Wearable underwater vital sign and environment monitoring system

Abstract

The invention relates to the technical field of vital sign measurement, in particular to a wearable underwater vital sign and environment monitoring system, which comprises a physiological signal acquisition module, a photoelectric volume pulse wave sensor, a bioelectrical impedance electrode, a piezoelectric sensor, a signal processing module, a signal processing module and a signal processing module, a wrist photoelectric volume pulse wave signal, a chest bioelectrical impedance signal and a neck piezoelectric signal are collected. According to the method, signal reconstruction is carried out based on energy proportion calculation, an electrocardiogram R wave peak value point position and an R-R interval time sequence are extracted, variability indexes are generated through frequency domain decomposition and frequency band energy statistics, time domain and frequency domain index values are calculated, weight distribution is carried out, and therefore comprehensive monitoring on the underwater physical sign state is achieved. The signal acquisition precision is improved by means of a multi-source signal acquisition mode, noise interference is eliminated through signal track matrix decomposition operation, and the signal reconstruction quality is ensured through feature vector energy proportion calculation.

Description

Technical Field

[0001] The present invention relates to the technical field of vital sign measurement, and particularly to a wearable underwater vital sign and environmental monitoring system. Background Art

[0002] The technical field of vital sign measurement is an important branch of medical health monitoring, mainly focusing on the acquisition, analysis, and monitoring of human vital signs. This field includes the measurement and evaluation of key physiological indicators such as human body temperature, heart rate, blood pressure, respiratory rate, blood oxygen saturation, etc., and a wearable underwater vital sign monitoring system is a portable device specifically used for monitoring human vital signs in an underwater environment.

[0003] However, there are many limitations in the actual application of the existing technology. Relying solely on a single sensor to collect physiological signals is difficult to ensure data integrity and accuracy. Affected by the underwater environment, signal distortion and information loss are likely to occur, resulting in a large deviation in the monitoring results. And for signal feature extraction, a simple threshold judgment method is adopted, which is difficult to handle the situation of signal distortion and severe fluctuations, resulting in inaccurate extraction of characteristic parameters. In the process of vital sign state evaluation, only the change of a single index is concerned, and the joint analysis of multi-dimensional physiological parameters is ignored, making the evaluation result one-sided and lacking reliability. Therefore, improvements are needed. Summary of the Invention

[0004] The object of the present invention is to solve the drawbacks existing in the prior art, and a wearable underwater vital sign and environment monitoring system is proposed. To achieve the above object, the present invention adopts the following technical solutions: A wearable underwater vital sign and environment monitoring system includes: a physiological signal acquisition module, which fixes a photoplethysmogram sensor at the wrist position, attaches a bioelectrical impedance electrode to the chest, and places a piezoelectric sensor at the neck position to collect the photoplethysmogram signal at the wrist, the bioelectrical impedance signal at the chest, and the piezoelectric signal at the neck, obtaining multi-source physiological signals, performing time series alignment and synchronous sampling on the multi-source physiological signals, and generating a physiological signal sequence; a signal denoising and reconstruction module, based on the physiological signal sequence, constructs a signal trajectory matrix, performs a block decomposition operation on the signal trajectory matrix, extracts signal eigenvalues and eigenvectors, obtains a set of eigencomponents, calculates the energy ratio of the eigenvectors in the set of eigencomponents, and generates a reconstructed signal group; a dynamic evaluation module, based on the reconstructed signal group, extracts the positions of electrocardiogram R-wave peak points and the electrocardiogram R-R interval time series, calculates the differences and fluctuation ranges of adjacent R-R intervals, obtains the heart rate change characteristics, performs frequency domain decomposition and frequency band energy statistics on the heart rate change characteristics, and generates heart rate variability indexes; a vital sign parameter fusion module, based on the heart rate variability indexes, calculates the numerical values of the time domain indexes and frequency domain indexes of heart rate variability respectively, statistics the change trends and fluctuation ranges of each index, obtains vital sign fluctuation parameters, performs weight assignment on the vital sign fluctuation parameters, and generates an underwater vital sign state quantity; an environment monitoring module, which is used to monitor the underwater environment temperature through a temperature sensor unit, monitor the underwater environment pressure through a pressure sensor unit, and monitor the underwater environment salinity through an inductive conductivity sensor unit, obtaining underwater environment monitoring data; a communication module, which is used to communicate the underwater vital sign state quantity and the underwater environment monitoring data with a wearable display terminal and a water surface monitoring device through an underwater cable or wireless communication method. Preferably, the steps for obtaining the physiological signal sequence are: based on the multi-source physiological signals, perform time series alignment, align each signal according to the time stamp, and adjust the signal data of each signal by interpolation or delay according to the acquisition time point of each signal, obtaining time series aligned multi-source physiological signals; according to the time series aligned multi-source physiological signals, perform data synthesis, integrate all aligned signals to a unified time point through data fusion, and generate a physiological signal sequence. Preferably, the steps for obtaining the set of eigencomponents are: according to the physiological signal sequence, generate a two-dimensional matrix by arranging the signal values sampled at each time point, take the data at each time point as a row vector and the signal channel data as a column vector, obtaining a signal trajectory matrix; according to the signal trajectory matrix, calculate the characteristic score of signal decomposition, and the calculation formula is: wherein, is the characteristic score, is the singular value of the signal trajectory matrix, is the corresponding left singular vector, is the number of blocks; according to the feature scores, the matching components of the eigenvalues and eigenvectors are extracted, and the block feature scores are aggregated to generate the corresponding signal eigenvalue and eigenvector set, obtaining the feature component set. Preferably, the step of obtaining the reconstructed signal group is: according to the feature component set, the eigenvectors and the corresponding singular values are extracted, and the eigenvectors and singular values are rearranged in matrix form to generate an eigenvector matrix; based on the eigenvector matrix, the energy ratio is calculated, and the calculation formula is: where, is the energy ratio, is the th component value of the reconstructed signal, is the th component value in the eigenvector set, is the number of selected principal components, is the total number of eigenvectors; the eigenvectors with the energy ratio higher than the threshold are weighted and combined with the corresponding singular values, and the inverse singular value decomposition is performed to generate the reconstructed signal group. Preferably, the step of obtaining the heart rate change feature is: according to the reconstructed signal group, each electrocardiogram signal channel is processed, and the peak point position of the R wave is extracted through waveform analysis and threshold determination, and the peak point positions are arranged in a time series manner to obtain the electrocardiogram R wave peak point position sequence; based on the electrocardiogram R wave peak point position sequence, the nonlinear fluctuation index is calculated, and the calculation formula is: where, is the nonlinear fluctuation index of adjacent R-R intervals, is the time position of the th R wave peak point, and are the time positions of the previous and the two previous R wave peak points respectively, is a dimensionless constant for positive adjustment to avoid the denominator being zero, is the total number of R waves; according to the nonlinear fluctuation index, combined with the time series analysis of the electrocardiogram signal, the heart rate change feature is generated by calculating the amplitude distribution characteristics of the R wave. Preferably, the step of obtaining the heart rate variability index is: according to the heart rate change feature, the heart rate change feature is transformed from the time domain to the frequency domain through Fourier transform, and different frequency components and amplitude values are extracted to obtain the frequency domain heart rate feature; based on the frequency domain heart rate feature, the band energy ratio of the target band is calculated, and the calculation formula is: where, is the band energy ratio, is the frequency at the power spectral density, and are the lower and upper limit frequencies of the target band, and are the minimum and maximum frequencies of the entire frequency spectrum; combining the information of the main frequency components in the frequency domain, the proportion of the frequency band energy in multiple frequency bands is statistically analyzed and summarized into an index set to generate heart rate variability indexes. Preferably, the obtaining step of the physical sign fluctuation parameter is as follows: classifying the heart rate variability indexes into two parts, namely time domain features and frequency domain features, statistically analyzing the time domain features including standard deviation and root mean square interval, classifying the frequency domain features including frequency band energy ratio and main frequency, and obtaining the classified heart rate variability indexes; based on the classified heart rate variability indexes, calculating the change trends and fluctuation ranges of each time domain feature and frequency domain feature, statistically analyzing the extreme values, ranges and average change amounts among the indexes, and obtaining the change trend of the statistical indexes; based on the change trend of the statistical indexes, calculating the physical sign fluctuation parameter, and the calculation formula is: Wherein, is the physical sign fluctuation parameter, is the change value of the th time domain feature, is the change value of the th frequency domain feature, is the proportion of the th frequency domain feature, is the total number of features. Preferably, the step of obtaining the underwater physical sign state quantity is as follows: According to the physical sign fluctuation parameter, it is divided into time-domain features and frequency-domain features according to the feature type, and each change value in the time-domain features and each energy value in the frequency-domain features are extracted respectively. Data normalization processing is performed according to a unified range, and the normalized data is recombined to form a normalized physical sign fluctuation parameter set; Based on the normalized physical sign fluctuation parameter set, through the correlation between features, the similarity between each feature is calculated, and the correlation relationship between features is represented in matrix form. The features are merged and recalculated to obtain a fused physical sign fluctuation parameter set; According to the fused physical sign fluctuation parameter set, weight values are assigned according to the classification criteria of time-domain features and frequency-domain features, and the features after weight assignment are summed up to generate an underwater physical sign state quantity. Compared with the prior art, the advantages and positive effects of the present invention are as follows: In the present invention, by fixing a photoplethysmogram sensor on the wrist, attaching a bioelectrical impedance electrode to the chest, and placing a piezoelectric sensor on the neck to collect multi-source physiological signals, the signals are aligned in time sequence and synchronously sampled, a signal trajectory matrix is constructed and block decomposition operation is performed, eigenvalue and eigenvector are extracted, signal reconstruction is performed based on energy ratio calculation, the position of the electrocardiogram R-wave peak point and the R-R interval time series are extracted, variability indexes are generated through frequency-domain decomposition and frequency-band energy statistics, the numerical values of time-domain and frequency-domain indexes are calculated and weight assignment is performed, so as to realize the comprehensive monitoring of the underwater physical sign state. The signal acquisition accuracy is improved by means of multi-source signal acquisition, noise interference is eliminated through signal trajectory matrix decomposition operation, the signal reconstruction quality is ensured by using eigenvector energy ratio calculation, the change trend of physical signs is accurately evaluated based on electrocardiogram R-wave feature extraction and frequency-domain decomposition, and the objectivity and credibility of the underwater physical sign state monitoring result are ensured by combining the weight assignment strategy. It has excellent performance in aspects such as signal acquisition, noise reduction processing, feature extraction and state evaluation, and provides a reliable guarantee for the vital sign monitoring of underwater operators. BRIEF DESCRIPTION OF THE DRAWINGS

