency band (0.15 - 0.4Hz) to obtain high-frequency energy. High-frequency energy is mainly related to vagus nerve activity and can reflect the degree of relaxation of the subject.

[0041] LF / HF ratio:

[0042] The ratio of low-frequency energy to high-frequency energy is calculated to assess the balance of the autonomic nervous system. An increased ratio may indicate increased sympathetic nerve activity, indicating a state of tension or anxiety. A decreased ratio indicates a dominant vagus nerve, indicating a state of relaxation.

[0043] Respiratory rate signal: Assume the respiratory cycle duration sequence:

[0044] Period mean:

[0045] is the number of respiratory cycles, indicating the duration of the th respiratory cycle. This formula calculates the average value of the durations of all respiratory cycles, reflecting the average rhythm of the subject's breathing during the measurement period. In normal adults at rest, the average value of the respiratory cycle is generally between 3 and 5 seconds.

[0046] Standard deviation:

[0047] This formula is used to evaluate the stability of the respiratory cycle. By calculating the square root of the average of the squares of the differences between the duration of each respiratory cycle and the mean value, it measures the fluctuation of the breathing rhythm. The larger the standard deviation, the more unstable the breathing rhythm, which may be related to factors such as emotional fluctuations and physical fatigue.

[0048] Rate of change:

[0049] This formula calculates the rate of change of the durations of adjacent respiratory cycles, which can reflect the dynamic change of the breathing frequency in real time. In an emotional stress scenario, such as being suddenly frightened, the rate of change of the breathing frequency usually increases.

[0050] Body temperature signal: Let the body temperature signal sequence be

[0051] Mean value:

[0052] is the number of sampling points of the body temperature signal, is the body temperature value at the th sampling point. This formula calculates the average value of the body temperature during the measurement period. The average value of the normal human body temperature is generally between 36 - 37 °C, and its abnormal change may be related to the physiological reactions caused by physical diseases or psychological stress.

[0053] Fluctuation range:

[0054] By finding the maximum and minimum values in the body temperature signal and calculating the difference between them, the fluctuation range of the body temperature is obtained. A larger fluctuation range may indicate that the body is in a stress state, such as infection, fever, or physiological reactions caused by excessive psychological stress.

[0055] Using linear regression fitting , where , are the regression coefficients, is the error term, reflecting the trend of change. Using the linear regression method to analyze the body temperature signal, As a regression coefficient, it reflects the changing trend of body temperature over time. For example, indicates an upward trend in body temperature, which may signal inflammation in the body or a state of high psychological stress.

[0056] Muscle tension signal: Let the tension signal sequence of a certain muscle group be

[0057] Peak value:

[0058] This formula finds the maximum value in the muscle tension signal, representing the maximum tension generated by the muscle during contraction. Under psychological stress, such as emotions like anxiety and fear, the muscles may tense up involuntarily, causing this peak value to increase.

[0059] Mean value:

[0060] is the number of sampling points of the muscle tension signal, is the signal value of the th sampling point. This formula calculates the average level of muscle tension over a period of time. A long-term high mean value may indicate that the tested person is under long-term stress or fatigue.

[0061] Variance:

[0062] It is used to measure the fluctuation of the muscle tension signal. A larger variance indicates that the muscle tension changes frequently, which may be related to emotional instability or body movement.

[0063] Let the tension signal sequences of different muscle groups be , Correlation coefficient: , where is the covariance, , are the standard deviations Calculate the correlation coefficient between the tension signals of different muscle groups to analyze the cooperative relationship between each muscle group. For example, under an anxious state, the correlation coefficient of the muscle tension signals of multiple muscle groups such as the neck and shoulders may increase, indicating that these muscle groups are simultaneously in a tense state.

[0064] Feature fusion: Let the feature vector of the heart rate signal be , the feature vector of the respiratory rate signal be , the feature vector of the body temperature signal be , the feature vector of the muscle tension signal be , then the multi-modal feature vector:

[0065] The features extracted from the heart rate, respiratory rate, body temperature, and muscle tension signals are respectively composed into corresponding feature vectors, and then concatenated in sequence into a multi-modal feature vector. This vector integrates the key information of multiple physiological signals, providing comprehensive data input for the subsequent mental health assessment model, enabling the model to more accurately reflect the mental health status of the detected person.

