Physiological signal acquisition method, device, equipment and medium in marine health care

Abstract

The invention relates to a physiological signal acquisition method, device and equipment in marine health care and a medium. The method comprises the following steps: collecting preliminary physiological data of a user through wearable equipment in a marine health environment, and carrying out noise suppression processing on the preliminary physiological data to obtain purified physiological data; according to the purification physiological data, performing time-frequency domain feature extraction processing on the purification physiological data to obtain multi-dimensional feature data; inputting the multi-dimensional feature data into a preset multi-level recognition model, and performing physiological state pattern recognition processing on the multi-dimensional feature data to obtain pattern recognition data; performing multi-source data fusion processing on the pattern recognition data and the purified physiological data to obtain fused physiological data; and performing health-care state evaluation processing on the fused physiological data to obtain physiological signal data. By adopting the method, the problems that the signals are easily polluted by noise, the state feature extraction is incomplete and the evaluation is inaccurate in the ocean health environment can be effectively solved.

Description

Technical Field

[0001] This invention belongs to the field of health management technology, and in particular relates to methods, devices, equipment and media for collecting physiological signals in marine health care. Background Technology

[0002] In the field of health management and wellness, marine-based health and wellness, as an innovative approach combining the natural environment with human health, is receiving increasing attention. It utilizes the unique marine environment, such as sea breezes, seawater, and negative oxygen ions, to provide physical and mental conditioning, offering significant health-promoting value. However, research and application in this field are still in their developmental stages, urgently requiring technological support to enhance the scientific rigor and personalization of its effects, thereby meeting the growing demand for health management.

[0003] While some methods exist to collect human data to aid in health and wellness, these methods often face adaptability issues in complex environments. Particularly in marine environments, external disturbances such as sea breezes and tidal sounds can easily contaminate the collected data, significantly reducing the accuracy of subsequent analyses. Furthermore, current technologies often fail to comprehensively capture subtle changes in the human body's state at different times and frequencies when processing this data, making it difficult to accurately reflect the body's dynamic responses during health and wellness activities.

[0004] A deeper technical challenge lies in effectively purifying the collected signals in complex environments and extracting diverse characteristics that represent the human body's condition. Interference factors in the marine environment directly reduce signal purity; for example, heartbeat or respiratory signals may be masked by environmental noise, thus affecting the assessment of the body's condition. If the issue of signal purity cannot be resolved, it will be difficult to further uncover the hidden patterns and characteristics behind the signals, such as fluctuations in heartbeat rhythm or subtle changes in respiratory rate—all crucial indicators for assessing the effectiveness of health and wellness programs.

[0005] Therefore, overcoming environmental interference in signal acquisition within the unique context of marine health and wellness, and extracting multi-dimensional feature information from purified signals to accurately reflect human health status, has become a critical issue that urgently needs to be addressed. Solving this problem not only concerns the reliability of data acquisition but also directly impacts the personalized design and actual effectiveness of health and wellness programs. Summary of the Invention

[0006] Therefore, it is necessary to provide methods, devices, equipment, and media for collecting physiological signals in marine health and wellness to address the aforementioned technical issues.

[0007] Firstly, this application provides a method for collecting physiological signals in marine health and wellness, including:

[0008] S1. In a marine health and wellness environment, wearable devices are used to collect preliminary physiological data from users, and noise suppression processing is applied to the preliminary physiological data to obtain purified physiological data. The preliminary physiological data includes electrocardiogram waveform data, blood oxygen waveform data, and respiratory waveform data.

[0009] S2. Based on the purification physiological data, perform time-frequency domain feature extraction processing on the purification physiological data to obtain multi-dimensional feature data; among which, the multi-dimensional feature data is used to characterize the fluctuation pattern and rhythm characteristics of physiological signals in the time domain and frequency domain.

[0010] S3. Input the multi-dimensional feature data into the preset multi-level recognition model, perform physiological state pattern recognition processing on the multi-dimensional feature data, and obtain pattern recognition data; wherein, the structure of the preset multi-level recognition model includes a shallow feature extraction layer and a deep pattern recognition layer connected in sequence.

[0011] S4. Perform multi-source data fusion processing on pattern recognition data and purification physiological data to obtain fused physiological data;

[0012] S5. Perform health and wellness status assessment processing on the fused physiological data to obtain physiological signal data; among which, the physiological signal data is used to characterize the user's real-time physiological status and health and wellness effects in the marine health and wellness environment.

[0013] In one embodiment, preliminary physiological data of users is collected via wearable devices in a marine health and wellness environment, and noise suppression processing is applied to the preliminary physiological data to obtain purified physiological data, including:

[0014] S11. Collect ECG waveform data through the ECG sensor in the wearable device to obtain the raw ECG signal;

[0015] S12. Collect blood oxygen waveform data through the blood oxygen sensor in the wearable device to obtain the raw blood oxygen signal;

[0016] S13. Collect respiratory waveform data through the respiratory sensor in the wearable device to obtain the raw respiratory signal;

[0017] S14. Perform wavelet denoising on the raw electrocardiogram signal, raw blood oxygen signal and raw respiratory signal to obtain the denoised physiological signal;

[0018] S15. Baseline drift correction is performed on the denoised physiological signals to obtain purified physiological data.

[0019] In one embodiment, based on the purification physiological data, time-frequency domain feature extraction processing is performed on the purification physiological data to obtain multi-dimensional feature data, including:

[0020] S21. Extract time-domain features from the purification physiological data to obtain time-domain feature data;

[0021] S22. The purification physiological data are transformed from the time domain to the frequency domain through the spectrum analysis method, and frequency domain feature data representing the rhythm distribution are extracted.

[0022] S23. The time-domain feature data and frequency-domain feature data are concatenated and normalized to form multi-dimensional feature data.

[0023] In one embodiment, time-domain features are extracted from the purification physiological data to obtain time-domain feature data, including:

[0024] S31. Calculate the mean of the purification physiological data to obtain the time-domain mean characteristics;

[0025] S32. Calculate the variance of the purification physiological data to obtain the time-domain variance characteristics;

[0026] S33. Perform peak detection on the purification physiological data to obtain the time-domain peak characteristics; wherein, the expression for peak detection is:

[0027]

[0028] In the formula, Indicates the peak characteristics in the time domain. Indicates the number of signal sampling points. Indicates the first The signal values ​​at each sampling point;

[0029] S34. Perform feature fusion on the time-domain mean feature, time-domain variance feature, and time-domain peak feature to obtain the time-domain feature vector.

[0030] In one embodiment, multi-dimensional feature data is input into a preset multi-level recognition model, and physiological state pattern recognition processing is performed on the multi-dimensional feature data to obtain pattern recognition data, including:

[0031] S41. Input the multi-dimensional feature data into the shallow feature extraction layer for shallow feature extraction processing to obtain shallow feature data;

[0032] S42. Input the shallow feature data into the convolutional neural network of the deep pattern recognition layer for convolutional feature extraction processing to obtain convolutional feature data;

[0033] S43. Input the convolutional feature data into the recurrent neural network of the deep pattern recognition layer for temporal feature extraction processing to obtain temporal feature data;

[0034] S44. Input the temporal feature data into the fully connected layer of the deep pattern recognition layer for pattern classification processing to obtain pattern recognition data.

[0035] In one embodiment, the fused physiological data is processed for health and wellness status assessment to obtain physiological signal data, including:

[0036] S51. Extract and process health and wellness indicators from the fused physiological data to obtain health and wellness indicator data;

[0037] S52. Calculate and score the health and wellness indicator data to obtain health and wellness score data;

[0038] S53. Perform trend analysis on the health and wellness scoring data to obtain health and wellness trend data;

[0039] S54. Generate a health and wellness report based on health and wellness score data and health and wellness trend data, and obtain physiological signal data.

[0040] In one embodiment, the health and wellness indicator data are scored and calculated to obtain health and wellness score data, including:

[0041] S61. Calculate the normalized values ​​of each health and wellness indicator based on the health and wellness indicator data to obtain normalized indicator data;

[0042] S62. Based on the preset weighting coefficients, the normalized index data are weighted and summed to obtain a comprehensive health and wellness score.

