Dual-lead finger clip electrode bio-wave detection method and device

Abstract

The application discloses a double-lead finger clip electrode biological wave detection method and device, and belongs to the technical field of biomedical engineering, and comprises a mechanical contact terminal, an intelligent adaptive module and a display terminal; the mechanical contact terminal comprises left and right two finger clip electrodes and is used for receiving biological wave signals in real time; the intelligent adaptive module is used for pre-processing the received biological wave signals, dividing the biological wave signals into low, medium and high frequency bands to extract key features, calculating cross-correlation coefficients and baseline drift amplitudes, constructing a classification model, identifying a user age range and automatically adjusting mode parameters, checking contact stability through a fast resampling mechanism and a dynamic threshold, and performing impedance compensation; and the display terminal is used for decomposing biological wave signals, extracting multi-domain features, modeling through a space-time attention LSTM network, calculating a physiological age and a health risk index, and generating an intervention measure recommendation list.

Description

TECHNICAL FIELD

[0001] The application relates to a double-lead finger clip electrode biological wave detection method and device and belongs to the technical field of biomedical engineering. BACKGROUND

[0002] Bioelectric signals contain rich physiological and pathological information of the human body and have important application value in the fields of health monitoring, disease diagnosis and rehabilitation medicine. With the progress of semiconductor technology, sensor technology and signal processing algorithms, multi-lead bioelectric detection devices have been widely used in clinical and scientific research fields. For example, a common four-to-six-lead electrocardiograph can simultaneously record signals of multiple cardiac electrical activity sites, providing more comprehensive information for doctors to evaluate the function of the heart.

[0003] However, due to the physical structural defects and fixed mode dependence of traditional large-area electrodes and the rigid signal processing logic, the existing technology still faces multiple challenges. The large-area electrodes are fixed by pasting or binding, which is easily disturbed by factors such as movement, sweating and skin oil, resulting in poor contact, large signal noise, reduced data reliability, and the fixed electrode position requires professional assistance for installation, which is difficult to meet the daily convenient detection requirements, especially in long-term monitoring in dynamic or home scenarios. The operation is complex and lacks flexibility, and at the same time, the existing technology ignores the applicability to the population and physiological differences. Children have thin skin and high impedance, and traditional fixed gain amplification circuits are prone to signal saturation or reduced signal-to-noise ratio. Factors such as skin folds in obese people and skin relaxation in the elderly can cause insufficient electrode contact area or uneven pressure. High-sweating sports enthusiasts can make the anti-sweat performance of traditional electrodes ineffective. In addition, the lack of small-sized, high-precision double-lead detection devices makes it difficult to solve the problems of contact consistency caused by different finger thicknesses and uneven signal quality caused by skin impedance differences. Therefore, the application provides a double-lead finger clip electrode biological wave detection method and device. SUMMARY

[0004] In view of the deficiencies of the prior art, the purpose of the present application is to provide a double-lead finger clip electrode biological wave detection method and device, which adjusts the pressure self-adaptively through a composite elastic clamping structure to ensure electrode adhesion, identifies age switching detection parameters using a random forest algorithm, combines fast resampling and impedance compensation to ensure signal quality, and uses high-order signal processing to analyze the health status, thereby solving the problems of unstable contact and insufficient adaptation.

[0005] To achieve the above-mentioned purpose, the application provides the following technical solutions:

[0006] The double-lead finger clip electrode biological wave detection method comprises the following steps:

[0007] The user puts fingers into the clamping opening of the detection device, sets an adaptive adjustment method, applies a basic clamping force, monitors the elastic clamping force in real time and automatically adjusts; wherein the detection device comprises two left and right finger clamping electrodes clamped on the user's left and right finger pads;

[0008] An alternating current excitation signal is injected to obtain double-lead bio-wave signals, a pattern recognition method is set, the contact impedance is analyzed in the frequency domain, the impedance modulus and phase characteristics are extracted, and different frequency bands are divided, the impedance feature vector is constructed in combination with the cross-correlation between the double-lead finger clamping electrodes and the baseline drift amplitude, and the user's age range is identified in real time, and the analog front-end parameters are automatically switched according to the identification result;

[0009] A fast resampling mechanism is triggered to update the bio-wave signals, an impedance compensation method is set, and the difference of the double-lead impedance is checked to determine the contact stability, and the impedance is independently compensated or linearly compensated;

[0010] After completing the impedance compensation, the bio-wave signals are reacquired, an evaluation method is set, the bio-wave signals are adaptively decomposed, early abnormality warning is performed, and a three-dimensional coordinate system is constructed to calculate the physiological age and health risk index, and a list of intervention measures is generated.