[0066] (III) Construction and Training of Transfer Learning Model The transfer learning model aims to address the domain difference problem between the source domain model and the target domain model

[0067] Source domain model : Data source: Public large-scale physiological signal and mental health-related datasets, including but not limited to MIMIC-III (Multiparameter Intelligent Monitoring in Intensive Care Database), BioVid (Bio-signal Video Database), etc. Data composition: Containing various physiological signals (such as ECG, EEG, EDA, etc.) and corresponding mental health annotations (such as depression scale scores, anxiety state classifications, etc.) Domain characteristics: Large data scale and rich annotations, but there are differences in signal acquisition devices and environmental conditions compared with the fiber Bragg grating sensor system of the present invention. Target domain model : Data source: Multi-modal physiological signal data collected by the fiber Bragg grating sensor system of the present invention. Data composition: Multi-modal feature vectors formed by signals such as heart rate, respiratory rate, body temperature, and muscle tension after preprocessing and feature extraction. Domain characteristics: Relatively small data scale, but highly matched with the actual application scenario. The signal acquisition device is the fiber Bragg grating sensor system independently developed by the present invention. When and , satisfying or , the performance of traditional machine learning methods deteriorates. The present invention adopts a method based on model parameter transfer, and constructs a cross-domain mapping function by sharing feature representations

[0068] Where: is the loss function of the target domain model, and the cross-entropy loss is used to measure the difference between the model prediction and the true label; is the target domain model parameter that needs to be optimized on the target domain model data; is the source domain model parameter that carries the knowledge learned by the source domain model; is the regularization term, using norm constraint and distance to control the degree of knowledge transfer; is the balance factor, which is optimized and selected within through cross-validation.

[0069] Source domain model training: The source domain model is trained on a publicly available large-scale physiological signal and mental health related dataset. The Transformer model is selected as the basic network architecture. This model is based on the self-attention mechanism, and its core calculation formula is:

[0070] Where: 、 、 are the query, key, and value matrices respectively; is the key vector dimension, set to 64; is the scaling factor to prevent the dot product result from being too large and causing gradient disappearance; Enhance the model's expressive ability through the multi-head attention mechanism:

[0071] Where:

[0072] 、 、 、 are learnable weight matrices; is the number of heads, set to 8; is the model dimension, set to 512; During the training process, the cross-entropy loss function is used as the optimization objective:

[0073] Where: is the number of batch samples; is the number of categories; is the sample Belonging to the category True label (one - hot encoding); For the model to predict the sample Belonging to the category Probability; Use the Adam optimizer for parameter update, and its update formula is as follows: ; Where: , Are the first - order and second - order moment estimates; Is Gradient at time , Are the decay coefficients; Is the learning rate; Is a very small value to prevent the denominator from being zero; Are the model parameters; Training configuration: Batch size: 32; Number of training epochs: 10 epochs; Learning rate scheduling: Cosine annealing strategy, decaying by 0.95 every epoch; Early stopping mechanism: Stop training if the validation set accuracy does not improve for 3 consecutive epochs.

[0074] Transfer learning adjustment: Use the labeled multi - modal feature vector data collected by the present invention as the target domain model data and input it into the pre - trained source domain model. To reduce the difference between the source domain model and the target domain model, an adversarial domain adaptation method is adopted, and through the domain classifier Implement:

[0075] Where: Is the feature extractor, composed of a Transformer encoder; Is the domain classifier, using a 2 - layer fully - connected network with a hidden layer dimension of 256; Gradient Reversal Layer (GRL): Introduced between the feature extractor and the domain classifier to achieve gradient reversal, and the formula is , , is the gradient reversal coefficient, with an initial value of 0.1, linearly increasing to 1 as the number of training steps increases.

[0076] In terms of parameter adjustment, an incremental strategy is adopted: Stage 1: Freeze the parameters of the first 6 layers of the Transformer, only train the last 3 layers and the fully connected layer, and train for 2 epochs Stage 2: Unfreeze the parameters of the last 3 layers of the Transformer and perform joint fine-tuning, training for 3 epochs Stage 3: Unfreeze all parameters, and perform fine-tuning with a smaller learning rate (1e - 6), training for 2 epochs Optimization objective in the fine-tuning stage: ; where: is the classification loss of the target domain model; is the maximum mean discrepancy, measuring the difference in feature distributions between the source domain model and the target domain model; is the adversarial training loss; , is the balance coefficient.

[0077] Model application: Input the feature vectors newly collected, preprocessed, and with multi-modal feature extraction into the model adjusted by transfer learning. Based on the parameters learned during training, the model calculates the input feature vectors and outputs the classification probabilities of each state of anxiety and depression, emotional stress, fatigue level, and stress sensitivity .

[0078] (IV) Calculation of evaluation indices Anxiety and depression and emotional stress indices According to the model output probabilities, calculate the comprehensive index through the following formula:

[0079] where: are the probabilities of "normal", "mild", "moderate", and "severe" for the anxiety and depression states respectively; are the probabilities of "low stress", "medium stress", and "high stress" for the emotional stress state respectively; Index range: .