[0043] The expression for the comprehensive health and wellness score is as follows:

[0044]

[0045] In the formula, Indicates the overall health and wellness score. Indicates the first The weighting coefficients of each health and wellness indicator. Indicates the first A normalized health and wellness indicator This represents the total number of health and wellness indicators;

[0046] S63. Classify the comprehensive health and wellness score into levels to obtain health and wellness level data;

[0047] S64. Generate a scoring report based on the comprehensive health and wellness score and health and wellness level data to obtain health and wellness score data.

[0048] Secondly, this application also provides a physiological signal acquisition device for marine health and wellness, comprising:

[0049] The preliminary data acquisition and noise reduction module is used to collect preliminary physiological data of users through wearable devices in a marine health and wellness environment, and to perform noise suppression processing on the preliminary physiological data to obtain purified physiological data; among which, the preliminary physiological data includes electrocardiogram waveform data, blood oxygen waveform data, and respiratory waveform data;

[0050] The data feature processing module is used to extract time-frequency domain features from the purified physiological data to obtain multi-dimensional feature data. The multi-dimensional feature data is used to characterize the fluctuation pattern and rhythm characteristics of physiological signals in the time and frequency domains.

[0051] The data pattern recognition module is used to input multi-dimensional feature data into a preset multi-level recognition model, perform physiological state pattern recognition processing on the multi-dimensional feature data, and obtain pattern recognition data; wherein, the preset multi-level recognition model includes a shallow feature extraction layer and a deep pattern recognition layer connected in sequence.

[0052] The multi-source data fusion module is used to perform multi-source data fusion processing on pattern recognition data and purification physiological data to obtain fused physiological data.

[0053] The physiological signal data module is used to process the fused physiological data to assess the health and wellness status and obtain physiological signal data. The physiological signal data is used to characterize the user's real-time physiological status and health and wellness effects in the marine health and wellness environment.

[0054] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect.

[0055] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0056] The aforementioned method, device, equipment, and medium for collecting physiological signals in marine health and wellness utilize wearable devices to collect preliminary physiological data such as electrocardiogram waveforms, blood oxygen waveforms, and respiratory waveforms from users in a marine health and wellness environment. First, noise suppression processing is applied to the data to obtain purified physiological data. Then, time-frequency domain feature extraction is performed based on the purified physiological data to obtain multi-dimensional feature data that characterizes the morphology and rhythmic properties of signal fluctuations. Subsequently, the multi-dimensional feature data is input into a multi-layered recognition model consisting of a shallow feature extraction layer and a deep pattern recognition layer for physiological state pattern recognition to obtain pattern recognition data. Next, the pattern recognition data and purified physiological data are fused from multiple sources to obtain fused physiological data. Finally, the fused physiological data undergoes health and wellness status assessment processing to obtain physiological signal data. This effectively overcomes the influence of interference such as sea breezes and tidal sounds in the marine environment on signal acquisition, improves the purity of physiological data, comprehensively captures the temporal and frequency-related changes in human body status, and accurately reflects the real-time physiological state and health and wellness effects of users in a marine health and wellness environment, providing reliable data support for the personalized design of health and wellness programs. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is a flowchart illustrating a physiological signal acquisition method in marine health and wellness, as shown in one embodiment.

[0059] Figure 2 This is a schematic diagram of the physiological signal acquisition device in marine health care in one embodiment. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0061] refer to Figure 1 The application presents a flowchart illustrating the physiological signal acquisition method for marine health and wellness provided in this application, which includes the following steps:

[0062] S1. In a marine health and wellness environment, wearable devices are used to collect preliminary physiological data from users, and noise suppression processing is applied to the preliminary physiological data to obtain purified physiological data.

[0063] Optionally, preliminary physiological data may include electrocardiogram waveform data, blood oxygen waveform data, and respiratory waveform data.

[0064] Optionally, the wearable device used needs to be specifically designed for the high humidity, high salt spray, and dynamic interference characteristics of the marine health and wellness environment. The main body of the device is encapsulated in stainless steel resistant to salt spray corrosion, and integrates a high-precision physiological sensor module and an anti-interference data transmission unit. Specifically, it includes a dry electrode ECG sensor for collecting ECG waveform data, a photoplethysmography sensor for collecting blood oxygen waveform data, and a piezoelectric thin-film respiratory sensor for collecting respiratory waveform data. The sampling frequency of each sensor is set to meet the requirements for high-frequency detail capture of physiological signals. In terms of wearing method, the ECG dry electrode is integrated on the inside of the wristband corresponding to the radial artery and the projection position of the apex of the heart. The photoplethysmography sensor adopts a transmissive design and is embedded in the ring-type acquisition end to ensure the stability of blood oxygen saturation detection. The respiratory sensor is integrated into a chest band module, which achieves accurate sensing of respiratory movements by fitting snugly against the chest.

[0065] During data collection, the wearable device establishes communication with the edge computing terminal via a low-power Bluetooth module, transmitting the real-time raw data to the terminal for preprocessing. The preprocessing stage first performs baseline drift correction, employing a high-pass filter to remove low-frequency drift interference caused by human movement and device shaking. This filter is based on an infinite impulse response Butterworth filter, with its order set to balance filtering effectiveness and signal distortion. For high-frequency noise in the marine environment, such as sea breeze and tidal sounds, an adaptive noise cancellation algorithm is used. This algorithm uses environmental vibration data collected by the device's built-in triaxial accelerometer as a reference noise signal, adjusting the filter coefficients in real-time to significantly improve the signal-to-noise ratio after cancellation. For potential electromyographic interference in the ECG waveform, a wavelet thresholding method is used. A specific base wavelet is selected to decompose the signal into multiple layers, and an improved soft thresholding function is applied to the high-frequency coefficients after decomposition. The threshold calculation formula is set as follows: ,in This represents the calculated threshold. This represents the noise standard deviation, which is estimated using the median absolute deviation of the signal. This represents the length of the signal to be processed. After the above multi-stage noise suppression processing, the final purified physiological data can effectively retain key features such as the ECG R wave, the main peak of the blood oxygen pulse wave, and the respiratory waveform cycle.

[0066] S2. Based on the purification physiological data, perform time-frequency domain feature extraction processing on the purification physiological data to obtain multi-dimensional feature data.

[0067] Optionally, multidimensional feature data can be used to characterize the fluctuation patterns and rhythmic characteristics of physiological signals in the time and frequency domains.

[0068] Optionally, time-domain feature extraction designs feature parameters for the different physiological meanings of ECG, blood oxygenation, and respiration signals. For ECG waveform data, core extraction indicators include RR interval, R-wave peak voltage, QRS complex width, and heart rate variability-related parameters. R-wave peak voltage refers to the voltage value corresponding to the R-wave peak within the QRS complex in the ECG signal, and is an important parameter reflecting the intensity of myocardial depolarization. QRS complex width refers to the time span from the start of the Q-wave to the end of the S-wave in the ECG signal, used to assess the speed and synchronicity of ventricular depolarization. RR interval refers to the time interval between the R-wave peaks within two adjacent QRS complexes in the ECG signal, and is a core indicator reflecting heart rate and heart rate variability. The heart rate variability parameter is further subdivided into time-domain statistics and nonlinear characteristics. The time-domain statistics include the standard deviation of the RR interval and the root mean square of the difference between adjacent RR intervals. The nonlinear characteristics are achieved by calculating the standard deviation of the short axis and the standard deviation of the long axis of the Poincaré plot. For blood oxygen waveform data, the focus is on extracting the mean blood oxygen saturation, pulse wave conduction time, the difference between the peak and trough of the pulse wave, and the slope of the rising edge of the pulse wave. The pulse wave conduction time is calculated by the time difference between the synchronously acquired ECG R wave and the peak of the blood oxygen pulse wave, and the slope of the rising edge is obtained by linear fitting of the rising segment of the pulse wave. For respiratory waveform data, the main parameters extracted are respiratory rate, respiratory depth, the ratio of inspiratory time to expiratory time, and respiratory rhythm stability.