[0011] Specifically, the pattern recognition method comprises:

[0012] An alternating current signal with a frequency range of [f low ,f high ] and an amplitude of I0 is injected to the double-lead electrodes;

[0013] The upper limit of the excitation signal amplitude is set to I max , and once it exceeds the upper limit of the excitation signal amplitude, I0 is automatically adjusted to I max ;

[0014] The bio-wave signals of the double electrodes are synchronously collected at a preset sampling rate, and after anti-aliasing filtering, fast Fourier transform is performed to calculate the impedance modulus and phase angle;

[0015] According to the frequency range [f low ,f high ], low, medium and high frequency bands are divided, and features in different frequency bands are extracted, the mean and standard deviation of the impedance modulus are calculated in the low frequency band, the median of the phase angle is extracted in the medium frequency band, the change rate and coefficient of variation of the impedance modulus are calculated in the high frequency band, and the least squares method is used to linearly fit the impedance modulus in the high frequency band to obtain the fitting slope;

[0016] In the full frequency band range, the cross-correlation coefficient of the left and right electrodes is calculated, and a sliding window is set to obtain the baseline drift amplitude by calculating the signal mean value in each window;

[0017] Linear regression is performed on the mean values of the continuous multiple windows to obtain the slope k of the regression straight line and the intercept b drift . drift If |k drift |>δ drift , it is determined that there is baseline drift, the time points of the start and end of the drift are recorded, and the baseline correction is performed by using the polynomial fitting method; otherwise, it is determined that there is no baseline drift; wherein, δ drift is a preset drift slope threshold value.

[0018] The calculated features are integrated to generate real-time impedance feature vectors, including frequency band features, cross-correlation coefficients, and baseline drift information.

[0019] Specifically, the pattern recognition method further comprises:

[0020] Obtaining test samples and dividing them by age, extracting features of each test sample, associating them with corresponding age labels, and generating a multi-dimensional feature database;

[0021] Calculating the contribution of each feature to age classification by Gini coefficient, screening out key features with a contribution higher than a preset contribution threshold, and generating a key feature library;

[0022] Based on a random forest classifier, the number of trees, the maximum number of features, and the minimum number of leaf node samples are configured, the random forest classifier is trained using the key feature library and age labels, branches with a contribution lower than the contribution threshold are removed through pruning operation, model parameters are compressed using bit quantization technology, and a classification model is output.

[0023] Specifically, the pattern recognition method further comprises:

[0024] The impedance feature vector is input into the classification model, the probability distribution of each age group is output, and the maximum value max(P) in the probability distribution is found out;

[0025] The probability threshold is set to δ P , and the classification result is determined; if max(P)≥δ P , the maximum value corresponds to the classification result; if max(P)<δ P , it is determined that the classification is uncertain, and a signal fast resampling mechanism is triggered.

[0026] According to the classification result, the corresponding mode parameters are called from the analog front-end parameter table, and the corresponding detection mode is automatically switched.

[0027] Specifically, the impedance compensation method comprises:

[0028] Start the fast resampling mechanism to obtain the dual-lead bio-wave signal under the newly added benefit and filtering parameters;

[0029] Calculate the impedance values of left and right leads to obtain the mean value Z of dual-lead impedance avg , impedance difference value ΔZ and pressure difference value ΔF.

[0030] Set the difference threshold as δ ΔZ , and the pressure uniformity threshold as δ ΔF . Compare the calculated impedance difference value and pressure difference value with the threshold value, wherein the threshold value is dynamically generated by the LSTM model trained by user historical detection data.

[0031] If ΔZ>δ ΔZ and ΔF>δ ΔF , it is determined that the contact is abnormal, and the abnormal handling process is triggered until the contact stable condition is met.

[0032] If ΔZ>δ ΔZ and ΔF≤δ ΔF , it is determined that it is a physiological difference, and independent compensation is enabled, and the left and right electrode compensation currents are calculated.

[0033] If ΔZ≤δ ΔZ , it is determined that the contact is stable, and the impedance compensation process is entered.