[0080] Grade division: : Good mental health status; : Mild psychological stress, it is recommended to adjust appropriately; : Moderate psychological stress, it is recommended to seek professional psychological counseling; : Severe psychological stress, it is recommended to seek medical treatment in time.

[0081] Fatigue level and stress sensitivity index: Calculate the comprehensive index according to the following formula:

[0082] Where: are the probabilities of "low fatigue", "medium fatigue", and "high fatigue" of the fatigue level respectively; are the probabilities of "low sensitivity", "medium sensitivity", and "high sensitivity" of stress sensitivity respectively; Index range: .

[0083] Grade division: : Good physical and mental state, strong stress resistance; : Slight fatigue, moderate stress sensitivity, pay attention to the combination of work and rest; : Moderate fatigue, high stress sensitivity, it is recommended to adjust the work and life rhythm; : High fatigue, extremely high stress sensitivity, it is recommended to rest immediately and seek professional help.

[0084] Generation of assessment report: According to the and index values obtained by calculation, combined with the above grade division criteria, generate a detailed mental health assessment report, including: Quantitative assessment of the current mental health state; Specific analysis of the four dimensions of anxiety and depression, emotional stress, fatigue level, and stress sensitivity; Professional suggestions and intervention measures based on the assessment results; Prediction of mental health trends (based on multiple measurement results).

[0085] In this embodiment, a complete, efficient, and accurate evaluation system is constructed. Physiological signals such as heart rate, respiratory rate, body temperature, and muscle tension are collected through fiber Bragg grating sensors. Signal processing algorithms such as wavelet transform and fast Fourier transform are comprehensively used, combined with multi-modal feature fusion and Transformer transfer learning models, to construct a quantitative evaluation system for anxiety, depression, emotional stress, fatigue level, and stress sensitivity, realizing high-precision and real-time monitoring and evaluation of mental health status.

[0086] Achieve high-precision signal acquisition: Utilize the unique advantages of fiber Bragg grating sensors to solve the defects of traditional electronic sensors, and stably and accurately acquire physiological signals such as heart rate, respiratory rate, body temperature, and muscle tension, providing a reliable data basis for subsequent evaluation.

[0087] Construct a multi-modal quantitative evaluation system: Through multi-modal feature extraction and fusion algorithms, integrate multi-source physiological signal information, and optimize the model performance by combining transfer learning technology to construct quantitative evaluation indices for anxiety, depression, emotional stress, fatigue level, and stress sensitivity, transforming the abstract mental health status into quantifiable and comparable numerical indicators.

[0088] Improve the evaluation efficiency and generalization ability: Based on transfer learning, reduce the model's dependence on large-scale labeled data, improve the model training efficiency and generalization ability, realize rapid and accurate evaluation of mental health status, and meet the application requirements of multiple scenarios such as clinical diagnosis, workplace health management, and educational psychological monitoring.

[0089] Promote the development of mental health assessment technology: Through algorithm innovation and technology integration, provide new technical ideas and methods for the field of mental health assessment, and promote the development of this field towards intelligence and precision.

[0090] Embodiment 2 Embodiment 2 includes all the technical features of Embodiment 1. As Figure 1 , Figure 2 shown, a mental health assessment method based on fiber optic sensors is provided, including the following steps: Step S1, collect the original physiological signals of the person to be detected through a fiber optic sensor, and perform data preprocessing on the original physiological signals to obtain preprocessed signals; Step S2: Construct a transfer learning model, obtain a publicly available physiological signal dataset, label multi-modal feature vectors for each physiological signal in the publicly available physiological signal dataset, set the transfer learning model to include a source domain model and a target domain model, use the data of the publicly available physiological signal dataset labeled with multi-modal feature vectors as the target domain model data, input it into the pre-trained source domain model, construct a transfer learning adjustment formula according to the feature extractor and the domain classifier, introduce a gradient reversal layer between the feature extractor and the domain classifier to achieve gradient reversal, reduce the difference between the source domain model and the target domain model through the domain classifier, and perform transfer learning adjustment on the transfer learning model to obtain an evaluation model; Step S3: Extract features of each physiological signal from the preprocessed signal, concatenate and fuse the features extracted from each physiological signal into a multi-modal feature vector, and input it into the evaluation model to obtain a classification probability output result; Step S4: Calculate the anxiety-depression and emotional stress index and the fatigue level and stress sensitivity index according to the classification probability output result, and generate a mental health assessment report according to the anxiety-depression and emotional stress index and the fatigue level and stress sensitivity index.