[0069] Optionally, frequency domain feature extraction employs a combination of short-time Fourier transform and wavelet packet transform. First, the purified physiological data is segmented into analysis windows of fixed duration, with the window overlap rate set to avoid feature loss. Short-time Fourier transform is performed on the signal within each window, using a Hanning window as the window function to reduce spectral leakage. For ECG signals, power spectral densities are extracted for low-frequency, mid-frequency, and high-frequency bands; for blood oxygen signals, power spectral densities are extracted for the pulse wave's baseline frequency band and the high-frequency noise-suppressed band; and for respiratory signals, power spectral density is extracted for the normal respiratory frequency band. Simultaneously, specific wavelet packets are used to perform multi-level decomposition on the three signals, calculating the energy entropy and singular entropy of each decomposition node. Energy entropy is calculated as the proportion of energy at each node to the total energy, while singular entropy is obtained based on the singular value distribution after singular value decomposition. The time-domain and frequency-domain feature parameters are normalized, and the minimum-maximum normalization method is used to map all feature values ​​to a fixed interval, ultimately forming a feature data matrix containing multiple dimensions. This matrix fully represents the amplitude and periodicity characteristics of physiological signals in the time domain and the energy distribution pattern in the frequency domain.

[0070] S3. Input the multi-dimensional feature data into the preset multi-level recognition model, perform physiological state pattern recognition processing on the multi-dimensional feature data, and obtain pattern recognition data.

[0071] Optionally, the structure of the preset multi-level recognition model includes a shallow feature extraction layer and a deep pattern recognition layer connected in sequence.

[0072] Optionally, the preset multi-level recognition model is built on a deep learning framework and implemented using a mainstream deep learning development platform. The training data of the model comes from physiological signal samples of subjects of different ages in a marine health and wellness environment. The samples cover a variety of scenarios such as calm rest, light activity, and health and wellness intervention. The sample size in each scenario meets the training requirements of the model. Furthermore, the samples are subjected to data augmentation processing by adding amplitude perturbation, time offset, and low-frequency noise to simulate signal changes in different marine environments in order to improve the generalization ability of the model.

[0073] Optionally, the shallow feature extraction layer consists of multiple convolutional neural network modules. The first convolutional neural network module contains a specific number of convolutional kernels and uses the ReLU activation function to extract local correlation information in the feature data, followed by a max pooling layer to reduce the feature dimensionality. The second convolutional neural network module increases the number of convolutional kernels compared to the first module, while maintaining the same kernel size, and also uses the ReLU activation function and a max pooling layer. The third convolutional neural network module sets even more convolutional kernels and is subsequently connected to a batch normalization layer and a max pooling layer. The batch normalization layer accelerates model convergence and alleviates overfitting. After processing by the shallow feature extraction layer, the dimensionality of the feature data is significantly reduced, and then it is converted into a one-dimensional feature vector through a flattening operation and input to the deep pattern recognition layer.

[0074] Optionally, the deep pattern recognition layer adopts a structure combining a bidirectional long short-term memory network (BSSN) and a fully connected layer. The BSSN layer contains a specific number of hidden units, and captures the positive and negative dependencies of feature sequences through forward and backward BSSN networks, respectively, effectively uncovering the dynamic changes of physiological features over time. The output of the BSSN layer is regularized through a dropout layer to avoid model overfitting, with the dropout rate set to balance model performance and generalization ability. The fully connected layer consists of two levels. The first level fully connected layer contains a specific number of neurons and uses the ReLU activation function. The second level fully connected layer contains neurons corresponding to the number of physiological state categories and uses the softmax activation function. The output corresponds to various physiological state categories, namely, relaxation, mild excitement, fatigue, stress, and stable state.

[0075] The model training process employs the cross-entropy loss function and uses the Adam optimizer. The initial learning rate is set according to standard model training settings, and is adjusted using a learning rate decay strategy. The number of training iterations is set to a fixed value, and training stops when the validation set accuracy shows no improvement for several consecutive iterations. Multi-dimensional feature data is input into the trained multi-level recognition model, which outputs the corresponding category probability distribution. The category with the highest probability and its corresponding confidence score are selected as the pattern recognition data. The confidence threshold is set according to the recognition reliability requirements; when the confidence score is below the threshold, it is marked as pending confirmation to improve the reliability of the recognition results.

[0076] S4. Perform multi-source data fusion processing on pattern recognition data and purification physiological data to obtain fused physiological data.

[0077] Optionally, a multi-source data fusion method based on DS evidence theory is employed. This method can effectively handle uncertain data and achieve complementary fusion of pattern recognition results and original physiological data. First, the data sources for fusion are clearly defined, including the ECG RR interval sequence, real-time blood oxygen saturation value, and respiratory rate sequence from the purified physiological data, as well as the physiological state category and confidence level from the pattern recognition data. Each data source is treated as an independent evidence body, and a basic probability allocation function is constructed for each evidence body. For continuous parameters in the purified physiological data, intervals are first divided. For example, blood oxygen saturation is divided into three intervals: normal, slightly low, and low; respiratory rate is divided into three intervals: slow, normal, and fast; and the heart rate corresponding to the RR interval is divided into three intervals: slow, normal, and fast. The basic probability allocation value is determined based on the probability that the actual value of each parameter falls within the corresponding interval. For example, when blood oxygen saturation is within the normal range, its basic probability allocation value in the normal interval is set to a higher level; in the slightly low interval, it is set to a lower level; and in the low interval, it is set to zero. For pattern recognition data, physiological state categories are used as focal elements in the recognition framework, with their basic probability allocation values ​​set as the corresponding confidence levels. The remaining confidence levels are assigned to uncertain focal elements. For example, when the pattern recognition result is a relaxed state and the confidence level reaches a certain level, the basic probability allocation value for the relaxed state is that confidence level, and the basic probability allocation value for uncertain focal elements is the difference between the total confidence level and that confidence level. During evidence fusion, the consistency of the basic probability allocation functions of each piece of evidence is first checked. The distance between any two pieces of evidence is calculated using the evidence distance formula, Jousselme distance. When the average distance between all pieces of evidence is less than a set threshold, the evidence is considered to have good consistency, and DS combination rules are directly applied for fusion. When the average distance is greater than or equal to the threshold, a weighted average method is used to correct the basic probability allocation values ​​of each piece of evidence. The correction weights are determined based on the credibility of the evidence, with the highest credibility weight set for pattern recognition data, followed by ECG data, blood oxygen data, and respiration data, in that order, to highlight the dominant role of the pattern recognition results while also considering the reference value of the original physiological data. During the fusion process, iterative calculations are used to synthesize multiple evidence bodies, ultimately yielding the fused basic probability allocation result. The physiological state information corresponding to the focal element with the largest basic probability allocation value is selected, and combined with key parameters from the purified physiological data, a fused physiological data structure is constructed. This structure includes state category, state confidence level, core physiological parameters, and parameter acquisition timestamps, ensuring data integrity and timeliness.

[0078] S5. Perform health and wellness status assessment on the fused physiological data to obtain physiological signal data.

[0079] Optionally, physiological signal data can be used to characterize the user's real-time physiological state and health benefits in a marine health environment.

[0080] Optionally, the health and wellness status assessment adopts a three-tiered evaluation system of indicator quantification, weight allocation, and comprehensive scoring. First, evaluation indicators are extracted based on integrated physiological data. These indicators are divided into two main categories: basic physiological indicators and health and wellness effect indicators. Basic physiological indicators include mean heart rate, mean blood oxygen saturation, mean respiratory rate, and the SDNN value (Standard Deviation of NN Intervals, the standard deviation of all sinus RR intervals (NN intervals)). Health and wellness effect indicators are constructed based on physiological state categories and indicator change trends, specifically including a state stability index, indicator improvement rate, and stress relief degree. The state stability index refers to the number of changes in physiological state categories within a continuous fixed time period; the indicator improvement rate is the percentage difference between the current indicator and the initial health and wellness indicator relative to the initial indicator; and the stress relief degree is the inverse indicator of the corresponding stress state confidence level. Each evaluation indicator is quantified and graded using a ten-point scoring system. A perfect score is awarded for a mean heart rate within the normal range, with points deducted for deviations. A perfect score is awarded for a mean blood oxygen saturation within the normal range, with points deducted for deviations. A perfect score is awarded for a mean respiratory rate within the normal range, with points deducted for deviations. A perfect score is awarded for a SDNN value within the normal range, with points deducted for deviations. A perfect score is awarded for a state stability index of zero, with points deducted for each additional change. A perfect score is awarded for an indicator improvement rate reaching a set improvement level, a moderate score is awarded within a certain improvement range, and zero score is awarded for no improvement or a decrease. The degree of stress relief is calculated by subtracting the stress state confidence level from one and then multiplying by ten to obtain a score out of ten. Regarding weight allocation, the analytic hierarchy process (AHP) is used to determine the weights of each indicator. By constructing a judgment matrix and calculating eigenvectors, the total weights of the basic physiological indicators and the health and wellness effect indicators are obtained. The weights of each sub-indicator in the basic physiological indicators are allocated according to their importance, and the weights of each sub-indicator in the health and wellness effect indicators are also allocated according to their importance, highlighting the comprehensive consideration of basic physiological state and health and wellness effect. The comprehensive score is calculated by summing the products of the quantitative scores of each indicator and their corresponding weights. Based on different intervals of the comprehensive score, the health and wellness effect is divided into levels: excellent, good, average, and requiring adjustment of the health and wellness plan. The final physiological signal data includes three parts: first, real-time basic physiological parameters and corresponding time-series curve data; second, the comprehensive health and wellness status score and the sub-scores of each indicator; and third, the health and wellness effect assessment conclusion and targeted suggestions. These suggestions can be given based on the assessment results; for example, if the health and wellness effect is excellent, it is recommended to maintain the current health and wellness environment; if the health and wellness plan needs adjustment, it is recommended to increase rest time. This data can be displayed in real time on the screen of a wearable device and simultaneously transmitted to the health and wellness management platform, providing data support for the dynamic adjustment of personalized health and wellness plans.