[0005] Figure 1 is the system flow chart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0006] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. Please refer to Figure 1, the present invention provides a technical solution: a wearable underwater vital sign and environment monitoring system includes: a physiological signal acquisition module, which fixes a photoplethysmogram sensor at the wrist position, attaches a bioelectrical impedance electrode to the chest, and places a piezoelectric sensor at the neck position to collect the photoplethysmogram signal at the wrist, the bioelectrical impedance signal at the chest, and the piezoelectric signal at the neck, obtaining multi-source physiological signals, performing time sequence alignment and synchronous sampling on the multi-source physiological signals, and generating a physiological signal sequence; a signal noise reduction and reconstruction module, which constructs a signal trajectory matrix based on the physiological signal sequence, performs a block decomposition operation on the signal trajectory matrix, extracts signal eigenvalues and eigenvectors, obtains a set of characteristic components, calculates the energy ratio of the eigenvectors in the set of characteristic components, and generates a reconstructed signal group; a dynamic evaluation module, which extracts the position of the R-wave peak point of the electrocardiogram and the time series of the R-R interval of the electrocardiogram based on the reconstructed signal group, calculates the difference and fluctuation range between adjacent R-R intervals, obtains the heart rate change characteristics, performs frequency domain decomposition and frequency band energy statistics on the heart rate change characteristics, and generates a heart rate variability index; a vital sign parameter fusion module, which calculates the numerical values of the time domain index and the frequency domain index of the heart rate variability respectively based on the heart rate variability index, statistically analyzes the change trend and fluctuation range of each index, obtains the vital sign fluctuation parameters, assigns weights to the vital sign fluctuation parameters, and generates an underwater vital sign state quantity; an environment monitoring module, which is used to monitor the underwater environmental temperature through a temperature sensor unit, monitor the underwater environmental pressure through a pressure sensor unit, and monitor the underwater environmental salinity through an inductive conductivity sensor unit, obtaining underwater environmental monitoring data; a communication module, which is used to communicate the underwater vital sign state quantity and the underwater environmental monitoring data with a wearable display terminal and a water surface monitoring device through an underwater cable or a wireless communication method. The steps for obtaining the physiological signal sequence are as follows: based on the multi-source physiological signals, perform time sequence alignment, align each signal according to the time stamp, and adjust the signal data of each signal by interpolation or delay according to the acquisition time point of each signal, obtaining the time sequence aligned multi-source physiological signals; according to the time sequence aligned multi-source physiological signals, perform data synthesis, integrate all the aligned signals to a unified time point through data fusion, and generate a physiological signal sequence.Specifically, based on the multi-source physiological signals obtained previously, first read the timestamps marked by each signal source during the acquisition phase one by one, convert the timestamps of different signals into milliseconds uniformly for subsequent comparison, and then check the amplitude range of each signal source. For example, the photoplethysmogram signal is compared in the range of 0.0 mV to 5.0 mV, the thoracic bioelectrical impedance signal is compared in the range of 0.0 Ω to 150.0 Ω, and the piezoelectric sensor signal is compared in the range of 0.0 Pa to 1000.0 Pa. The upper and lower limits of these values are set empirically according to the specifications of the sensors used. If the amplitude of a certain record exceeds the corresponding range, it is marked as an abnormal record and not included in the interpolation process for the time being. Then, align each record in chronological order. If a gap with a time difference exceeding 10 ms is found between adjacent sampling points, the missing position is filled by linear interpolation. Assume the adjacent known sampling points are and , and the corresponding times are and , the time point to be interpolated is , then the interpolated value is calculated according to , where and represent the measured values of adjacent known points, is between and If the interpolated amplitude still falls within the aforementioned abnormal range, it is recorded as data to be confirmed for subsequent correction. After all interpolation is completed, all time points are rearranged in ascending order to form a unified time series. During this process, it is also necessary to check whether the time interval between any two records is less than 1 ms. If it is less than 1 ms, it is determined as a repeated sampling point and the mean is merged. Finally, all the records with time stamps and amplitudes matched are summarized to obtain a multi-source physiological signal with time alignment. According to the multi-source physiological signal completed with the above alignment, first determine the effective data index of each signal under the unified time reference. If it is found that the photoplethysmogram signal is missing at certain moments while the bioelectrical impedance signal and the piezoelectric sensor signal are effective, retain the effective signals and generate marker entries at the missing positions to prevent omission. Then, combine all the aligned signal records in order according to the time points. For multiple amplitude values that appear at the same moment, the average value fusion method can be used for synthesis. At the same time, judge whether there is a significant deviation according to the amplitude difference threshold set according to experience. For example, if the amplitude difference between the photoplethysmogram and the piezoelectric sensor at the same time point exceeds 10.0 units, and this threshold is obtained from the statistical analysis of historical observation data, it is marked as a difference entry during fusion. Next, re-check all the merged entries and compare whether there are new abnormalities in the range of intervals such as 0.0 to 5.0 mV, 0.0 Ω to 150.0 Ω, and 0.0 Pa to 1000.0 Pa. If there are no new abnormalities, continue the synthesis operation. If new abnormalities occur, record their moments and briefly compare the values at adjacent moments. If the amplitude difference is still greater than the empirical threshold after comparison, it is temporarily retained as an abnormality for subsequent correction. After completing the fusion of all time points, a time series can be sorted out. This series contains the data values of the three signals at the unified time point, and finally a physiological signal series is generated. The steps for obtaining the set of characteristic components are as follows: According to the physiological signal series, generate a two-dimensional matrix by arranging the signal values sampled at each time point. Take the data at each time point as a row vector and the signal channel data as a column vector to obtain a signal trajectory matrix; According to the signal trajectory matrix, calculate the characteristic score of signal decomposition. The calculation formula is: Wherein, is the characteristic score, is the singular value of the signal trajectory matrix, is the corresponding left singular vector, is the number of blocks; according to the feature scores, the matching components of the feature values and feature vectors are extracted, and the block feature scores are aggregated to generate the corresponding signal feature value and feature vector set, obtaining the feature component set. Specifically, based on the physiological signal sequence obtained previously, arrange them in sequence according to the values at each sampling moment and ensure that each time point matches the corresponding signal amplitude data. Among them, the photoplethysmogram signal can be detected in the effective range of 0 mV to 5 mV, the thoracic bioelectrical impedance signal can be compared in the range of 0 Ω to 150 Ω, and the cervical piezoelectric signal can be inspected in the range of 0 Pa to 1000 Pa. These interval data are obtained by organizing multiple consecutive acquisitions and records of the actual sensor data. Then, horizontally combine all signal amplitude values according to the index order of time points, and organize the three signal amplitude values corresponding to each timestamp into a row. During this process, the time-series synchronous data needs to be re-verified. If there are individual timestamps exceeding the tolerance deviation of 1 ms, linear interpolation is performed additionally. Combined with the amplitude comparison detection of adjacent data, for example, when the amplitude difference between the photoplethysmogram and the cervical piezoelectric signal at the same moment exceeds 10, the threshold of this difference is obtained from the cumulative observation statistics and will be marked as potentially abnormal and enter the subsequent comparison. If it is found to be an