[0091] Specifically, construct a complete, efficient and accurate evaluation system. After collecting the original physiological signals through fiber Bragg grating sensors and performing preprocessing, combine multi-modal feature fusion and transfer learning models to construct a quantitative evaluation system for anxiety-depression and emotional stress, and fatigue level and stress sensitivity, so as to realize high-precision and real-time monitoring and evaluation of mental health status.

[0092] As Figure 3 shown, in this embodiment, the collection of the original physiological signals of the person to be detected through the fiber optic sensor and the data preprocessing of the original physiological signals to obtain the preprocessed signal include: Set the collected original physiological signals to include heart rate, respiratory rate, body temperature, and muscle tension physiological signals, and perform wavelet transform denoising, band-pass filtering, and normalization processing on the original physiological signals to obtain the preprocessed signal; When using fiber Bragg grating sensors to collect heart rate and respiratory rate signals, based on the Bragg reflection condition of fiber Bragg gratings, the reflection center wavelength satisfies the formula ; where is the wavelength offset, is the effective refractive index of the optical fiber, is the grating period, is the photoelastic coefficient, is the strain; When using a fiber optic temperature sensor to measure body temperature, the formula used is ; where is the wavelength change caused by temperature, is the temperature coefficient, is the temperature change; When using a distributed fiber optic strain sensor to collect muscle tension signals, strain signals of optical fibers at different positions are collected at a preset sampling rate.

[0093] As Figure 4 shown, in this embodiment, the extraction of various physiological signal features from the preprocessed signal includes: When the physiological signal is a heart rate signal, time-domain features and frequency-domain features of the preprocessed heart rate signal are extracted; When extracting time-domain features of the preprocessed heart rate signal, the mean of the time-domain features is , represents the number of sampling points, is the th sampling point of the heart rate signal value; the variance of the time-domain features is ; When extracting frequency-domain features of the preprocessed heart rate signal, after converting the time-domain heart rate signal to the frequency domain using the fast Fourier transform (FFT), the low-frequency band range c1 - c2 and the high-frequency band range c2 - c3 are obtained. The low-frequency energy obtained by integrating the power spectral density within the low-frequency band is , and the high-frequency energy obtained by integrating the power spectral density within the high-frequency band is , where represents the power spectral density at the frequency of ; By calculating the ratio of the low-frequency energy to the high-frequency energy, it is used to evaluate the balance state of the autonomic nervous system; When the physiological signal is a respiratory frequency signal, features of the preprocessed respiratory frequency signal are extracted; Let the respiratory cycle duration sequence be , then the mean of the respiratory cycle duration is , the standard deviation of the respiratory cycle duration is , and the change rate of adjacent respiratory cycle durations is ; where is the number of respiratory cycles, represents the th respiratory cycle duration; When the physiological signal is a body temperature signal, features of the preprocessed body temperature signal are extracted; Let the body temperature signal sequence be , then the mean of the body temperature signal is , the fluctuation range of the body temperature signal is , is the number of sampling points of the body temperature signal, is the body temperature value at the th sampling point, is the maximum value in the body temperature signal, is the minimum value in the body temperature signal; Linear regression fitting is adopted , where , are regression coefficients, is the error term, and the linear regression method is used to analyze the changing trend of body temperature over time for the body temperature signal; When the physiological signal is a muscle tension signal, feature extraction is performed on the preprocessed muscle tension signal; Suppose a muscle group tension signal sequence is , find the maximum value in the muscle tension signal, the mean value of the muscle tension signal is , the variance of the muscle tension signal is , is the number of sampling points of the muscle tension signal, is the th sampling point of the muscle tension signal value; Suppose different muscle group tension signal sequences are , then the correlation coefficient between different muscle group tension signals is , where is the covariance, , are the standard deviations.

[0094] In this embodiment, the concatenated fusion of the features extracted from each physiological signal into a multi-modal feature vector includes: The features extracted from heart rate, respiratory rate, body temperature, and muscle tension signals are respectively formed into corresponding feature vectors, and then concatenated in sequence into a multi-modal feature vector.

[0095] As Figure 5 shown, in this embodiment, the construction of the transfer learning model, obtaining a public physiological signal dataset, annotating multi-modal feature vectors for each physiological signal in the public physiological signal dataset, setting the transfer learning model to include a source domain model and a target domain model, using the data of the public physiological signal dataset annotated with multi-modal feature vectors as the target domain model data, inputting it into the pre-trained source domain model, constructing a transfer learning adjustment formula according to the feature extractor and the domain classifier, and introducing a gradient reversal layer between the feature extractor and the domain classifier to achieve gradient reversal includes: Set the transfer learning model to include a source domain model and a target domain model , the source domain model The multimodal feature vectors formed by the heart rate, respiratory rate, body temperature, and muscle tension signals included in are respectively The target domain model The multimodal feature vectors formed by the heart rate, respiratory rate, body temperature, and muscle tension signals included in are respectively