[0034] Specifically, the specific steps of the impedance compensation process include:

[0035] According to the current amplification gain and ADC chip parameters, the best amplitude interval [V min , V max ] is obtained.

[0036] Under the constraint condition that the signal amplitude is in the best amplitude interval, the compensation current I * is calculated to minimize the deviation of the compensation current and the reference current.

[0037] A humidity-impedance correction curve is established. Once the environmental humidity exceeds the humidity threshold, a correction coefficient is calculated based on the environmental humidity and the humidity threshold, and the compensation current is updated.

[0038] The electrode temperature T is monitored in real time, and the skin impedance measurement value is corrected.

[0039] Inject a standard signal to verify the amplitude error after compensation. When the error exceeds the preset limit, trigger a secondary compensation cycle. If the adjustment is still not satisfactory after multiple adjustments, automatically roll back to the default parameters and prompt to check the contact.

[0040] Specifically, the adaptive adjustment method includes:

[0041] Initialize the pressure on the detection device, so that the finger electrode is in a ready-to-grip state. After the user places his finger in the electrode clamping port, the basic clamping force F0 is applied;

[0042] Obtain the normal pressure F and the deformation amount Δd of the contact sheet caused by the pressure, and calculate the contact area S;

[0043] Set the optimal contact area interval as [S opt.min ,S opt.max ] to determine whether to perform adaptive adjustment of the pressure;

[0044] If S opt.min ≤S≤S opt.max , the contact area does not exceed the optimal contact area interval, and the original pressure is maintained;

[0045] If S<S opt.min , calculate the pressure adjustment amount ΔF PID through the PID controller, and drive the spring to increase the clamping force by a step size of F' based on the pressure adjustment amount until the contact area meets the standard;

[0046] If S>S opt.max , the spring is driven in the opposite direction to relax, and the adjustment step size is limited to F' each time;

[0047] During the adjustment process, if the pressure F≥F TH and lasts for more than a preset period of time, it is determined that there is a risk of finger artery compression, triggering a two-stage protection mechanism; wherein F TH is the clamping warning threshold.

[0048] Specifically, the evaluation method comprises:

[0049] Obtain the biological wave signal after impedance compensation, and perform mirror extension and empirical mode decomposition to generate multiple intrinsic mode functions;

[0050] Calculate the mutual information value of each intrinsic mode function and the original signal, filter the effective intrinsic mode function, and generate an enhanced time-frequency energy matrix using Hilbert transform and wavelet interpolation;

[0051] Extract the time domain, frequency domain, and nonlinear features of the biological wave signal, and fuse environmental data to construct a fusion feature vector through a self-attention mechanism;

[0052] Obtain the fusion feature sequence of the latest multiple detections, analyze using a spatiotemporal attention LSTM network, capture time dependence and dual-lead collaborative features, and combine reconstruction error analysis and phase-locked value detection to identify synchronous mutation points, thereby achieving early abnormal warning;

[0053] Construct a three-dimensional visualization map to support gesture interaction and feature point labeling;

[0054] The physiological age and health risk index are calculated based on multi-feature weighted calculation, and preset schemes are called according to risk levels to generate personalized intervention measures.

[0055] The double-lead finger clip electrode bio-wave detection device comprises a mechanical contact terminal, an intelligent adaptive module and a display terminal.

[0056] The mechanical contact terminal comprises left and right finger clip electrodes for clamping the left and right finger pads of a user, and is used for real-time monitoring and self-adaptive adjustment of clamping force, output of alternating current excitation signals through the double-lead finger clip electrodes after pressure adjustment, and real-time reception of bio-wave signals.

[0057] The intelligent adaptive module is used for pre-processing the received bio-wave signals, dividing the bio-wave signals into low, medium and high frequency bands to extract key features, calculating cross-correlation coefficients and baseline drift amplitudes, constructing a classification model, identifying the age range of the user and automatically adjusting mode parameters, checking contact stability through a fast resampling mechanism and a dynamic threshold, and performing impedance compensation.

[0058] The display terminal is used for decomposing bio-wave signals, extracting multi-domain features and modeling through a spatiotemporal attention LSTM network, calculating physiological age and health risk index, and generating an intervention measure recommendation list.