[0081] The aforementioned method for collecting physiological signals in marine health and wellness involves first collecting preliminary physiological data, including electrocardiogram, blood oxygen, and respiration, in the marine environment and then suppressing noise to obtain interference-resistant purified physiological data. This purified data is then subjected to feature extraction in both the time and frequency domains to obtain multi-dimensional feature data that comprehensively reflects the fluctuation patterns and rhythmic characteristics. This multi-dimensional feature data is then input into a pre-defined recognition model with shallow and deep structures to accurately identify physiological state patterns such as relaxation, tension, or fatigue, obtaining pattern recognition data. Subsequently, this pattern recognition data is calibrated and fused with the initial purified physiological data through complementary multi-source fusion to obtain fused physiological data that more accurately reflects human responses. Finally, based on this fused data, a health and wellness status assessment is completed, and physiological signal data characterizing real-time status and health and wellness effects is output. This achieves high-fidelity acquisition, deep feature mining, and accurate status assessment of human physiological signals in the complex marine environment, effectively improving the scientific rigor and personalization of marine health and wellness effect monitoring.

[0082] In an optional embodiment, preliminary physiological data of users are collected via wearable devices in a marine health and wellness environment, and the preliminary physiological data is subjected to noise suppression processing to obtain purified physiological data, including:

[0083] S11. Collect ECG waveform data through the ECG sensor in the wearable device to obtain the raw ECG signal.

[0084] Optionally, the ECG sensor employs a dry electrode design, achieving stable contact without the need for conductive gel. Its electrode material utilizes a biocompatible nano-silver coating, effectively reducing skin contact impedance and improving signal acquisition stability. The sensor uses a differential amplifier circuit to initially amplify the weak ECG signal from the human body surface. This circuit, constructed using an instrumentation amplifier, features high input impedance and a high common-mode rejection ratio, effectively suppressing common-mode interference. During acquisition, the sensor's acquisition electrodes are positioned against the radial artery side of the user's wrist and the projection of the apex of the heart on the chest. The elastic structure of the wristband and chest strap ensures that displacement does not occur during marine health activities, guaranteeing the continuity of the acquired signal. The acquired raw ECG signal contains various waveform components generated by cardiac electrical activity, as well as noise components such as electromyographic interference, environmental electromagnetic interference, and motion artifacts, requiring subsequent processing for purification.

[0085] S12. Collect blood oxygen waveform data through the blood oxygen sensor in the wearable device to obtain the raw blood oxygen signal.

[0086] Optionally, the blood oxygen sensor is designed based on the principle of photoelectric volume plethysmography, employing a dual-source emission module of red and infrared light, coupled with a high-sensitivity photodiode receiving module, to acquire signals related to blood oxygen saturation. The sensor is embedded in a ring-shaped acquisition end, with an elastic pad on the inner side of the ring ensuring the light source and receiving module are in close contact with the skin at the fingertip, guaranteeing effective penetration and reception of the light signal. The red and infrared wavelengths are selected to distinguish the absorption characteristics of oxyhemoglobin from deoxyhemoglobin. When the light signal penetrates the finger tissue, changes in the hemoglobin content in the blood cause fluctuations in light absorption. The receiving module converts these light signal fluctuations into electrical signals, thus obtaining the raw blood oxygen signal. This signal not only contains pulse wave components related to blood oxygen saturation but is also affected by finger movement, skin temperature changes, and ambient light interference, requiring targeted noise reduction processing.

[0087] S13. Collect respiratory waveform data through the respiratory sensor in the wearable device to obtain the raw respiratory signal.

[0088] Optionally, the breathing sensor employs a piezoelectric thin-film sensor, which features fast response and high sensitivity, converting chest deformation caused by respiratory movements into electrical signals. The sensor is integrated into the center of an elastic chest band, fitting snugly against the user's chest. When the user breathes, the expansion and contraction of the chest creates periodic pressure changes on the piezoelectric thin film. The piezoelectric effect of the film converts these pressure changes into a corresponding voltage signal, i.e., the raw breathing signal. To prevent the chest band from being too tight and affecting breathing, or too loose and causing signal loss, the chest band uses an adjustable-length Velcro fastening structure and incorporates a pressure-sensing feedback unit to ensure appropriate contact pressure between the sensor and the chest. The raw breathing signal may contain interference signals from heartbeat vibrations and noise from body movement, requiring subsequent processing to separate the pure breathing waveform components.

[0089] S14. Perform wavelet denoising on the original electrocardiogram signal, original blood oxygen signal and original respiratory signal to obtain the denoised physiological signal.

[0090] Optionally, wavelet denoising, based on the multi-resolution analysis characteristics of wavelet transform, can distinguish signals from noise at different scales, achieving accurate denoising. Before processing, corresponding wavelet basis functions are selected according to the characteristics of the three signals: Daubechies (db) wavelets are used for ECG and blood oxygenation signals, while approximately symmetric compactly supported orthogonal (sym) wavelets are used for respiratory signals. The selection of wavelet basis functions is based on the similarity between the signal and the wavelet, ensuring effective extraction of signal features. Subsequently, the original signal undergoes multi-level wavelet decomposition. The number of decomposition levels is determined based on the signal's sampling frequency and noise distribution characteristics. The decomposition divides the signal into approximation coefficients and detail coefficients. Approximation coefficients correspond to the low-frequency useful components of the signal, while detail coefficients correspond to high-frequency noise components. A soft-thresholding method is used to suppress the detail coefficients obtained from the decomposition. During processing, detail coefficients with absolute values ​​less than the threshold are set to zero, and detail coefficients with absolute values ​​greater than the threshold are subtracted from the threshold to retain the high-frequency useful components in the signal. Finally, the processed approximation coefficients and detail coefficients are reconstructed by inverse wavelet transform to obtain the denoised physiological signal. The high-frequency noise components in the signal are significantly suppressed, while the key feature information of the signal is preserved.

[0091] S15. Baseline drift correction is performed on the denoised physiological signals to obtain purified physiological data.

[0092] Optionally, baseline drift is a common low-frequency interference in physiological signals, mainly caused by factors such as human respiratory movements, changes in body position, and sensor drift. It leads to an overall shift in the signal waveform, affecting the accuracy of subsequent feature extraction; therefore, baseline drift correction is necessary. The correction process combines high-pass filtering and polynomial fitting. First, a high-pass filter is used to initially filter the denoised physiological signal, removing some low-frequency drift components. This filter is an infinite impulse response Butterworth filter, with its cutoff frequency set to the lowest frequency that retains the useful components of the signal, ensuring no distortion is caused to the signal itself. Then, a polynomial fitting method is used to further extract and correct the baseline of the filtered signal. The baseline trend of the signal is fitted using the least squares method to obtain the baseline curve. Specifically, several feature points in the signal are selected, corresponding to the troughs of the signal waveform, ensuring they do not contain useful waveform components. A polynomial model is constructed based on these feature points, and the model parameters are solved to obtain the baseline fitting curve. Finally, the baseline fitting curve is subtracted from the denoised physiological signal to eliminate the influence of baseline drift, resulting in purified physiological data with a stable baseline and sufficient noise suppression.