extreme mutation in the subsequent analysis, record this moment and prepare for subsequent data correction. At the same time, it is also necessary to check whether the bioelectrical impedance signal has too large a difference in amplitude from the previous and subsequent timestamps in the same row. If it exceeds the 20 Ω threshold obtained from the previous acquisition statistics three times in a row, mark this row as an abnormal row for further verification. Then continue to organize all sampling points in the form of row vectors. Arrange the three-channel amplitude values of each row from left to right in sequence. The leftmost column is the photoplethysmogram data, the middle column is the bioelectrical impedance data, and the rightmost column is the piezoelectric sensor data. After completion, number each row sequentially to correspond to the chronological order. Then, divide the column vectors in the vertical direction according to the channel category. The number of column vectors formed is equal to the number of signal channels and keeps the upper and lower positions consistent. Each column in the file from top to bottom is the amplitude value of the same signal at different time points. Since the anomalies have been marked through threshold detection previously, only the index association of the values of each row and column is required here. Finally, matrix storage of these rows and columns can obtain the signal trajectory matrix. The benefit of the formula is that by accumulating the product of the block singular values and the amplitudes of the corresponding left singular vectors, it can highlight the proportion of different blocks in the signal space, and quantify the contribution degree of each block to the overall signal change into feature scores, so as to use this value to compare with the threshold in the subsequent steps to extract the corresponding effective features; The acquisition steps are as follows: First, perform singular value decomposition on the signal trajectory matrix and record each element in the singular value vector. These elements are from the re-decomposition results of the matrix after multiple blockings, and are obtained by calculating the data amplitude and time index in each block. The specific blocking method is to cut the original matrix by rows or columns, keep the sub-matrices after cutting non-overlapping, then separately obtain the singular value vector for each sub-matrix, and finally merge the singular values of all sub-matrices according to the index position to obtain... , where The value of... will vary depending on different block sizes. For example, after multiple measurements and statistics, if the block is set to 3 blocks, it is recorded that... = 2.3, = 1.8, = 0.9; The acquisition steps of... are as follows: First, calculate the amplitude of the left singular vector... obtained when decomposing each sub-matrix, and obtain the value by accumulating the squares of the data amplitudes. For example, regard... as a vector of length m, and its elements can be formed by reading each row of the blocked matrix and normalizing it. The corresponding... , and... is obtained in the same way. , and... and , where The specific value of... is determined by the number of rows of the sub-matrix after blocking. Assume that... = 200 is obtained through preliminary statistics, and... is calculated through actual measured data. Calculation process: First step, list the known parameters: Second step, multiply and accumulate each term according to the formula: Third step, calculate each product value and add them in sequence: The result shows that the feature score after overall chunking is 5.463. When the feature score after chunking is higher than a certain range, it indicates that the corresponding fluctuation of the signal trajectory matrix may be relatively large. If the value falls between 2 and 4, it means that the contribution of the chunks within this range is relatively balanced. If it exceeds 5, it means that some chunks may have a higher amplitude concentration. Therefore, in the follow-up, more refined feature extraction can be carried out in combination with other indicators. According to the previously obtained feature scores, first read the corresponding index information of the singular values and left singular vectors in each chunk, and find the index range corresponding to the chunk position in the previously recorded singular value sequence. For example, if the number of chunks is 3, retrieve the data at positions 1 to 3 in the index cut-off points of the singular value sequence, and cross-check in combination with the previously calculated feature scores to confirm whether the values of each chunk are within the previously set chunk amplitude range. If the chunk amplitude of chunk 1 exceeds 3.0 in the corresponding index range, it is marked as a large value. This threshold of 3.0 is obtained by statistical induction after continuously collecting multiple groups of matrices. Find all the index rows with high or low amplitudes in all chunks and compare them one by one with the squared amplitudes of the left singular vectors. If the squared amplitude continuously exceeds 1.0, record its position and compare the data differences between the previous and subsequent rows. If three consecutive rows are all greater than 1.0, include them in the candidate sequence of feature components and continuously observe the subsequent indexes. Extract and combine the above-recorded chunk information and singular value information to form a matching list of eigenvalues and eigenvectors. During this process, it is necessary to ensure that each pair of eigenvalues and left singular vectors has a unique index identifier to avoid mismatching. Then merge the eigenvalues and eigenvectors of all chunks in order and form a unified set of matching items based on the same index mechanism as before. Keep separate annotations for the already marked abnormal or overly high amplitude indexes for later verification, and further complete the aggregation of the feature scores of each chunk. When aggregating, if it is found that the singular values corresponding to a certain segment of indexes are significantly higher than those of other index segments, for example, exceeding 2 times the statistically averaged value, record these index segments and maintain their corresponding relationship with the original sequence, and finally wait to be used in the subsequent steps. After completing all chunks, obtain the corresponding signal eigenvalue and eigenvector set, and after summarization, the final feature component set can be formed. The steps to obtain the reconstructed signal group are as follows: According to the feature component set, extract the eigenvectors and the corresponding singular values, rearrange the eigenvectors and singular values in matrix form to generate an eigenvector matrix; Based on the eigenvector matrix, calculate the energy ratio, and the calculation formula is: where is the energy ratio, is the th component value of the reconstructed signal, is the th component value in the eigenvector set, is the number of selected principal components, is the total number of eigenvectors; perform a weighted combination of the eigenvectors with energy proportions higher than the threshold and the corresponding singular values, perform inverse singular value decomposition, and generate a reconstructed signal group. Specifically, according to the previously obtained set of eigencomponents, read the corresponding vectors and singular values of each eigencomponent and list them one by one, and check the arrangement relationship of these vectors in the index order one by one. If it is detected that the source of continuous eigencomponents is inconsistent with the index of the previous block result, it is necessary to review the construction method of those eigenvectors to avoid dislocation, and then rearrange all the corresponding eigenvectors and singular values in column or row order. During the arrangement process, first record the dimension of each vector, for example, record the length of one vector as 100 and the length of the other vector as 120. These data of different lengths need to be supplemented with blank entries or redundant entries are truncated in the corresponding order of the feature items. For example, if it is found that there is If the terminal index of some vectors exceeds the required range, the excess part will be removed and its specific position will be recorded. The singular values need to be aligned accordingly. If a singular value does not match the index marked in front of the eigenvector, it will be rechecked before the final merge. Then, a row-column corresponding matrix is constructed for all the checked eigenvectors and singular values. The eigenvector content is placed in the column direction, and the row direction corresponds to the serial number of each singular value. At this time, the exact number of columns and rows must be determined to avoid empty columns or rows. After all alignments are completed, these elements are uniformly indexed and formed into the final eigenvector matrix. The final matrix will be used in the subsequent energy weight calculation. The formula is beneficial in that by selecting the number of principal components The ratio operation of the sum of squares of the eigenvector components in the eigenvector to the sum of squares of all eigenvector components can directly determine the energy contribution of the currently selected principal component to the overall eigenvector set, and use this ratio to