[0096] When or is satisfied, a cross-domain mapping function is constructed to improve the performance of the transfer learning model, where is the loss function of the target domain model, are the parameters of the target domain model, is the shared feature, are the parameters of the source domain model, is the regularization term, and the norm constraint and are used to measure the distance between is the balance factor; The Transformer model is selected as the source domain model in the transfer learning model. The calculation formula of the source domain model based on the self-attention mechanism is where , , , Q, K, and V are the query matrix, key matrix, and value matrix respectively, is the dimension of the key vector, is the scaling factor; The expression ability of the source domain model is enhanced through the multi-head attention mechanism. The calculation formula of the multi-head attention mechanism is where: , , , , , , , are learnable weight matrices, is the number of heads, is the model dimension; During the training process of the source domain model, the cross-entropy loss function is used as the optimization objective. The cross-entropy loss function is where is the number of batch samples, is the number of classes, is the sample belongs to the class true label, For the model prediction sample Belonging to the category Probability; The transfer learning adjustment formula constructed according to the feature extractor and the domain classifier is , where G is the feature extractor, Is the domain classifier; Introduce a gradient reversal layer GRL between the feature extractor and the domain classifier to achieve gradient reversal. The formula is , , Is the gradient reversal coefficient, The initial value is 0.1 and linearly increases to 1 with the number of training steps.

[0097] In this embodiment, when obtaining the evaluation model by performing transfer learning adjustment on the transfer learning model, it further includes: Update the target domain model parameters and source domain model parameters of the transfer learning model, where the Adam optimizer is used for parameter update. The parameter update formula is as follows: ; Among them, , Are the first-order and second-order moment estimates, Is The gradient at time , Are the decay coefficients, Is the learning rate, Is a minimum value to prevent the denominator from being zero, Is The model parameters at time Is The model parameters at time

[0098] In this embodiment, for the feature extraction of each physiological signal from the preprocessed signal, the features extracted from each physiological signal are concatenated and fused into a multi-modal feature vector, and input into the evaluation model to obtain the classification probability output result, including: The evaluation model calculates the input feature vector and outputs the classification probabilities of the state indices of anxiety and depression, emotional stress, fatigue level, and stress sensitivity ; As Figure 6 Shown, calculating the anxiety and depression and emotional stress indices, fatigue level and stress sensitivity indices according to the classification probability output result, and generating a mental health assessment report according to the anxiety and depression and emotional stress indices, fatigue level and stress sensitivity indices, including: According to the classification probabilities of the anxiety and depression and emotional stress indices output by the evaluation model, calculate the first comprehensive index through the following formula: ; wherein, respectively represent the probabilities when the anxiety and depression state is normal, mild, moderate, and severe; respectively represent the probabilities when the emotional stress state is low stress, medium stress, and high stress; According to the value of the first comprehensive index, the mental health state level is judged according to the threshold range divided by the preset level. The mental health state level includes good mental health state, mild mental stress, moderate mental stress, and severe mental stress; According to the fatigue level and stress sensitivity index classification probability output by the evaluation model, the second comprehensive index is calculated by the following formula: ; wherein respectively represent the probabilities when the fatigue level is low fatigue, medium fatigue, and high fatigue, respectively represent the probabilities when the stress sensitivity is low sensitivity, medium sensitivity, and high sensitivity; According to the value of the second comprehensive index, the physical health state level is judged according to the threshold range divided by the preset level. The physical health state level includes good physical health state, mild fatigue, moderate fatigue, and high fatigue; According to the calculated values of the first comprehensive index and the second comprehensive index, the corresponding mental health state level and the physical health state level are obtained, and a mental health assessment report is generated.

[0099] In the above mental health assessment method based on fiber optic sensors, a complete, efficient, and accurate assessment system is constructed. After collecting the original physiological signals through fiber Bragg grating sensors and performing preprocessing, combined with the multi-modal feature fusion and transfer learning model, a quantitative assessment system for anxiety and depression, emotional stress, fatigue level, and stress sensitivity is constructed to realize high-precision and real-time monitoring and assessment of the mental health state.

[0100] Achieve high-precision signal acquisition: Utilize the unique advantages of fiber Bragg grating sensors to solve the defects of traditional electronic sensors, and realize the stable and accurate acquisition of physiological signals such as heart rate, respiratory rate, body temperature, and muscle tension, providing a reliable data basis for subsequent evaluation.