[0093] In an optional embodiment, based on the purification physiological data, time-frequency domain feature extraction processing is performed on the purification physiological data to obtain multi-dimensional feature data, including:

[0094] S21. Extract time-domain features from the purification physiological data to obtain time-domain feature data.

[0095] Optionally, temporal feature extraction revolves around the core characteristics of the signal in the time dimension, designing extraction indicators based on the physiological attribute differences of the three types of signals: ECG, blood oxygenation, and respiration. For ECG signals, the focus is on extracting basic morphological features such as RR interval, R wave peak value, and QRS complex width, as well as heart rate variability features such as the standard deviation of the RR interval and the root mean square of the difference between adjacent RR intervals. For blood oxygenation signals, the focus is on indicators related to oxygenation status and vascular function, such as the mean blood oxygen saturation, pulse wave conduction time, and the pulse wave peak-to-trough difference. For respiration signals, the focus is on extracting features directly reflecting respiratory function, such as respiratory rate, respiratory depth, and the inspiratory-to-expiratory ratio. All extracted temporal indicators are integrated to form temporal feature data, fully preserving the intuitive information of signal changes in the time dimension.

[0096] S22. The purification physiological data are converted from the time domain to the frequency domain by using the spectrum analysis method, and frequency domain feature data that characterizes the rhythm distribution are extracted.

[0097] Optionally, a joint analysis using short-time Fourier transform and wavelet packet transform is employed. First, the signal is divided and superimposed into fixed windows, with a Hanning window used to suppress spectral leakage. The power spectral density of different frequency bands is obtained through short-time Fourier transform, accurately capturing the energy distribution of characteristic frequency bands in ECG, blood oxygen, and respiration signals. Then, wavelet packets are used to perform multi-level decomposition of the signal, dividing it into multiple sub-bands. The energy entropy and singular entropy of each sub-band are calculated to quantify the uniformity of energy distribution and signal complexity. Integrating the power spectral density and entropy characteristics forms frequency domain feature data that reflects the deep rhythmic characteristics of the signal.

[0098] S23. The time-domain feature data and frequency-domain feature data are concatenated and normalized to form multi-dimensional feature data.

[0099] Optionally, the two types of feature data are first converted into one-dimensional vectors and then concatenated sequentially to form a joint feature vector containing both time and frequency information. Since the dimensions of the different features vary significantly, the minimum-maximum normalization method is used to eliminate the influence of dimensions. The core formula is: ,in These are the normalized eigenvalues. These are the original eigenvalues. and These are the minimum and maximum values ​​of the sample for this feature index, respectively. After normalization, redundant features with variances approaching zero are removed, resulting in multi-dimensional feature data that combines time-domain intuitiveness with frequency-domain depth, providing high-quality input for subsequent physiological state recognition.

[0100] In an optional embodiment, time-domain features are extracted from the purification physiological data to obtain time-domain feature data, including:

[0101] S31. Calculate the mean of the purification physiological data to obtain the time-domain mean characteristics.

[0102] Optionally, mean calculation is a fundamental statistical method for reflecting the overall amplitude level of a signal, effectively suppressing the interference of random noise on the overall trend of the signal. For three types of purification physiological signals—ECG, blood oxygenation, and respiration—the arithmetic mean method is used to calculate the mean, with the sampling point sequence of the signal as the calculation object. Before calculation, outlier preprocessing is required on the signal sequence. Extreme outliers in the sampling points are identified and removed using the 3σ criterion to avoid deviations in the mean result. Specifically, the calculation process involves summing the preprocessed signal sampling point sequence and then dividing by the total number of sampling points. The result is the time-domain mean characteristic of the corresponding signal, which directly reflects the baseline level of the ECG signal, the basic fluctuation amplitude of the blood oxygenation signal, and the static reference value of the respiration signal.

[0103] S32. Calculate the variance of the purification physiological data to obtain the time-domain variance characteristics.

[0104] Optionally, variance is a statistic that measures the degree to which signal sampling points deviate from the mean. It can quantify the dispersion of the signal and indirectly reflect the stability of physiological signals. Based on the time-domain mean characteristics, for each type of purified physiological signal, the difference between each sampling point and the mean is calculated and squared. The sum of all squared differences is then divided by the total number of sampling points to obtain the time-domain variance characteristics. For electrocardiogram (ECG) signals, the variance characteristics reflect the baseline level of heart rate fluctuations; for blood oxygenation signals, its magnitude is related to the stability of blood oxygen saturation; for respiratory signals, it corresponds to the baseline fluctuation of respiratory rhythm. The smaller the variance, the more stable the signal and the more stable the physiological state.

[0105] S33. Perform peak detection on the purification physiological data to obtain the time-domain peak characteristics; wherein, the expression for peak detection is:

[0106]

[0107] In the formula, Indicates the peak characteristics in the time domain. Indicates the number of signal sampling points. Indicates the first The signal value at each sampling point.

[0108] In the formula, The peak feature in the time domain, which is the final extracted feature, is used to characterize the maximum fluctuation amplitude of the signal and is a key indicator reflecting the intensity of physiological activity. This represents the number of sampling points for purified physiological signals within the current analysis window, and its value is determined by the signal sampling frequency and the analysis window duration. Represents the first in the signal sequence The specific signal values ​​at each sampling point, by taking the absolute value and then solving for the maximum value, can simultaneously cover both the positive and negative peak values ​​of the signal, ensuring that the peak characteristics can fully reflect the maximum fluctuation range of the signal. In actual detection, the signal needs to be smoothed first to eliminate high-frequency glitches, and then the signal sequence is traversed through a sliding window. The peak values ​​are calculated using the above expression to obtain the maximum amplitude of the ECG R wave, the peak intensity of the blood oxygen pulse wave, and the maximum motion amplitude of the respiratory waveform.

[0109] S34. Perform feature fusion on the time-domain mean feature, time-domain variance feature, and time-domain peak feature to obtain the time-domain feature vector.

[0110] Optionally, before feature fusion, the three types of features need to be consistent. The mean, variance, and peak values ​​of the ECG, blood oxygenation, and respiration signals are labeled with classification tags to ensure a one-to-one correspondence between features and signal types. Then, in the order of ECG features, blood oxygenation features, and respiration features, the mean, variance, and peak values ​​of each signal are arranged sequentially to form a one-dimensional feature sequence. To avoid the influence of differences in numerical ranges between different features on subsequent processing, this sequence is standardized to map all feature values ​​to a unified interval, ultimately forming a time-domain feature vector with a clear structure and well-defined dimensions, fully preserving the basic statistical characteristics and intensity information of the three types of signals.

[0111] In an optional embodiment, multi-dimensional feature data is input into a preset multi-level recognition model, and physiological state pattern recognition processing is performed on the multi-dimensional feature data to obtain pattern recognition data, including:

[0112] S41. Input the multi-dimensional feature data into the shallow feature extraction layer for shallow feature extraction processing to obtain shallow feature data.

[0113] Optionally, the core function of the shallow feature extraction layer is to quickly filter redundant information in the feature data and extract basic local correlation features, reducing the computational load for deep processing. This layer consists of multiple progressively structured convolutional units, each containing convolution operations, activation functions, and pooling operations. After multi-dimensional feature data is input, feature mapping is first performed through the convolutional kernel of the first convolutional unit. The kernel size is set according to the dimension of the feature data, effectively capturing the local dependencies between adjacent features. Subsequently, a non-linear transformation is introduced through the ReLU activation function, enhancing the model's ability to express complex features. Finally, max pooling is used to compress the feature dimension, preserving key features while improving the model's robustness. Subsequent convolutional units gradually deepen feature extraction by increasing the number of convolutional kernels, ultimately outputting shallow feature data with reduced dimensions and concentrated information, providing high-quality input for deep processing.

[0114] S42. Input the shallow feature data into the convolutional neural network of the deep pattern recognition layer for convolutional feature extraction processing to obtain convolutional feature data.

[0115] Optionally, the convolutional neural network module of the deep pattern recognition layer focuses on mining the complex correlation information hidden in shallow features, and uses a multi-scale convolutional structure to improve the comprehensiveness of feature extraction. This module sets convolutional kernels of different sizes to process shallow feature data in parallel. Large-sized convolutional kernels are used to capture long-distance dependencies between features, while small-sized convolutional kernels focus on local fine features. The outputs of convolutional kernels of each scale are fused through feature concatenation to form multi-dimensional intermediate features. To avoid overfitting during training, batch normalization is introduced after the convolution operation. By standardizing and adjusting the feature data, the model convergence is accelerated and the generalization ability is improved. After multiple rounds of convolution and normalization operations, the output is convolutional feature data that reflects the complex correlation characteristics of features.