[0059] The beneficial effects of the present application are:

[0060] Through the self-adaptive pressure adjustment mechanism, the composite elastic clamping structure and the multi-modal sensor are used to real-time monitor the pressure distribution and contact area, solving the problem of unstable contact caused by the dependence of traditional electrodes on adhesion, so that the electrodes can be closely attached to fingers of different thicknesses, ensuring the stability of signal acquisition from the physical level and avoiding noise interference caused by poor contact; through frequency domain analysis of double-lead impedance and modeling of the random forest algorithm, the simulation front-end parameters are dynamically switched according to the age range, combined with fast resampling and impedance compensation technology, solving the signal adaptation problem caused by physiological differences of different populations, ensuring that the excitation signal strength and impedance mean are linearly related, avoiding signal clipping caused by gain mismatch, improving the accuracy of cross-population detection, and using high-order signal processing methods such as Hilbert-Huang transform, fusing multi-domain features and environmental data, constructing a physiological state classification model and generating a visual report, realizing the leap from single signal acquisition to multi-dimensional health state analysis, providing personalized health suggestions for users, making up for the lack of traditional technology in health management, and significantly improving the stability, accuracy and practicality of bio-wave detection. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 It is a schematic diagram of a double-lead finger clip electrode bio-wave detection method.

[0062] Figure 2 It is a flowchart of the pattern recognition method of the present application.

[0063] Figure 3 Flow chart of the impedance compensation method of the present application;

[0064] Figure 4 Structure diagram of the dual-lead finger clip electrode biological wave detection device. DETAILED DESCRIPTION

[0065] The technical solutions of the present application will be described in detail below with reference to the drawings and specific embodiments. It should be understood that the embodiments of the present application and the specific features in the embodiments are detailed descriptions of the technical solutions of the present application, and are not limitations of the technical solutions of the present application. In the case of no conflict, the technical features in the embodiments of the present application and the embodiments can be combined with each other.

[0066] Embodiment 1:

[0067] Reference Figures 1 to 3 As shown in the figure, the embodiment introduces a dual-lead finger clip electrode biological wave detection method, including the following steps:

[0068] The user puts his finger into the clamping opening of the detection device, sets the adaptive adjustment method, applies a basic clamping force to the skin surface of the fingertip through the elastic clamping force of the finger clip, monitors the elastic clamping force in real time and automatically adjusts, so that the electrode is tightly attached to the skin of the finger pulp of different thickness of the finger, solves the unstable contact problem of the traditional large-area electrode relying on pasting from the physical layer, and avoids the blood flow being blocked due to being too tight or the poor contact due to being too loose;

[0069] After the physical contact is stable, the detection device injects an alternating current excitation signal into the dual electrodes through a constant current source, conducts through the human body tissue, and is recovered by the electrode and transmitted to the terminal such as a portable terminal or an independent host through a wire or wireless transmission mode. The two electrodes are connected with the mainboard through wires, forming a dual-lead loop of the left hand finger-right hand finger, synchronously collecting the biological wave signals of both hands at a specific sampling frequency, thereby generating dual-lead biological wave signals, including electrocardio, electrodermal electrical physiological signals, setting a pattern recognition method, due to the dynamic range difference of biological wave signals of different people, the contact impedance of the dual-lead finger clip electrode and the skin of the finger pulp is analyzed in frequency domain, the impedance modulus and phase characteristics in the preset range are extracted, and different frequency bands are divided. Combined with the cross-correlation between the dual-lead finger clip electrodes and the baseline drift amplitude, an impedance feature vector is constructed, a sample database including age and impedance characteristics is used, a random forest algorithm is used to establish a classification model including the mapping relationship between age and impedance characteristics, to identify the age classification result of the user in real time, and automatically switch the analog front-end parameters including the amplification gain and the high-pass filter cutoff frequency according to the classification result.

[0070] Since the impedance data measured at the original gain will lead to deviation in the calculation of the excitation current if directly used at the new gain, after the mode switching is completed, a fast resampling mechanism is triggered to obtain the real-time bio-wave signal at the new gain and the filtering parameters, an impedance compensation method is set, and the difference between the dual-lead impedance is checked to determine the contact stability. Once the contact is abnormal, a buzzer alarm is triggered and the user is prompted to adjust the clamping force to ensure that the bilateral signal compensation and the difference remain within the physiological range, avoiding the pollution of the overall analysis result by the distorted signal of the single side. After confirming the stable contact, the impedance is compensated independently or linearly. If the impedance compensation is performed first, such as adjusting the excitation strength of the constant current source, and then switching the mode, the compensated signal will exceed the upper limit of ADC quantization at high gain, causing clipping distortion.