guide the subsequent reconstruction or filtering process; and The acquisition step is to read each vector element in the eigenvector matrix. The values of these elements come from the amplitude data obtained after multiple signal acquisitions and decompositions. Each vector element can be regarded as an amplitude component within a sampling range. For example, in some test scenarios, the pulse wave amplitude recorded falls between 0mV and 5mV, the impedance amplitude is between 0Ω and 150Ω, and the piezoelectric amplitude is between 0Pa and 1000Pa. The amplitudes of different channels are merged into the eigenvector to form component values. Statistics show that there will be n total components. The steps of obtaining are: first, analyze the variance contribution of all feature components, observe the arrangement of components from high to low according to variance contribution, and then select the first m components with the highest proportion for subsequent principal component processing. The specific value of m can be determined after comprehensively comparing the energy proportion curve and the recognition accuracy requirements. For example, when it is found that selecting the first three components in several groups of test data can achieve a more comprehensive coverage, m is set to 3. It can also be added or reduced later to match the new data distribution; The acquisition steps are as follows: count all the component numbers in the feature vector matrix, the value of which is determined by the scale of the previous singular value decomposition and vector merging. For example, in a certain experiment, multi-source signals are collected, and each signal generates several feature vectors during the block processing. After summarization, they are merged into a total of 8 vectors, and each vector contains several components. After totaling all the components, n may reach several hundred or thousands, and this value needs to be confirmed in combination with the actual data acquisition volume; Calculation process: First step, list the numerical values of each parameter in the test case, such as selecting the number of principal components , the total number of components of the feature vector , first record all the component values as Second step, calculate the denominator : Third step, take the first 3 components as the principal components, and the numerator is calculated as: Fourth step, substitute into the formula: The result shows that the three selected principal components account for approximately 70.9% of the overall component energy. When this ratio exceeds 80%, it may indicate that the number of selected principal components covers the overall energy more adequately. When the ratio is too low, the number of principal components should be increased or decreased according to the actual situation to meet the subsequent reconstruction or separation requirements. Based on the energy ratio values calculated previously, first match all eigenvectors with their corresponding singular values in the order of their indices. Then, read the indices of each vector with an energy ratio greater than the threshold in the record and mark its position in the eigenvector matrix. The threshold value needs to be set in combination with the previously obtained energy ratio distribution. For example, after multiple tests, it is found that when the energy ratio is higher than 0.8, there is a more stable reconstruction quality, so the threshold can be set to 0.8. Then, multiply the eigenvectors corresponding to these indices by the singular values to complete the weighting operation. If an eigenvector is repeatedly included during the index traversal, count the number of times and accumulate the corresponding coefficients during subsequent weighting. The size of the coefficients can refer to the amplitude range of the previous block results. For example, for photoplethysmogram signals, the stability can be checked between 0 mV and 5 mV, and for thoracic bioimpedance signals, the amplitude fluctuations can be monitored between 0 Ω and 150 Ω. After normalizing these fluctuation values and combining them with the singular values, a more refined weighting result can be obtained. After finishing the arrangement, perform the inverse singular value decomposition on all the selected weighted eigenvectors. During this process, the length and position of each vector need to be kept consistent, arrange them one by one and perform matrix operations with the weighted singular values. After the operation, a series of time-sequence amplitude sequences will be generated. Summarize these sequences to obtain the reconstructed signal group. Finally, subsequent analysis and feature extraction can be performed on this reconstructed signal group in the following steps. The steps for obtaining the heart rate change characteristics are as follows: According to the reconstructed signal group, process each electrocardiogram signal channel, extract the peak point positions of the R waves through waveform analysis and threshold determination, and arrange the peak point positions in a time series manner to obtain the electrocardiogram R wave peak point position sequence; Based on the electrocardiogram R wave peak point position sequence, calculate the nonlinear fluctuation index, and the calculation formula is: where is the nonlinear fluctuation index of adjacent R-R intervals, is the time position of the th R wave peak point, and are the time positions of the previous and the two previous R wave peak points respectively, is a dimensionless constant for positive adjustment to avoid a zero denominator, is the total number of R waves; according to the non-linear fluctuation index, combined with the time series analysis of the electrocardiogram signal, by calculating the amplitude distribution characteristics of the R waves, the heart rate change characteristics are generated. Specifically, according to the reconstructed signal group obtained previously, first, for each electrocardiogram signal channel, the amplitude values at different sampling times are read row by row and index comparison is performed. If it is observed that the amplitude is continuously higher than the empirical threshold, it is marked. For example, the R wave amplitude detection threshold is set to 0.5 mV, which is obtained by multi-day recording and grouped statistics of the electrocardiogram data collected from the actual population sample. Then, all amplitude records are arranged in chronological order and the adjacent amplitude differences are monitored. If there is a sudden increase in amplitude at certain moments, the waveforms within about 20 ms before and after this moment are further scanned to determine whether there is a prominently shaped R wave peak. During the scanning, a morphological standard for determination can be referred to. For example, the main feature of the R wave in the QRS complex is an upward sharp waveform and the peak position has a relatively high amplitude. Each already marked potential peak point is re-verified by combining the slope change rate of the waveforms before and after. If the slope change of the peak point in the adjacent neighborhood exceeds the set slope threshold, then the peak point is confirmed as the R wave peak. The slope threshold can be determined by extracting the eigenvalue distribution from the daily monitoring data of multiple different subjects. For example, after multiple inductions, the critical slope range is between 0.3 mV / ms and 0.6 mV / ms. If the observed value exceeds this range, it is initially determined that the peak has sufficient discrimination. The time positions of all peak points passing the threshold and slope verification are recorded item by item, and the software aligns the time positions at the millisecond level. If there are a small number of time indexes with too small intervals in the records, it can be checked in combination with the adjacent amplitude differences. If it is found that the adjacent peak interval is less than 200 ms, the amplitudes and slopes are compared to select a peak that more conforms to the R wave morphology and the other one is excluded to avoid duplication. After the peak identification of all channels is completed, the peak points are rearranged in chronological order to form a unified peak index sequence, and finally the electrocardiogram R wave peak point position sequence is obtained. The advantage of the formula is that by performing difference and non-linear correction on the time positions of adjacent R wave peak points, the rhythm fluctuation amplitude in different time periods can be evaluated and the potential abnormal change degree can be prompted, so that it has a good fluctuation measurement when monitoring the heart rate regularity or rhythm in the follow-up; The acquisition step of is to sequentially extract the peak time positions with index k from the electrocardiogram R wave peak point position sequence obtained previously. These time positions are synchronized with the electrocardiogram data through a millisecond timer. For example, during a 24-hour acquisition, the occurrence time of each R wave is recorded as a certain moment among tens of thousands of data points; and The acquisition step of is to push forward several indexes within the same recording sequence and read the corresponding time positions, which can be retrieved forward in the existing peak position list without additional measurement; The acquisition steps are as follows. When processing multi-day electrocardiogram data, if it is found that the denominator part may have