[0101] Construct a multi-modal quantitative assessment system: Through multi-modal feature extraction and fusion algorithms, integrate multi-source physiological signal information, and combine transfer learning technology to optimize the model performance, construct quantitative assessment indexes for anxiety and depression, emotional stress, fatigue level, and stress sensitivity, and transform the abstract mental health state into quantifiable and comparable numerical indexes.

[0102] Improve evaluation efficiency and generalization ability: Based on transfer learning, reduce the model's dependence on large-scale labeled data, improve the model training efficiency and generalization ability, achieve rapid and accurate evaluation of mental health status, and meet the application requirements of multiple scenarios such as clinical diagnosis, workplace health management, and educational psychological monitoring.

[0103] Promote the development of mental health assessment technology: Through algorithm innovation and technology integration, provide new technical ideas and methods for the field of mental health assessment, and promote the development of this field towards the direction of intelligence and precision.

[0104] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as Figure 7 shown. The computer device includes a processor, a memory, a network interface, and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store mental health assessment data. The network interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it implements a mental health assessment method based on fiber optic sensors.

[0105] Those skilled in the art can understand that Figure 7 the structure shown in

[0106] is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements. Collect the original physiological signals of the person to be detected through a fiber optic sensor, and perform data preprocessing on the original physiological signals to obtain preprocessed signals; Construct a transfer learning model, obtain a publicly available physiological signal dataset, label multi-modal feature vectors for each physiological signal in the publicly available physiological signal dataset, set the transfer learning model to include a source domain model and a target domain model, use the data of the publicly available physiological signal dataset labeled with multi-modal feature vectors as the target domain model data, input the pre-trained source domain model, construct a transfer learning adjustment formula according to the feature extractor and the domain classifier, and introduce a gradient reversal layer between the feature extractor and the domain classifier to achieve gradient reversal, reduce the difference between the source domain model and the target domain model through the domain classifier, and perform transfer learning adjustment on the transfer learning model to obtain an evaluation model; Extract the features of each physiological signal from the preprocessed signal, concatenate and fuse the features extracted from each physiological signal into a multi-modal feature vector, and input it into the evaluation model to obtain the classification probability output result; Calculate the anxiety-depression and emotional stress index, and the fatigue level and stress sensitivity index according to the classification probability output result, and generate a mental health assessment report according to the anxiety-depression and emotional stress index, and the fatigue level and stress sensitivity index.

[0107] For the specific limitations on the steps implemented when the processor executes the computer program, reference can be made to the limitations on the method for mental health assessment in the above text, which will not be elaborated here.

[0108] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented: Collect the original physiological signals of the person to be detected through an optical fiber sensor, and perform data preprocessing on the original physiological signals to obtain preprocessed signals; Construct a transfer learning model, obtain a publicly available physiological signal dataset, label multi-modal feature vectors for each physiological signal in the publicly available physiological signal dataset, set the transfer learning model to include a source domain model and a target domain model, use the data of the publicly available physiological signal dataset labeled with multi-modal feature vectors as the target domain model data, input the pre-trained source domain model, construct a transfer learning adjustment formula according to the feature extractor and the domain classifier, and introduce a gradient reversal layer between the feature extractor and the domain classifier to achieve gradient reversal, reduce the difference between the source domain model and the target domain model through the domain classifier, and perform transfer learning adjustment on the transfer learning model to obtain an evaluation model; Extract the features of each physiological signal from the preprocessed signal, concatenate and fuse the features extracted from each physiological signal into a multi-modal feature vector, and input it into the evaluation model to obtain the classification probability output result; Calculate the anxiety and depression and emotional stress index, and the fatigue level and stress sensitivity index based on the output result of the classification probability, and generate a mental health assessment report based on the anxiety and depression and emotional stress index, and the fatigue level and stress sensitivity index.

[0109] For the specific limitations on the steps implemented when the computer program is executed by the processor, reference can be made to the limitations on the method for mental health assessment in the above text, and details will not be repeated here.

[0110] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memories can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0111] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0112] The above-described embodiments merely represent several implementation manners of the present application. Their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A mental health assessment method based on fiber optic sensors, characterized in that, Including: Collecting the original physiological signals of the person to be detected through an optical fiber sensor, and performing data preprocessing on the original physiological signals to obtain preprocessed signals; Constructing a transfer learning model, obtaining a public physiological signal dataset, annotating multi-modal feature vectors for each physiological signal in the public physiological signal dataset, setting the transfer learning model to include a source domain model and a target domain model, using the data of the public physiological signal dataset with annotated multi-modal feature vectors as the target domain model data, inputting it into the pre-trained source domain model, constructing a transfer learning adjustment formula according to the feature extractor and the domain classifier, introducing a gradient reversal layer between the feature extractor and the domain classifier to achieve gradient reversal, reducing the difference between the source domain model and the target domain model through the domain classifier, and performing transfer learning adjustment on the transfer learning model to obtain an evaluation model; Performing feature extraction on each physiological signal for the preprocessed signals, concatenating and fusing the features extracted from each physiological signal into a multi-modal feature vector, and inputting it into the evaluation model to obtain a classification probability output result; Calculating the anxiety depression and emotional stress index, and the fatigue level and stress sensitivity index according to the classification probability output result, and generating a mental health assessment report according to the anxiety depression and emotional stress index, and the fatigue level and stress sensitivity index.