[0071] After completing the impedance compensation, the bio-wave signal is reacquired, an evaluation method is set, the bio-wave signal is adaptively decomposed, the feature parameters are extracted, and the environmental data are fused to construct a fusion feature vector through a self-attention mechanism. Early abnormality warning is realized by using a spatiotemporal attention LSTM network. A three-dimensional coordinate system is constructed to visualize the signal, physiological age and health risk index are calculated, an intervention measure recommendation list is generated and encrypted for storage.

[0072] The detection device includes two left and right finger clamp electrodes, adopts a composite elastic clamping structure, and each electrode has a conductive material contact piece on the inner side. A variable clamping force is formed by a stainless steel main spring and a shape memory alloy auxiliary spring to clamp the user's left and right finger pads. A distributed micro piezoresistive sensor and a laser displacement sensor are integrated on the inner side of the finger clamp electrode clamping arm to monitor the normal pressure distribution and the deformation amount of the contact piece in real time at a preset sampling frequency.

[0073] Preferably, the adaptive adjustment method comprises:

[0074] The pressure on the detection device is initialized, at which time the finger clamp electrode is in a default clamping state. After the user places the left and right finger pads into the electrode clamp, the finger clamp spring automatically applies a basic clamping force F0 to lightly press the contact piece against the finger pad skin, avoiding discomfort caused by excessive initial pressure or poor contact caused by insufficient initial pressure.

[0075] At a preset acquisition frequency, the micro piezoresistive sensor array integrated on the inner side of the finger clamp electrode clamping arm acquires the pressure value of the finger pad skin in real time, extracts the pressure mean value as the normal pressure F, and synchronously acquires the deformation amount Δd of the contact piece caused by the pressure using the laser displacement sensor. The contact area S is estimated in real time through a skin elasticity mechanics model, and the expression is as follows:

[0076] S = S0 + k (F - F0) - β sign (F - F0) Δd crit )|F-F crit |

[0077] wherein S0 is the initial contact area under the basic clamping force, sign(·) is a sign extraction function, taking values including -1, 0, 1, F is the normal force, F is the blood flow warning pressure for judging whether the normal force is pressing the digital artery, β is the contact area attenuation coefficient in the overpressure state, used for correcting the contact area calculation when the normal force exceeds the blood flow warning pressure, to avoid blood flow blockage caused by excessive clamping force pressing the digital artery, and k is the skin elasticity coefficient, representing the contact area increment under unit pressure, determined through a three-in-one calibration experiment of normal force, deformation, and contact area, and Δd and the actual contact area under different normal forces F are recorded synchronously to establish a three-dimensional mapping relationship, and k is obtained by fitting a large number of historical samples; crit is the blood flow warning pressure for judging whether the normal force is pressing the digital artery, β is the contact area attenuation coefficient in the overpressure state, used for correcting the contact area calculation when the normal force exceeds the blood flow warning pressure, to avoid blood flow blockage caused by excessive clamping force pressing the digital artery, and k is the skin elasticity coefficient, representing the contact area increment under unit pressure, determined through a three-in-one calibration experiment of normal force, deformation, and contact area, and Δd and the actual contact area under different normal forces F are recorded synchronously to establish a three-dimensional mapping relationship, and k is obtained by fitting a large number of historical samples;

[0078] The optimal contact area interval [S opt.min ,S opt.max ] is set to judge whether to perform adaptive adjustment of the pressure; wherein S opt.min is the lower limit of the optimal contact area, and S opt.max is the upper limit of the optimal contact area;

[0079] If S opt.min ≤ S ≤ S opt.max , the contact area does not exceed the optimal contact area interval, and the pressure is not adjusted at this time, and the original pressure is maintained;

[0080] If S < S opt.min , the current pressure is insufficient, the pressure adjustment amount ΔF PID of the spring is calculated through the PID controller, the spring is driven to increase the clamping force by a step size F' based on the pressure adjustment amount, and the contact area is adjusted until it meets the standard; the expression is as follows:

[0081]