a minimum value, leading to a division-by-zero problem, a positive adjustment value is selected. For example, after analyzing thousands of R-wave peak records, it is summarized that if is set, it can balance extreme cases and prevent the denominator from becoming zero; The acquisition steps are as follows. Count the total number of R-wave peaks marked in this batch of electrocardiogram data. For example, after obtaining 80,000 data points through 24-hour monitoring and identifying approximately 90,000 sampling peaks, among which there may be 85,000 valid R-wave peaks, then at this time xm = 85,000; Calculation process: First step, select the electrocardiogram R-wave peak time series in a certain test scenario. For example, take out 5 moments in sequence as , and let , xm = 5; Second step, calculate the numerator : Third step, calculate the denominator : For simplification, it can be roughly calculated first: ln(801) , ln(1601) , ln(2401) , ln(3151) , ln(4051) , and 1 / 800 , 1 / 1600 , 1 / 2400 , 1 / 3150 , 1 / 4050 , add these up to get approximately 38.210, and then multiply by , to get 0.038210; The denominator is comprehensively ; Fourth step, substitute into the formula to get: The results show that this group of R-wave peak sequences has a certain non-linear fluctuation amplitude within this interval. When the H value exceeds 0.6, it may indicate an enhanced rhythm fluctuation. When the H value is below 0.3, it indicates that the fluctuation within this interval is relatively stable. In this example, H is approximately 0.4396, indicating that the degree of fluctuation is in the medium range. Combining with the analysis of the R-wave amplitude distribution characteristics in the subsequent steps, the characteristics of heart rate changes can be further evaluated. According to the non-linear fluctuation index obtained previously, first read the actual amplitude of each peak from the recorded list of R-wave peaks and summarize it into a separate distribution sequence. When checking the amplitude range of this distribution sequence, the amplitude can be compared with the range of 0.0 mV to 5.0 mV, and the situation where any single peak amplitude is continuously higher than 4.5 mV for three times can be marked. This 4.5 mV value is selected after statistically analyzing the peak conditions of more than a hundred people during daily monitoring. If the phenomenon of peak amplitude lower than 0.1 mV appears in the record at the same time, it is confirmed that these peaks may have noise or equipment contact problems and are marked independently. Then, all R-wave amplitudes are arranged in ascending order and their distribution frequencies are counted. If there are high-density peak concentration paragraphs within the visible range, the frequencies of these paragraphs need to be compared in the subsequent operations to provide a reference for analyzing the actual amplitude changes. At the same time, the change rate of adjacent peak amplitudes is tracked in chronological order. For example, it is detected whether the amplitude difference between three to five consecutive peaks exceeds the previously set threshold of 0.8 mV. After recording the samples that exceed this threshold, their time indices are recorded and paired with the non-linear fluctuation index obtained previously. It is compared whether there is a large H value fluctuation at the same moment. If both are in the high-value range, this period of time can be listed as a high-fluctuation paragraph. After the above recording work is completed, the characteristics of heart rate changes can be integrated and output according to the distribution, and finally the characteristics of heart rate changes are generated. The steps to obtain the heart rate variability index are as follows: According to the characteristics of heart rate changes, the characteristics of heart rate changes are transformed from the time domain to the frequency domain through Fourier transform, different frequency components and amplitude values are extracted to obtain the frequency-domain heart rate characteristics; Based on the frequency-domain heart rate characteristics, calculate the proportion of the energy of the target frequency band, and the calculation formula is: where is the proportion of the energy of the frequency band, is the frequency at the power spectral density, and are the lower and upper limit frequencies of the target frequency band, and are the minimum and maximum frequencies of the entire frequency spectrum; by combining the information of the main frequency components in the frequency domain, the energy ratio of multiple frequency bands is statistically calculated and summarized into an index set to generate heart rate variability indexes. Specifically, according to the heart rate change characteristics obtained previously, first read the timing information of each heart rate change record and recombine it in an equally spaced form. After aligning the heart rate amplitudes of adjacent records at the millisecond level, perform discrete Fourier transform successively to obtain the amplitude distribution corresponding to different frequency points. In specific operations, it can be marked according to the time index of each heart rate record. If the sampling time of a certain record exceeds the preset 5ms interval compared with the previous one, and this 5ms value is determined by observing the heart rate sampling distribution of multiple subjects, then linear interpolation is performed at this position and the amplitude after interpolation is recorded. Subsequently, during the discrete Fourier transform, unified processing is performed on the entire time series. Arrange all the sampled amplitude values in order and use them as the transform input. When the transform is completed, frequency components covering the range from 0Hz to dozens of Hz can be obtained. The extremely low frequency region below 0.05Hz and the ultra-high frequency region above 40Hz are excluded or separately marked. These frequency intervals are determined according to the common analysis intervals in the study of human heart rate characteristics. For example, through multiple verifications of the heart rate data monitored continuously for one day (24 hours), it is found that most of the main information is concentrated in the range of 0.05Hz to 2.0Hz. The components outside this range are mostly related to environmental interference or relatively extreme situations. After excluding or marking these sections, record the amplitude values of the remaining part and match them one by one with the frequency index. If frequency points with amplitude values continuously exceeding the established threshold are detected, for example, exceeding 1.2 times the maximum amplitude mean obtained in the previous statistics, then these points are marked as frequencies of concern. Finally, summarize all the retrieval results to form a frequency domain data containing amplitude and corresponding frequency information, and classify according to the detected multiple frequency components and amplitude sizes. For example, the section from 0.05Hz to 0.15Hz is recorded as the ultra-low frequency band, 0.15Hz to 0.4Hz is recorded as the low frequency band, 0.4Hz to 1Hz is recorded as the middle frequency band, and above 1Hz can be regarded as the high frequency band. These segments are divided according to the conventional research range of human heart rate variability. After the processing is completed, the frequency domain heart rate characteristics are obtained. The benefit of the formula is that by calculating the ratio of the power spectral density within a specific interval to the power spectral density of the overall spectral interval , it can quantify the energy proportion occupied by the target frequency band and distinguish the relative contribution degrees of each segment; is obtained by squaring the amplitude spectrum obtained after Fourier transform of the heart rate time series signal to obtain the power spectral density. Each frequency point corresponds to an amplitude value, and the amplitude value can be obtained by the sampling device in multiple measurements. Squaring the amplitude again can obtain the power; and are obtained by selecting the target frequency band through the previous analysis of the heart rate change law. For example, when studying the low frequency band of heart rate variability, it can be set Hz, Hz, and the specific value is usually obtained based on clinical or experimental observation and statistical methods; and The acquisition steps of are as follows: Statistically analyze the entire frequency range obtained after Fourier transform, and take the minimum and maximum values as boundaries. For example, if the sampling frequency is 100 Hz and the data length is fixed, the frequency spectrum range may be from 0 Hz to 50 Hz. In population heart rate analysis, the interval from 0 Hz to 2 Hz is often of concern. Therefore, it can be set that Hz, Hz; Calculation process: First step, in actual measurement, assume that the heart rate change of a certain subject is monitored and the power spectrum covering the range from 0 Hz to 2 Hz is obtained. At the same time, the target frequency band is [0.04, 0.15], and the values within this interval are integrated; Second step, the values of the entire frequency spectrum interval (here it is from 0 Hz to 2 Hz) are also integrated in the same way to obtain the denominator; Third step, divide the two integration results to obtain the energy proportion ; In a given example, let (estimated from the actual integral values