2. The mental health assessment method based on an optical fiber sensor according to claim 1, characterized in that The collecting the original physiological signals of the person to be detected through an optical fiber sensor, and performing data preprocessing on the original physiological signals to obtain preprocessed signals includes: Setting the collected original physiological signals to include heart rate, respiratory rate, body temperature, and muscle tension physiological signals, and performing wavelet transform denoising, band-pass filtering, and normalization processing on the original physiological signals to obtain preprocessed signals; When using a fiber Bragg grating sensor to collect heart rate and respiratory rate signals, based on the Bragg reflection condition of the fiber Bragg grating, its reflection center wavelength satisfies the formula ; where is the wavelength offset,[ is the effective refractive index of the optical fiber,[ is the grating period,[ is the photoelastic coefficient,[ is the strain; When using an optical fiber temperature sensor for body temperature measurement, the formula relied on is ; where is the wavelength change caused by temperature, is the temperature coefficient, is the temperature change amount; When using a distributed optical fiber strain sensor to collect muscle tension signals, collecting strain signals of optical fibers at different positions according to a preset sampling rate.

3. The mental health assessment method based on an optical fiber sensor according to claim 2, wherein The performing feature extraction on each physiological signal for the preprocessed signals includes: When the physiological signal is a heart rate signal, performing time-domain feature extraction and frequency-domain feature extraction on the preprocessed heart rate signal; When extracting time-domain features from the preprocessed heart rate signal, the mean of the time-domain features is , represents the number of sampling points, is the heart rate signal value at the -th sampling point; the variance of the time-domain features is ; When extracting the frequency-domain features of the preprocessed heart rate signal, after converting the time-domain heart rate signal to the frequency domain using the Fast Fourier Transform (FFT), the low-frequency band range c1 - c2 and the high-frequency band range c2 - c3 are obtained. The low-frequency energy obtained by integrating the power spectral density within the low-frequency band is , and the high-frequency energy obtained by integrating the power spectral density within the high-frequency band is , where represents the power spectral density at the frequency ; By calculating the ratio of low-frequency energy to high-frequency energy , which is used to evaluate the balance state of the autonomic nervous system; When the physiological signal is a respiratory rate signal, performing feature extraction on the preprocessed respiratory rate signal; Let the sequence of respiratory cycle durations be , then the mean of the respiratory cycle durations is , the standard deviation of the respiratory cycle durations is , and the change rate of adjacent respiratory cycle durations is ; where is the number of respiratory cycles, represents the duration of the th respiratory cycle; When the physiological signal is a body temperature signal, performing feature extraction on the preprocessed body temperature signal; Let the body temperature signal sequence be , then the mean value of the body temperature signal is , and the fluctuation range of the body temperature signal is , is the number of sampling points of the body temperature signal, is the th body temperature value at the sampling point, is the maximum value in the body temperature signal, is the minimum value in the body temperature signal; Using linear regression fitting , where , are regression coefficients, is the error term. The linear regression method is used to analyze the changing trend of body temperature over time for the body temperature signal; When the physiological signal is a muscle tension signal, performing feature extraction on the preprocessed muscle tension signal; Suppose a muscle group tension signal sequence is , find the maximum value in the muscle tension signal . The mean value of the muscle tension signal is , and the variance of the muscle tension signal is . is the number of sampling points of the muscle tension signal, is the muscle tension signal value at the -th sampling point; Let the tension signal sequences of different muscle groups be , then the correlation coefficient between the tension signals of different muscle groups is , where is the covariance, , are the standard deviations.

4. The mental health assessment method based on an optical fiber sensor according to claim 2, wherein The concatenating and fusing the features extracted from each physiological signal into a multi-modal feature vector includes: Respectively forming corresponding feature vectors from the features extracted from heart rate, respiratory rate, body temperature, and muscle tension signals, and then concatenating them in order into a multi-modal feature vector.