[0116] S43. Input the convolutional feature data into the recurrent neural network of the deep pattern recognition layer for temporal feature extraction processing to obtain temporal feature data.

[0117] Optionally, the core advantage of the recurrent neural network module lies in capturing the temporal dynamics of feature data, which highly matches the characteristic of continuous changes in physiological signals over time. Before inputting convolutional feature data, dimensionality adjustment is required to convert it into a sequence format that meets the input requirements of the recurrent neural network. The sequence length is determined by the temporal dimension of the feature data. This step employs a bidirectional recurrent structure: the forward recurrent unit traverses from the beginning of the sequence backward, capturing the positive temporal dependencies of the features; the backward recurrent unit traverses from the end of the sequence forward, capturing the negative temporal dependencies. The outputs of both are concatenated and fused to fully preserve the temporal correlation information of the features. To address the vanishing gradient problem in long sequence training, a gating mechanism is introduced within the network. Through the synergistic effect of the forget gate, input gate, and output gate, temporal information is adaptively retained or discarded, ultimately outputting temporal feature data that accurately reflects the dynamic changes of physiological characteristics.

[0118] S44. Input the temporal feature data into the fully connected layer of the deep pattern recognition layer for pattern classification processing to obtain pattern recognition data.

[0119] Optionally, the fully connected layer serves as the classification terminal of the model, responsible for mapping temporal feature data to specific physiological state categories. The temporal feature data is first converted into a one-dimensional vector through a flattening operation and then input into multi-level fully connected units. The preceding fully connected units construct complex feature mapping relationships through a large number of neurons, converting the temporal features into a higher-dimensional feature representation more suitable for classification. A dropout layer is set after each unit for regularization, randomly discarding some neurons to avoid model overfitting. The number of neurons in the final fully connected unit is consistent with the preset number of physiological state categories. The softmax activation function converts the feature vector into probability values ​​corresponding to each category; the category with the highest probability value is the recognition result of the current physiological state. Simultaneously, the confidence score of this category is output. When the confidence score is lower than a set threshold, it is marked as a state to be confirmed, ultimately forming pattern recognition data containing physiological state categories and confidence scores, providing a reliable basis for state judgment for subsequent data fusion.

[0120] In an optional embodiment, the fused physiological data is processed for health and wellness status assessment to obtain physiological signal data, including:

[0121] S51. Extract and process health and wellness indicators from the fused physiological data to obtain health and wellness indicator data.

[0122] Optionally, the integrated physiological data combines physiological state assessment with core physiological parameters. Indicator extraction needs to construct a two-tiered system around basic health assurance and rehabilitation effect feedback to ensure coverage of both static and dynamic states. Basic health assurance indicators are directly selected and statistically analyzed from the integrated data, including mean heart rate, mean blood oxygen saturation, mean respiratory rate, and SDNN values ​​of heart rate variability. These indicators are core representations of basic human health status. Extraction is performed using a rolling statistical method within a fixed time window to balance data timeliness and stability. Rehabilitation effect feedback indicators are constructed by combining state identification results with parameter change characteristics, specifically including a state stability index, indicator improvement rate, and stress relief degree. The state stability index is obtained by statistically analyzing the number of changes in physiological state categories within a continuous fixed time period; a lower frequency of change indicates a more stable rehabilitation state. The indicator improvement rate is the percentage of the difference between the current indicator and the initial baseline value relative to the initial value; a higher positive percentage indicates a more significant rehabilitation intervention effect. The stress relief degree is calculated as 1 minus the stress state confidence level; the value directly reflects the relief of the stress response. The two types of indicators are integrated in the order of basic indicators first and effect indicators second, and the data collection timestamps and calculation basis of each indicator are marked to form a complete and logically clear health and wellness indicator data.

[0123] S52. Calculate the scores for the health and wellness indicator data to obtain the health and wellness score data.

[0124] Optionally, a ten-point quantitative scoring system can be adopted, combining the physiological significance of each indicator with differentiated standards based on health and wellness goals to ensure that the scoring results accurately reflect the quality of the indicators. Basic health assurance indicators are based on the medically normal range. Full marks are awarded when the mean heart rate, mean blood oxygen saturation, mean respiratory rate, and SDNN value are within the normal range. Deviations from the normal range result in linear deductions based on the degree of deviation, with the deduction magnitude positively correlated with health risk. In the health and wellness effect feedback indicators, a state stability index of zero receives full marks, with each additional state change deducting corresponding points. Full marks are awarded when the indicator improvement rate is greater than a set positive threshold, a moderate score is awarded when it is between zero and the threshold, and zero marks are awarded when it is less than zero. Stress relief is converted to a ten-point score. Before scoring, the indicator data must be validated. Outliers caused by temporary equipment malfunctions are eliminated using the 3σ criterion, and outliers are replaced by the mean of adjacent indicators within the same time period. Subsequently, the weights of each indicator were determined using the analytic hierarchy process. The total weight of the basic health security indicators was higher than that of the health care effect feedback indicators, highlighting the core position of basic health. Finally, the scores of each indicator were multiplied by their corresponding weights and summed to obtain the comprehensive health care score and the scores of each sub-item, thus forming the health care score data.

[0125] S53. Perform trend analysis on the health and wellness scoring data to obtain health and wellness trend data.

[0126] Optionally, the health and wellness scoring data is first divided into continuous analysis units according to time sequence. Each unit contains a fixed number of scoring samples. A sliding window is used to traverse all data, calculating the mean and standard deviation of the scores within each window. The mean reflects the overall scoring level during that period, while the standard deviation reflects the fluctuation range of the scores. A linear fit is performed on the mean of each window to obtain a trend line, and the slope of the line is used to determine the trend direction. A positive slope indicates an upward trend, representing continuous improvement in health and wellness effects; a negative slope indicates a downward trend, suggesting the need to monitor changes in health and wellness status; a slope close to zero indicates a stable trend, indicating stable health and wellness effects. Simultaneously, the trend confidence score is calculated, expressed as the absolute value of the correlation coefficient of the fitted line. The larger the correlation coefficient, the higher the trend confidence. Combined with the standard deviation data of each window, outliers in the trend curve are marked. These outliers correspond to windows with a standard deviation greater than a set threshold, ultimately forming health and wellness trend data that includes trend direction, slope, confidence score, and outlier markings, fully presenting the dynamic changes in the scores.

[0127] S54. Generate a health and wellness report based on health and wellness score data and health and wellness trend data, and obtain physiological signal data.

[0128] Optionally, the data presentation section visually displays the overall score, sub-scores, and raw values ​​of each indicator in tabular form, simultaneously embedding a trend curve of score changes, marking the trend direction and key fluctuation points, making data changes readily apparent. The conclusion analysis section combines both score and trend information, dividing the overall score range into four levels of health and wellness effectiveness: excellent, good, average, and requiring adjustment. Supplementary conclusions are also provided based on trend characteristics. For example, an excellent overall score with an upward trend indicates a continuous improvement in health and wellness effectiveness; an average overall score with a downward trend suggests the need for timely intervention. The recommendation output section proposes personalized solutions based on the conclusions, providing targeted conditioning suggestions when basic indicators are abnormal; for example, increasing rest time may be recommended when blood oxygen levels are low. A downward trend clarifies the direction for adjusting the health and wellness plan, such as optimizing the duration of marine environmental exposure or adjusting activity intensity. The report content is integrated with real-time basic physiological parameters, timestamps, and other information to form the final physiological signal data, which can be displayed locally via wearable devices or transmitted in real-time to the health and wellness management platform, providing data support for the dynamic adjustment of personalized health and wellness plans.

[0129] In an optional embodiment, the health and wellness indicator data are scored and calculated to obtain health and wellness score data, including:

[0130] S61. Calculate the normalized values ​​of each health and wellness indicator based on the health and wellness indicator data to obtain normalized indicator data.