of the power spectra of a large number of heart rate samples), and the integration result of the entire interval ; Fourth step, substitute into the formula: The results show that the current target frequency band (0.04 Hz to 0.15 Hz) accounts for 30% of the overall energy. If the energy proportion of a certain frequency band is higher than 50%, it may indicate that the activity of this frequency band is relatively active. If it is lower than 10%, it means that the contribution of this frequency band to the whole is small. Based on this value, the significance of the distribution of heart rate variability in different frequency bands can be further judged. Combining the main frequency component information in the frequency domain previously, first, select each main frequency position from the amplitude spectrum data obtained by Fourier transform and record the corresponding frequency and amplitude values. It is necessary to check the surrounding frequency bands of each main frequency position. If it is detected that the amplitude value continuously exceeds 1.2 times the average amplitude obtained from the previous statistics, then this section is determined as a high-amplitude interval and marked. At the same time, calculate the energy proportion section by section according to the overall range from 0 Hz to 2 Hz. For example, several adjacent intervals can be set, where 0.04 Hz to 0.15 Hz is the low-frequency band, 0.15 Hz to 0.4 Hz is the middle-frequency band, 0.4 Hz to 1 Hz is the high-frequency band, and 1 Hz to 2 Hz is the higher-frequency band. Then calculate the energy proportion for each interval. If it is found that the proportion continuously exceeds the previously recorded historical average value twice in a certain interval, then it is marked as an active section again. After completing the statistics of all frequency bands, collect these proportion values and combine them with their corresponding main frequency components to form an index set for this section of data. Finally, organize and output them into heart rate variability indicators based on the multi-frequency band energy proportion and main frequency position data contained in this index set. The steps to obtain the physical sign fluctuation parameter are as follows: Classify the heart rate variability indicators into two parts: time-domain characteristics and frequency-domain characteristics. Statistically analyze the time-domain characteristics including standard deviation and root mean square interval, and classify the frequency-domain characteristics including frequency band energy ratio and main frequency to obtain the classified heart rate variability indicators; Based on the classified heart rate variability indicators, calculate the change trends and fluctuation ranges of each time-domain characteristic and frequency-domain characteristic, and statistically analyze the extreme values, ranges, and average change amounts between the indicators to obtain the change trends of the statistical indicators; Based on the change trends of the statistical indicators, calculate the physical sign fluctuation parameter, and the calculation formula is: Among them, is the physical sign fluctuation parameter, is the change value of the th time-domain characteristic, is the change value of the th frequency-domain characteristic, is the proportion of the th frequency-domain characteristic, is the total number of features. Specifically, the heart rate variability indexes are classified into two parts: time-domain features and frequency-domain features. First, confirm the statistical methods required for the two indexes of standard deviation and root mean square interval in time-domain features. Extract the continuously recorded heart rate values or R-R intervals from the record dataset and arrange them in chronological order. Read each item of time-domain features one by one and calculate the standard deviation. When calculating, you can first square the difference between all heart rate values and their average value and then take the average, and finally take the square root of the result to obtain the standard deviation. The root mean square interval can also be obtained by performing a secondary operation on the difference between adjacent heart rate values or R-R intervals, taking the average, and then taking the square root. If any record lacks valid values, mark this place as data missing and summarize it in subsequent comparisons. If it is found that some time-domain features deviate greatly or extremely, conduct additional inspections and record the number of times the threshold is exceeded. These thresholds are determined by the statistical results of continuous monitoring of the same population for several weeks. For example, the situation where the standard deviation is greater than 70 ms or the root mean square interval is greater than 100 ms is recorded as a significant deviation. Then, compare the frequency-domain information in the heart rate data to obtain the frequency-domain features. For the frequency-domain features, first select several key intervals from the power spectrum data obtained earlier, such as the low-frequency band from 0.04 Hz to 0.15 Hz and the high-frequency band from 0.15 Hz to 0.4 Hz, and count the power ratio within these intervals. Then, read the amplitude information of the sampling points at the main frequency position. If the power ratio observed within a specific interval exceeds 1.3 times the historical average, record it as a band with large fluctuations. The low-frequency power ratio and high-frequency power ratio obtained above, together with the main frequency value, are summarized as frequency-domain features. When summarizing the results, it is necessary to ensure that each feature item has a corresponding time index for subsequent comparison. If individual intervals lack valid frequency components, mark these intervals as null values and do not include them in the mean calculation. Through this process, feature collection can be carried out in terms of time domain and frequency domain. Record and classify the obtained standard deviation, root mean square interval, band energy ratio, and main frequency in sequence to form a set of classified heart rate variability indexes.Based on the classified heart rate variability metrics, first read the standard deviation and root mean square interval in the time domain features one by one and record their maximum and minimum values at multiple sampling stages. If the standard deviation exceeds the set upper limit of 150 ms within a certain period, it is marked as a maximum value record. This 150 ms is determined by conducting multiple grouped statistics on a large amount of data. At the same time, if the root mean square interval is lower than 10 ms during continuous monitoring for several hours, it is marked as a minimum value interval. These extreme values are incrementally statistically analyzed in the software and combined with the time range of each sampling period to obtain a characteristic change curve. Then, the same operation is performed on the frequency domain features, extracting the maximum or minimum values of the energy ratio of each frequency band at different times. If the energy ratio reaches or exceeds 50% within any period and exceeds twice the average value obtained from previous continuous observations, it is recorded as a high-energy segment and marked with a time index. If the main frequency position shows continuous upward or downward trends in several detections and the amplitude spans more than 0.1 Hz, it also needs to be recorded separately. The threshold of 0.1 Hz is obtained based on the conventional fluctuation range found in multiple groups of human body monitoring. Then, a horizontal comparison is made of the above-mentioned metrics, and the peak-valley range and average change value they experience during the sampling period are statistically analyzed. If it is found that both the root mean square interval of the time domain features and the high-frequency band energy of the frequency domain features show significant fluctuations, the change amplitude and time period are summarized and recorded, and extreme value screening is performed. Finally, based on the maximum values, minimum values, ranges, and mean fluctuations of all time domain features and frequency domain features, the overall index change trend is plotted. After finishing the collation, the statistical index change trend is obtained. The benefit of the formula is that it comprehensively considers the change values of time domain features. , the change values of frequency domain features and the proportion of frequency domain features , comprehensively measure the linkage between various features and correct the frequency domain changes using a logarithmic function, so as to more precisely reflect the status of different types of features in the overall fluctuation; The acquisition steps of are as follows: Select the difference of the th time domain feature (such as standard deviation or root mean square interval) within several monitoring periods. The specific value can be obtained by subtracting the measurement results of different periods and performing statistics. For example, in a certain observation, the th time domain feature is 30 ms in the early monitoring stage and 50 ms in the later monitoring stage, then ms; The acquisition steps of are as follows: Extract the difference before and after for the corresponding frequency domain feature (such as the low-frequency band energy ratio or high-frequency band energy ratio) from multiple