[0082] wherein K p , K i , and K d are PID adjustment amounts, set by a person skilled in the art, and in this embodiment, K p = 0.3, K i = 0.1, and K d = 0.05;

[0083] If S > S opt.max , the current pressure is too large, the spring is then driven to relax in the opposite direction, and the adjustment step size is also limited to F' each time to avoid signal interference caused by sudden changes in the pressure;

[0084] Meanwhile, if the pressure F ≥ F THAnd last more than 10 seconds, it is determined that there is a risk of digital artery compression, triggering a two-stage protection mechanism, first through the built-in vibration mechanism of the electrode, such as 100Hz low-frequency vibration, continuous vibration to prompt the user to "lightly adjust the finger position", at the same time the spring maintains the current pressure and does not increase, to carry out the first stage of protection, if the pressure is still not decreased, the spring is automatically released to the basic clamping force, and the detection is stopped and marked "abnormal contact pressure" on the terminal interface, suggesting checking the electrode clamping angle; wherein, F TH is the clamping warning threshold.

[0085] Preferably, the pattern recognition method comprises:

[0086] A high-precision programmable constant current source is used as an excitation signal generator, and a small amplitude alternating current signal with a frequency range of [f low ,f high ] and an amplitude of I0 is injected into the human body through double-lead electrodes, and at the same time, in order to avoid discomfort or potential risks caused by the excitation signal to the human body, an upper limit I max of the excitation signal amplitude is set, and if I0 calculated exceeds the upper limit, I0 is automatically adjusted to I max , and an adjustment log is recorded.

[0087] The biological wave signal V(t) of the double electrodes is synchronously collected at f s , the collected signal is first subjected to anti-aliasing filtering, and an 8th order Butterworth low-pass filter is used, with a cutoff frequency f cut According to the Nyquist criterion, f s is dynamically adjusted to f , which effectively filters out high-frequency noise and prevents aliasing from occurring, and the filtered signal is sent to a digital signal processor for fast Fourier transform, converting the time domain signal to a frequency domain complex representation V(f) = A(f) + jB(f), and then calculating the impedance modulus |Z(f)| and the phase angle θ(f); The expression is as follows:

[0088]

[0089] In the formula, V(f) is the biological wave frequency domain signal of the double electrodes, f represents the frequency, which is the independent variable of frequency domain analysis, A(f) is the real part of the complex number, B(f) is the imaginary part, both of which correspond to the components of the signal on the real axis and the imaginary axis at frequency f, and j is the imaginary unit;

[0090] Different physiological structures and tissues of human body show different frequency response characteristics in the process of electrical signal conduction. A single frequency band signal cannot fully reflect the diversity and complexity of human physiological state. By dividing different frequency bands for analysis, information related to specific physiological characteristics can be extracted, thereby more accurately identifying physiological differences of different age groups and improving the accuracy and effectiveness of biological wave signal analysis. Firstly, according to the frequency range [f low ,f high ], the low frequency band, the medium frequency band and the high frequency band are divided, so as to facilitate the subsequent extraction of frequency band features. The frequency range of the low frequency band is , the frequency range of the medium frequency band is , and the frequency range of the high frequency band is , and

[0091] In the low frequency band range, the impedance modulus value is calculated point by point, so as to obtain the mean value of the impedance modulus value and the standard deviation to reflect the permeability of subcutaneous tissue fluid. The content and distribution of subcutaneous tissue fluid of different age groups are different. For example, children have relatively high low-frequency impedance due to high water content, and the impedance modulus value of the elderly fluctuates greatly due to tissue relaxation. By analyzing the low-frequency impedance, the physiological changes related to age can be effectively captured, providing an important basis for age identification. At the same time, the sliding window technology is introduced, and the window size is set to W L frequency points, and the window is slid by a step S L , the mean value and the standard deviation of the impedance modulus value in each window are calculated to form a dynamic feature sequence of the low-frequency impedance. The expression is as follows:

[0092]

[0093]

[0094] In the formula, is the frequency point in the low frequency band, N L is the number of frequency points in the low frequency band;