[0131] Optionally, the health and wellness indicator data includes two categories: basic health and effect feedback indicators. Different indicators have different dimensions and numerical ranges, and direct calculation can lead to biased results. Therefore, normalization is necessary. Normalization uses the physiological normal range or historical data interval of each indicator as a benchmark. For indicators with clearly defined medical normal ranges, such as mean heart rate and mean blood oxygen saturation, the upper and lower limits of the normal range are used as benchmark values. For indicators without clear standards, such as the state stability index and indicator improvement rate, the historical statistical interval of similar health and wellness populations is used as the benchmark value. The original values ​​of each indicator are converted into normalized values ​​within a fixed interval using a normalization formula, eliminating the influence of dimensions, making different indicators comparable, and ultimately forming normalized indicator data.

[0132] S62. Based on the preset weighting coefficients, the normalized index data are weighted and summed to obtain a comprehensive health and wellness score.

[0133] The expression for the comprehensive health and wellness score is as follows:

[0134]

[0135] In the formula, Indicates the overall health and wellness score. Indicates the first The weighting coefficients of each health and wellness indicator. Indicates the first A normalized health and wellness indicator This represents the total number of health and wellness indicators.

[0136] In the formula, The comprehensive health and wellness score is a core quantitative indicator that reflects the user's overall health and wellness status. The higher the score, the better the health and wellness status. Representing the The weighting coefficients of each health and wellness indicator were determined by combining the analytic hierarchy process with expert experience. The total weighting coefficients of the basic health security indicators were higher than those of the health and wellness effect feedback indicators, in order to highlight the core position of basic health. Representing the A normalized health and wellness indicator, that is, the indicator value after normalization processing; The score represents the total number of health and wellness indicators, covering all basic health and effectiveness feedback indicators to ensure a comprehensive reflection of health and wellness status. The calculation involves multiplying the weight coefficient of each indicator by its corresponding normalized value according to the formula, and then summing all the products to obtain the final comprehensive health and wellness score.

[0137] S63. Classify the comprehensive health and wellness score into levels to obtain health and wellness level data.

[0138] Optionally, the grading is based on the distribution characteristics of the comprehensive health and wellness score, combined with setting a grading threshold according to the target effect of health and wellness intervention. Through statistical analysis of a large amount of health and wellness sample data, the health and wellness status corresponding to different scoring intervals is determined, forming standardized grading rules. The graded levels include qualitative descriptions such as excellent, good, average, and needing adjustment, and also correspond to clear scoring interval ranges, making the grading results both intuitive and objective. Matching the calculated comprehensive health and wellness score to the corresponding grade interval yields health and wellness grade data containing both scoring and qualitative grades.

[0139] S64. Generate a scoring report based on the comprehensive health and wellness score and health and wellness level data to obtain health and wellness score data.

[0140] Optionally, the scoring report aims to clearly present the scoring results and core criteria, and includes three parts: first, the original values, normalized values, and corresponding weighting coefficients of each health and wellness indicator, clarifying the basic data for the scoring calculation; second, a detailed description of the comprehensive health and wellness score and the calculation process, enhancing the credibility of the results; and third, a qualitative description of the health and wellness level and the basis for level classification, allowing users to intuitively understand their own health and wellness status. Integrating the report content with the data collection timestamps of each indicator forms a complete and structured health and wellness scoring data set, providing accurate quantitative evidence for subsequent trend analysis.

[0141] In the aforementioned method for collecting physiological signals in marine health and wellness, preliminary physiological data, including electrocardiogram, blood oxygen saturation, and respiratory waveforms, are first collected using wearable devices in the marine health and wellness environment. This data is then subjected to noise suppression processing to filter out environmental noise such as sea breezes and tidal sounds, resulting in purified physiological data. Subsequently, time-frequency domain feature extraction processing is performed on this purified physiological data to comprehensively capture the fluctuation patterns and rhythmic characteristics of physiological signals in both the time and frequency domains, obtaining multi-dimensional feature data. Finally, this multi-dimensional feature data is input into a pre-defined multi-layered recognition model with a shallow feature extraction layer and a deep pattern recognition layer structure for physiological state pattern recognition. The process involves differentiating different states, such as relaxation, tension, or fatigue, to obtain pattern recognition data. Then, the pattern recognition data and the purified physiological data undergo multi-source data fusion processing. The state information identified by the model is calibrated and complemented with the pure original waveform to obtain fused physiological data. Finally, the fused physiological data undergoes health and wellness status assessment processing to generate physiological signal data that accurately characterizes the user's real-time physiological state and health and wellness effects in a marine environment. This achieves the technical effects of effectively purifying physiological signals, deeply extracting multi-dimensional features, accurately identifying physiological states, and ultimately realizing a scientific and personalized health and wellness effect assessment in complex marine environments.

[0142] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0143] Based on the same inventive concept, this application also provides a physiological signal acquisition device for marine health care, used to implement the above-mentioned method for acquiring physiological signals in marine health care. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the physiological signal acquisition device for marine health care provided below can be found in the limitations of the physiological signal acquisition method for marine health care described above, and will not be repeated here.

[0144] In one exemplary embodiment, such as Figure 2 As shown, a schematic diagram of a physiological signal acquisition device 20 for marine health and wellness is provided, comprising:

[0145] The preliminary data acquisition and noise reduction module 21 is used to collect preliminary physiological data of users through wearable devices in a marine health and wellness environment, and to perform noise suppression processing on the preliminary physiological data to obtain purified physiological data; among which, the preliminary physiological data includes electrocardiogram waveform data, blood oxygen waveform data and respiratory waveform data;

[0146] The data feature processing module 22 is used to perform time-frequency domain feature extraction processing on the purified physiological data to obtain multi-dimensional feature data; wherein, the multi-dimensional feature data is used to characterize the fluctuation pattern and rhythm characteristics of physiological signals in the time domain and frequency domain.

[0147] The data pattern recognition module 23 is used to input multi-dimensional feature data into a preset multi-level recognition model, perform physiological state pattern recognition processing on the multi-dimensional feature data, and obtain pattern recognition data; wherein, the structure of the preset multi-level recognition model includes a shallow feature extraction layer and a deep pattern recognition layer connected in sequence.

[0148] The multi-source data fusion module 24 is used to perform multi-source data fusion processing on pattern recognition data and purification physiological data to obtain fused physiological data.

[0149] The physiological signal data module 25 is used to process the fused physiological data to assess the health status and obtain physiological signal data; wherein, the physiological signal data is used to characterize the user's real-time physiological status and health effects in the marine health environment.

[0150] Furthermore, the preliminary data acquisition and noise reduction module 21 is also used for:

[0151] S11. Collect ECG waveform data through the ECG sensor in the wearable device to obtain the raw ECG signal;

[0152] S12. Collect blood oxygen waveform data through the blood oxygen sensor in the wearable device to obtain the raw blood oxygen signal;

[0153] S13. Collect respiratory waveform data through the respiratory sensor in the wearable device to obtain the raw respiratory signal;

[0154] S14. Perform wavelet denoising on the raw electrocardiogram signal, raw blood oxygen signal and raw respiratory signal to obtain the denoised physiological signal;

[0155] S15. Baseline drift correction is performed on the denoised physiological signals to obtain purified physiological data.

[0156] Furthermore, the data feature processing module 22 is also used for:

[0157] S21. Extract time-domain features from the purification physiological data to obtain time-domain feature data;

[0158] S22. The purification physiological data are transformed from the time domain to the frequency domain through the spectrum analysis method, and frequency domain feature data representing the rhythm distribution are extracted.

[0159] S23. The time-domain feature data and frequency-domain feature data are concatenated and normalized to form multi-dimensional feature data.

[0160] Furthermore, the data feature processing module 22 is also used for:

[0161] S31. Calculate the mean of the purification physiological data to obtain the time-domain mean characteristics;

[0162] S32. Calculate the variance of the purification physiological data to obtain the time-domain variance characteristics;

[0163] S33. Perform peak detection on the purification physiological data to obtain the time-domain peak characteristics; wherein, the expression for peak detection is:

[0164]

[0165] In the formula, Indicates the peak characteristics in the time domain. Indicates the number of signal sampling points. Indicates the first The signal values ​​at each sampling point;

[0166] S34. Perform feature fusion on the time-domain mean feature, time-domain variance feature, and time-domain peak feature to obtain the time-domain feature vector.