measurements. If the low-frequency band energy ratio rises from 20% to 32% within a certain period, then the corresponding . It is necessary to ensure that the data comes from the same feature item and take the difference of the monitoring results of the same order; The acquisition steps of are as follows: Read the proportion of the frequency domain feature. For example, the overall energy proportion of a certain frequency band is obtained from the previous spectrum integration result. If the proportion of this frequency band is 25%, then , this value can be regarded as the relative weight of the feature in the frequency domain; The acquisition steps of are as follows: After counting the total number of all features included in the analysis, for example, if there are 2 time-domain features (standard deviation, root mean square interval) and 2 frequency-domain features (low-frequency band energy ratio, high-frequency band energy ratio), then ; Calculation process: First step, list each parameter obtained from actual observations: Let , assuming that these are the differences and proportions obtained for two time-domain features of standard deviation and root mean square interval, and two frequency-domain features of low-frequency band and high-frequency band; Second step, calculate the specific values of respectively, for : when: For when: Third step, square and add the two and then take the square root: The results show that after synthesizing the changes of the above two time-domain characteristics and two frequency-domain characteristics, the physical sign fluctuation parameter is approximately 6.151. If this value is higher than 10, it indicates that the changes of multiple characteristics are large and the weight of the frequency-domain proportion is high. If it is lower than 2, it means that the overall change range is limited and the differences of most characteristics are small. In actual scenarios, thresholds can be further established to determine whether the physical signs enter the high-fluctuation range or remain stable. The steps for obtaining the underwater physical sign state quantity are as follows: According to the physical sign fluctuation parameter, it is divided into time-domain characteristics and frequency-domain characteristics according to the characteristic type. The change value of each item in the time-domain characteristics and the energy value of each item in the frequency-domain characteristics are extracted respectively, and data normalization processing is performed according to a unified range. The normalized data is recombined to form a set of normalized physical sign fluctuation parameters; based on the set of normalized physical sign fluctuation parameters, through the correlation between characteristics, the similarity between each characteristic is calculated, and the correlation relationship between characteristics is represented in matrix form. The characteristics are merged and recalculated to obtain a set of fused physical sign fluctuation parameters; according to the set of fused physical sign fluctuation parameters, weight values are assigned according to the classification criteria of time-domain characteristics and frequency-domain characteristics, and the weighted characteristics are summed up to generate the underwater physical sign state quantity. Specifically, according to the previously obtained physical sign fluctuation parameter, first, the numerical ranges of each time-domain characteristic and frequency-domain characteristic are counted and the upper and lower limits of these ranges are recorded. These upper and lower limits can be obtained by grouping and screening the continuous multi-day sampling results and taking their minimum and maximum values. Then, it is divided into time-domain characteristics and frequency-domain characteristics according to the characteristic type. The change value of each item in the time-domain characteristics and the energy value of each item in the frequency-domain characteristics are read separately and data normalization processing is performed according to a unified reference interval. For example, the data in the region from 0 to the minimum value is uniformly mapped to 0, and the part from the maximum value to a higher value, if it exists, is linearly extended or removed. At the same time, it is ensured that special markings are given to extremely over-limit data during the normalization process. For example, when a certain time-domain characteristic shows a recorded value higher than the previously statistically obtained upper limit of 200 ms, this value is uniformly set to 1 and marked as an over-limit point during normalization. This 200 ms upper limit is obtained through percentile analysis of the heart rate sampling records of continuous monitoring for hundreds of hours. After normalization, the final data of each characteristic item is indexed and checked item by item to avoid corresponding confusion. If some characteristics are missing during a certain sampling stage, corresponding empty records are reserved in the normalization list and their time indexes are marked. Finally, all processed time-domain characteristics and frequency-domain characteristics are merged into a new sequence in the same time order. When merging, if it is found that the number of time-domain characteristics is inconsistent with the number of frequency-domain characteristics, interpolation or zero-padding processing is performed according to the previously recorded time indexes to ensure that the final number of time-domain and frequency-domain characteristics can correspond one by one. After completing this step, the normalized data can be recombined to form a set of normalized physical sign fluctuation parameters.Based on the normalized set of physical sign fluctuation parameters, first, compare the numerical similarity between each pair of feature items one by one and select a suitable metric to calculate the similarity. For example, for the change value of time-domain features, the ratio of the absolute value of the numerical difference can be used for comparison. For the energy value of frequency-domain features, the difference and its proportion can be comprehensively considered. If the differences of the same feature in multiple adjacent time periods are all lower than the 0.05 threshold set by statistics, it is considered to have a high similarity. Here, this 0.05 is obtained by summarizing the variance distribution of multiple measured values after normalization and selecting a standard that deviates from the mean by about 5%. Then, conduct pairwise comparison for each pair of features and record the obtained similarity in the corresponding position in the matrix. If the similarity of some feature pairs is lower than 0.1 in multiple samplings, they can be regarded as relatively independent. This 0.1 is also set with reference to the average difference of multiple previous groups of heart rate samplings. Then, carry out feature merging according to the similarity matrix, group the features with high similarity into the same category or smooth-combine their results. During this process, if a certain feature contains multiple numerical components and has a large distribution span, this feature needs to be split into multiple sub-items and then compared separately. After the comparison is completed, recalculate the merged result. For example, uniformly weight or take the weighted average of multiple time-domain features with high similarity, perform the same steps for frequency-domain features, and correct the weighting coefficient with the previous proportion information. Finally, record and assemble these recombined and recalculated results one by one into the fused set of physical sign fluctuation parameters. According to the fused set of physical sign fluctuation parameters, first, read the corresponding category of each feature in turn according to the classification criteria of time-domain features and frequency-domain features and assign corresponding weight values to them. This weight assignment can be implemented by combining the importance scores of time-domain and frequency-domain indicators in historical monitoring data. For example, the weight of time-domain features can be set between 0.4 and 0.6 and slightly adjusted up or down according to the previous deviation degree from the standard deviation and root mean square interval. The weight of frequency-domain features can be placed in the range of 0.3 to 0.5 and use increasing or decreasing factors according to the actual proportion of the low-frequency band and high-frequency band. The specific values of these weights are determined by statistically comparing multiple measurements and comprehensively using the expert experience method. If it is found in multiple batches of data that the energy of the low-frequency band has a greater impact on the overall heart rate state change, the weight of the low-frequency band can be increased accordingly. Finally, sum up the weighted feature values at the same time index. When summing up, first read the combined results of time-domain and frequency-domain item by item and multiply each by its corresponding weight, and then sum up all the products to obtain a representative value. If any feature value is missing at a certain moment, mark this moment as vacant during weighting and skip the summation or assign its weight to other features. After all moments have been calculated, the summation results in time series can be further recorded. Thus, the underwater physical sign state quantity is generated.