5. The mental health assessment method based on fiber optic sensors according to claim 1, wherein Construct the transfer learning model, obtain a publicly available physiological signal dataset, label multi-modal feature vectors for each physiological signal in the publicly available physiological signal dataset, set the transfer learning model to include a source domain model and a target domain model, use the data of the publicly available physiological signal dataset labeled with multi-modal feature vectors as the target domain model data, input it into the pre-trained source domain model, construct a transfer learning adjustment formula according to the feature extractor and the domain classifier, and introduce a gradient reversal layer between the feature extractor and the domain classifier to achieve gradient reversal, including: The set migration learning model includes a source domain model and a target domain model . The multi-modal feature vectors formed after preprocessing and feature extraction of the heart rate, respiratory rate, body temperature, and muscle tension signals included in the source domain model are respectively . Then . The multi-modal feature vectors formed after preprocessing and feature extraction of the heart rate, respiratory rate, body temperature, and muscle tension signals included in the target domain model are respectively . Then ; When or is satisfied, a cross-domain mapping function is constructed to improve the performance of the transfer learning model, where is the target domain model loss function, are the target domain model parameters, is the shared feature, are the source domain model parameters, is the regularization term, using norm constraint and distance, is the balance factor; Select the Transformer model as the source domain model in the transfer learning model. The calculation formula of the source domain model based on the self-attention mechanism is , where , , , Q, K, and V are the query matrix, key matrix, and value matrix respectively, is the key vector dimension, is the scaling factor; Enhance the expressive ability of the source domain model through the multi-head attention mechanism. The calculation formula of the multi-head attention mechanism is , where: , , , , , , , are learnable weight matrices, is the number of heads, is the model dimension; During the training process of the source domain model, the cross-entropy loss function is used as the optimization objective. The cross-entropy loss function is , where is the number of batch samples, is the number of categories, is the sample belonging to the category true label, is the probability that the model predicts that the sample belongs to the category ; The transfer learning adjustment formula constructed based on the feature extractor and the domain classifier is , where G is the feature extractor, is the domain classifier; Introduce a Gradient Reversal Layer (GRL) between the feature extractor and the domain classifier to achieve gradient reversal. The formula is , , where is the gradient reversal coefficient, with an initial value of 0.1, linearly increasing to 1 as the number of training steps increases.

6. The mental health assessment method based on an optical fiber sensor according to claim 5, characterized in that, When performing transfer learning adjustment on the transfer learning model to obtain an evaluation model, it further includes: Update the parameters of the target domain model and the source domain model of the transfer learning model, where the Adam optimizer is used for parameter update, and the parameter update formula is as follows: ； Among them, and are the first-order and second-order moment estimates, is the gradient at time , and are the attenuation coefficients, is the learning rate, is the model parameters at time and are the model parameters at time 7. The mental health assessment method based on an optical fiber sensor according to claim 1, wherein Extract the features of each physiological signal from the preprocessed signal, concatenate and fuse the features extracted from each physiological signal into a multi-modal feature vector, and input it into the evaluation model to obtain the classification probability output result, including: The evaluation model calculates the input feature vectors and outputs the classification probabilities of the state indices of anxiety and depression, emotional stress, fatigue level, and stress sensitivity , , , ; Calculate the anxiety-depression and emotional stress index, and the fatigue level and stress sensitivity index according to the classification probability output result, and generate a mental health assessment report according to the anxiety-depression and emotional stress index, and the fatigue level and stress sensitivity index, including: According to the classification probabilities of anxiety, depression, and emotional stress index output by the evaluation model, the first comprehensive index is calculated through the following formula: ; where respectively represent the probabilities when the anxiety and depression states are normal, mild, moderate, and severe; respectively represent the probabilities when the emotional stress states are low stress, medium stress, and high stress; Judge the mental health status level according to the value of the first comprehensive index according to the threshold range divided by the preset level. The mental health status levels include good mental health status, mild mental stress, moderate mental stress, and severe mental stress; According to the fatigue level and stress sensitivity index classification probability output by the evaluation model, the second comprehensive index is calculated through the following formula: ; where respectively represent the probabilities when the fatigue level is low fatigue, medium fatigue, and high fatigue, respectively represent the probabilities when the stress sensitivity is low sensitivity, medium sensitivity, and high sensitivity; Judge the physical health status level according to the value of the second comprehensive index according to the threshold range divided by the preset level. The physical health status levels include good physical health status, slight fatigue, moderate fatigue, and high fatigue; Obtain the corresponding mental health status level and physical health status level according to the calculated values of the first comprehensive index and the second comprehensive index, and generate a mental health assessment report.

8. The mental health assessment method based on an optical fiber sensor according to claim 1, characterized in that The method further includes: Obtain the comfort state mark feedback by the user, update the label in the classification probability output result to form historical data annotation, and then feedback it into the transfer learning model for retraining to obtain a new evaluation model.

9. A computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1 to 7.

Images