[0167] Furthermore, the data pattern recognition module 23 is also used for:

[0168] S41. Input the multi-dimensional feature data into the shallow feature extraction layer for shallow feature extraction processing to obtain shallow feature data;

[0169] S42. Input the shallow feature data into the convolutional neural network of the deep pattern recognition layer for convolutional feature extraction processing to obtain convolutional feature data;

[0170] S43. Input the convolutional feature data into the recurrent neural network of the deep pattern recognition layer for temporal feature extraction processing to obtain temporal feature data;

[0171] S44. Input the temporal feature data into the fully connected layer of the deep pattern recognition layer for pattern classification processing to obtain pattern recognition data.

[0172] Furthermore, the physiological signal data module 25 is also used for:

[0173] S51. Extract and process health and wellness indicators from the fused physiological data to obtain health and wellness indicator data;

[0174] S52. Calculate and score the health and wellness indicator data to obtain health and wellness score data;

[0175] S53. Perform trend analysis on the health and wellness scoring data to obtain health and wellness trend data;

[0176] S54. Generate a health and wellness report based on health and wellness score data and health and wellness trend data, and obtain physiological signal data.

[0177] Furthermore, the physiological signal data module 25 is also used for:

[0178] S61. Calculate the normalized values ​​of each health and wellness indicator based on the health and wellness indicator data to obtain normalized indicator data;

[0179] S62. Based on the preset weighting coefficients, the normalized index data are weighted and summed to obtain a comprehensive health and wellness score.

[0180] The expression for the comprehensive health and wellness score is as follows:

[0181]

[0182] In the formula, Indicates the overall health and wellness score. Indicates the first The weighting coefficients of each health and wellness indicator. Indicates the first A normalized health and wellness indicator This represents the total number of health and wellness indicators;

[0183] S63. Classify the comprehensive health and wellness score into levels to obtain health and wellness level data;

[0184] S64. Generate a scoring report based on the comprehensive health and wellness score and health and wellness level data to obtain health and wellness score data.

[0185] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the physiological signal acquisition method in marine health care as described above.

[0186] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0187] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0188] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for collecting physiological signals in marine health and wellness, characterized in that, The method includes: S1. In a marine health and wellness environment, wearable devices are used to collect preliminary physiological data from users, and noise suppression processing is applied to the preliminary physiological data to obtain purified physiological data; wherein, the preliminary physiological data includes electrocardiogram waveform data, blood oxygen waveform data, and respiratory waveform data; S2. Based on the purification physiological data, perform time-frequency domain feature extraction processing on the purification physiological data to obtain multi-dimensional feature data; wherein, the multi-dimensional feature data is used to characterize the fluctuation pattern and rhythm characteristics of physiological signals in the time domain and frequency domain. S3. Input the multi-dimensional feature data into a preset multi-level recognition model, and perform physiological state pattern recognition processing on the multi-dimensional feature data to obtain pattern recognition data; wherein, the structure of the preset multi-level recognition model includes a shallow feature extraction layer and a deep pattern recognition layer connected in sequence. S4. Perform multi-source data fusion processing on the pattern recognition data and the purification physiological data to obtain fused physiological data; S5. Perform health and wellness status assessment processing on the fused physiological data to obtain physiological signal data; wherein, the physiological signal data is used to characterize the user's real-time physiological status and health and wellness effects in the marine health and wellness environment.

2. The method according to claim 1, characterized in that, The process involves collecting preliminary physiological data from users via wearable devices in a marine health and wellness environment, and then performing noise suppression processing on the preliminary physiological data to obtain purified physiological data, including: S11. Collect ECG waveform data through the ECG sensor in the wearable device to obtain the raw ECG signal; S12. Collect blood oxygen waveform data through the blood oxygen sensor in the wearable device to obtain the raw blood oxygen signal; S13. Collect respiratory waveform data through the respiratory sensor in the wearable device to obtain the raw respiratory signal; S14. Perform wavelet denoising on the original electrocardiogram signal, the original blood oxygen signal and the original respiratory signal to obtain the denoised physiological signal; S15. The denoised physiological signal is subjected to baseline drift correction processing to obtain the purified physiological data.

3. The method according to claim 1, characterized in that, The step involves performing time-frequency domain feature extraction processing on the purification physiological data to obtain multi-dimensional feature data, including: S21. Extract time-domain features from the purification physiological data to obtain time-domain feature data; S22. The purification physiological data are converted from the time domain to the frequency domain using a spectrum analysis method, and frequency domain feature data characterizing the rhythm distribution are extracted. S23. The time-domain feature data and the frequency-domain feature data are concatenated and normalized to form the multi-dimensional feature data.

4. The method according to claim 3, characterized in that, The extraction of time-domain features from the purified physiological data to obtain time-domain feature data includes: S31. Calculate the mean value of the purification physiological data to obtain the time-domain mean value characteristics; S32. Calculate the variance of the purification physiological data to obtain the time-domain variance characteristics; S33. Perform peak detection on the purified physiological data to obtain time-domain peak characteristics; wherein, the expression for peak detection is: In the formula, Indicates the peak characteristics in the time domain. Indicates the number of signal sampling points. Indicates the first The signal values ​​at each sampling point; S34. Perform feature fusion on the time-domain mean feature, the time-domain variance feature, and the time-domain peak feature to obtain the time-domain feature vector.

5. The method according to claim 1, characterized in that, The step of inputting the multi-dimensional feature data into a preset multi-level recognition model and performing physiological state pattern recognition processing on the multi-dimensional feature data to obtain pattern recognition data includes: S41. Input the multi-dimensional feature data into the shallow feature extraction layer for shallow feature extraction processing to obtain shallow feature data; S42. Input the shallow feature data into the convolutional neural network of the deep pattern recognition layer for convolutional feature extraction processing to obtain convolutional feature data; S43. Input the convolutional feature data into the recurrent neural network of the deep pattern recognition layer for temporal feature extraction processing to obtain temporal feature data; S44. Input the time-series feature data into the fully connected layer of the deep pattern recognition layer for pattern classification processing to obtain pattern recognition data.

6. The method according to claim 1, characterized in that, The process of assessing the health status of the fused physiological data to obtain physiological signal data includes: S51. Extract health and wellness indicators from the fused physiological data to obtain health and wellness indicator data. S52. The health and wellness indicator data are scored and calculated to obtain health and wellness score data; S53. Perform trend analysis on the health and wellness scoring data to obtain health and wellness trend data; S54. Generate a health and wellness report based on the health and wellness score data and health and wellness trend data, and obtain the physiological signal data.

7. The method according to claim 6, characterized in that, The process of scoring and calculating the health and wellness indicator data to obtain health and wellness score data includes: S61. Calculate the normalized value of each health and wellness indicator based on the health and wellness indicator data to obtain normalized indicator data; S62. The normalized index data is weighted and summed based on preset weight coefficients to obtain a comprehensive health and wellness score; The expression for the comprehensive health and wellness score is as follows: In the formula, Indicates the overall health and wellness score. Indicates the first The weighting coefficients of each health and wellness indicator. Indicates the first A normalized health and wellness indicator This represents the total number of health and wellness indicators; S63. Divide the comprehensive health and wellness score into levels to obtain health and wellness level data; S64. Generate a scoring report based on the comprehensive health and wellness score and the health and wellness level data to obtain the health and wellness score data.

8. A physiological signal acquisition device for marine health and wellness, characterized in that, The device includes: The preliminary data acquisition and noise reduction module is used to collect preliminary physiological data of users through wearable devices in a marine health and wellness environment, and to perform noise suppression processing on the preliminary physiological data to obtain purified physiological data; wherein, the preliminary physiological data includes electrocardiogram waveform data, blood oxygen waveform data and respiratory waveform data; The data feature processing module is used to perform time-frequency domain feature extraction processing on the purified physiological data to obtain multi-dimensional feature data; wherein, the multi-dimensional feature data is used to characterize the fluctuation pattern and rhythm characteristics of physiological signals in the time domain and frequency domain. The data pattern recognition module is used to input the multi-dimensional feature data into a preset multi-level recognition model, perform physiological state pattern recognition processing on the multi-dimensional feature data, and obtain pattern recognition data; wherein, the preset multi-level recognition model includes a shallow feature extraction layer and a deep pattern recognition layer connected in sequence. A multi-source data fusion module is used to perform multi-source data fusion processing on the pattern recognition data and the purification physiological data to obtain fused physiological data. The physiological signal data module is used to perform health and wellness status assessment processing on the fused physiological data to obtain physiological signal data; wherein, the physiological signal data is used to characterize the user's real-time physiological status and health and wellness effects in the marine health and wellness environment.

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

Images