Claims

1. A wearable underwater vital sign and environmental monitoring system, characterized in that, The system includes: a physiological signal acquisition module, which fixes a photoplethysmogram sensor on the wrist position, attaches a bioelectrical impedance electrode to the chest, and places a piezoelectric sensor at the neck position to collect the photoplethysmogram signal at the wrist, the bioelectrical impedance signal at the chest, and the piezoelectric signal at the neck, obtaining multi-source physiological signals, performing time series alignment and synchronous sampling on the multi-source physiological signals, and generating a physiological signal sequence; a signal denoising and reconstruction module, which constructs a signal trajectory matrix based on the physiological signal sequence, performs a block decomposition operation on the signal trajectory matrix, extracts signal eigenvalues and eigenvectors, obtains a set of eigen-components, calculates the energy ratio of the eigenvectors in the set of eigen-components, and generates a reconstructed signal group; a dynamic evaluation module, which extracts the position of the electrocardiogram R-wave peak point and the electrocardiogram R-R interval time series based on the reconstructed signal group, calculates the difference and fluctuation range between adjacent R-R intervals, obtains the heart rate change characteristics, performs frequency domain decomposition and frequency band energy statistics on the heart rate change characteristics, and generates a heart rate variability index; a physical sign parameter fusion module, which calculates the numerical values of the time domain index and the frequency domain index of the heart rate variability respectively based on the heart rate variability index, statistically analyzes the change trend and fluctuation range of each index, obtains the physical sign fluctuation parameters, assigns weights to the physical sign fluctuation parameters, and generates an underwater physical sign state quantity; an environmental monitoring module, which is used to monitor the underwater environmental temperature through a temperature sensor unit, monitor the underwater environmental pressure through a pressure sensor unit, and monitor the underwater environmental salinity through an inductive conductivity sensor unit, obtaining underwater environmental monitoring data; a communication module, which is used to communicate the underwater physical sign state quantity and the underwater environmental monitoring data with a wearable display terminal and a water surface monitoring device through an underwater cable or a wireless communication method.

2. The wearable underwater vital sign and environmental monitoring system according to claim 1, characterized in that, The steps for obtaining the physiological signal sequence are as follows: Based on the multi-source physiological signals, perform time series alignment, align each signal according to the time stamp, and adjust the signal data of each signal through interpolation or delay according to the acquisition time point of each signal to obtain the multi-source physiological signals with time series alignment. According to the multi-source physiological signals with time series alignment, perform data synthesis, integrate all aligned signals to a unified time point through data fusion, and generate a physiological signal sequence.

3. The wearable underwater vital sign and environmental monitoring system according to claim 1, characterized in that, The steps for obtaining the set of characteristic components are as follows: according to the physiological signal sequence, a two-dimensional matrix is generated by arranging the signal values sampled at each time point, with the data at each time point as a row vector and the signal channel data as a column vector to obtain a signal trajectory matrix; according to the signal trajectory matrix, the characteristic score of signal decomposition is calculated, and the calculation formula is: where is the characteristic score, is the singular value of the signal trajectory matrix, is the corresponding left singular vector, is the number of blocks; according to the characteristic score, the matching components of the eigenvalue and eigenvector are extracted, and the block characteristic scores are aggregated to generate the corresponding signal eigenvalue and eigenvector set, thereby obtaining the set of characteristic components.

4. The wearable underwater vital sign and environment monitoring system according to claim 1, characterized in that The steps for obtaining the reconstructed signal group are as follows: According to the set of eigen-components, extract the eigenvectors and the corresponding singular values, rearrange the eigenvectors and the singular values in matrix form, and generate an eigenvector matrix. Based on the eigenvector matrix, calculate the energy ratio, and the calculation formula is: where is the energy ratio, is the -th component value of the reconstructed signal, is the -th component value in the eigenvector set, is the number of selected principal components, is the total number of eigenvectors; Weightedly combine the eigenvectors with energy ratios higher than the threshold with the corresponding singular values, and perform inverse singular value decomposition to generate a reconstructed signal group.

5. The wearable underwater vital sign and environmental monitoring system according to claim 1, characterized in that, The steps for obtaining the heart rate change characteristics are as follows: Based on the reconstructed signal group, each electrocardiogram signal channel is processed, and the peak point positions of the R waves are extracted through waveform analysis and threshold determination. The peak point positions are arranged in a time series manner to obtain an electrocardiogram R wave peak point position sequence; Based on the electrocardiogram R wave peak point position sequence, a non-linear fluctuation index is calculated, and the calculation formula is: where, is the non-linear fluctuation index of adjacent R-R intervals, is the time position of the th R wave peak point, and are the time positions of the previous and the previous two R wave peak points respectively, is a dimensionless constant for positive adjustment to avoid a zero denominator, is the total number of R waves; Based on the non-linear fluctuation index, combined with the time series analysis of the electrocardiogram signal, the heart rate change characteristics are generated by calculating the amplitude distribution characteristics of the R waves.

6. The wearable underwater vital sign and environmental monitoring system according to claim 1, wherein The steps for obtaining the heart rate variability index are as follows: According to the heart rate change characteristics, the heart rate change characteristics are transformed from the time domain to the frequency domain through Fourier transform, different frequency components and amplitude values are extracted to obtain the frequency domain heart rate characteristics; Based on the frequency domain heart rate characteristics, the proportion of the frequency band energy of the target frequency band is calculated, and the calculation formula is: Wherein, is the proportion of the frequency band energy, is the frequency at the power spectral density, and are the lower and upper limit frequencies of the target frequency band, and are the minimum and maximum frequencies of the entire frequency spectrum range; Combining the information of the main frequency component in the frequency domain, the proportion of the frequency band energy of multiple frequency bands is statistically analyzed and summarized into an index set to generate the heart rate variability index.

7. The wearable underwater vital sign and environment monitoring system according to claim 1, wherein The steps for obtaining the physical sign fluctuation parameter are as follows: Classify the heart rate variability index into two parts, namely time domain features and frequency domain features. Statistically analyze the time domain features including standard deviation and root mean square interval, and classify the frequency domain features including band energy ratio and dominant frequency to obtain the classified heart rate variability index. Based on the classified heart rate variability index, calculate the change trends and fluctuation ranges of each time domain feature and frequency domain feature, and statistically analyze the extreme values, ranges, and average change amounts between the indexes to obtain the change trend of the statistical indexes. Based on the change trend of the statistical indexes, calculate the physical sign fluctuation parameter, and the calculation formula is: Wherein, is the physical sign fluctuation parameter, is the change value of the th time domain feature, is the change value of the th frequency domain feature, is the proportion of the th frequency domain feature, is the total number of features.

8. The wearable underwater vital sign and environmental monitoring system according to claim 1, characterized in that The steps for obtaining the underwater physical sign state quantity are as follows: According to the physical sign fluctuation parameters, divide them into time domain characteristics and frequency domain characteristics according to the characteristic type, respectively extract each change value in the time domain characteristics and each energy value in the frequency domain characteristics, perform data normalization processing within a unified range, and recombine the normalized data to form a set of normalized physical sign fluctuation parameters. Based on the set of normalized physical sign fluctuation parameters, calculate the similarity between each pair of features through the correlation between features, represent the correlation relationship between features in matrix form, merge and recalculate the features to obtain a set of fused physical sign fluctuation parameters; according to the set of fused physical sign fluctuation parameters, assign weight values according to the classification criteria of time-domain features and frequency-domain features, and perform summation calculation on the features with assigned weights to generate an underwater physical sign state quantity.

Images