[0095] In the medium frequency band range, the phase angle data is extracted and sorted, and different methods are used to calculate the median med(θ M), if the number of data is odd, the median is the phase angle value at the middle position after sorting, if it is even, the median is the average of the two middle phase angle values, which represents the capacitance characteristics of the cell membrane. The structure and function of cell membranes in different age groups are different, resulting in different capacitance effects. For example, the cell membrane capacitance characteristics of adults and children show obvious differences in this frequency band. The phase angle of adults is concentrated in a certain range, and the phase angle of children is larger due to the more significant capacitance effect of thin skin. Analysis of the phase angle in the medium frequency band helps to distinguish the physiological characteristics of people in different age groups;

[0096] In the high frequency range, the rate of change of the impedance modulus is calculated to evaluate the integrity of the stratum corneum. With age, the thickness of the stratum corneum changes. The stratum corneum of the elderly is thickened, so that the high-frequency impedance change is relatively flat. The stratum corneum of children is thin, and the high-frequency impedance change rate is larger. Therefore, the analysis of the high-frequency impedance change rate can be used as one of the important reference indicators for judging the age group. In order to analyze the trend of high-frequency impedance change more carefully, the least squares method is used to linearly fit the impedance modulus in the high-frequency range, and the slope k H and the intercept b H of the fitting straight line are obtained. The slope is used as a supplementary feature of the high-frequency impedance change rate. In addition, the coefficient of variation of the impedance modulus in the high-frequency range is calculated to reflect the relative fluctuation degree of the high-frequency impedance; the expression is as follows:

[0097]

[0098] In the formula, and are the standard deviation and mean of the high-frequency impedance modulus, respectively;

[0099] In the full frequency range, the impedance signals Z left (f) and Z right (f) of the left and right electrodes are calculated, and the cross-correlation coefficient p Z is calculated, which is used to reflect the symmetry of the left and right physiological signals; the expression is as follows:

[0100]

[0101] In the formula, Cov(Z left (f), Z right (f)) is the covariance of the left and right lead impedance signals, and are the standard deviations of the left and right lead impedance signals, respectively;

[0102] At the same time, in the full frequency range, the signal baseline drift amplitude s drift is calculated by the sliding window method, and the window size is set to W d sampling points with a step size of S dSliding window, within each window, calculate the mean μ of the signal. d Then, the standard deviation of the mean of all windows is calculated, which gives the baseline drift magnitude σ. drift This is to assess the stability of the electrode in contact with the skin;

[0103] To more accurately determine baseline drift, baseline drift trend analysis is introduced. Linear regression is performed on the mean of multiple consecutive windows to obtain the slope k of the regression line. drift and intercept b drift The direction and extent of baseline drift are determined by the sign and magnitude of the slope; if |k drift |>δ drift If the baseline drift is positive, a baseline shift is determined to exist, and the start and end times of the drift are recorded. To suppress the impact of baseline drift on signal feature extraction, a polynomial fitting method is used to perform baseline correction on the original signal. The order of the fitting polynomial is automatically adjusted according to the complexity of the signal. Otherwise, a baseline shift is determined to not exist. Wherein, δ drift The preset drift slope threshold;

[0104] The calculated values ​​are integrated to generate a real-time impedance characteristic vector, including frequency band characteristics, cross-correlation coefficient, baseline drift amplitude, baseline drift trend information, and baseline-corrected signal (if baseline drift exists). The baseline drift trend information includes slope, intercept, and drift time point.

[0105] A large number of test samples were obtained and divided into m groups according to age, with n samples in each group. At the same time, the feature vectors corresponding to each sample were obtained and associated with the corresponding age group labels to build a labeled multidimensional feature database.

[0106] Using the Gini coefficient built into the random forest algorithm, the contribution of each feature in the multidimensional feature database to the age group classification is calculated. A contribution threshold is set, and key features with a contribution higher than the contribution threshold are selected. Redundant features in the multidimensional feature database are removed to form an optimized key feature library.

[0107] A random forest classifier is selected, with n trees configured. trees To prevent overfitting, the maximum number of features is set to 1. The minimum number of leaf node samples is determined by automatically selecting the optimal feature combination. s.min To enhance the model's generalization ability, the model is trained based on the selected key feature library and age group labels; where q is the dimension of the key feature library.

[0108] The trained random forest model is simplified by using a cost-complexity pruning method. Branches with contributions below a preset value are removed by pruning operations, reducing the complexity and number of parameters of the model. 8-bit quantization technology is used to compress the trained model parameters and store them in the microcontroller's flash memory. At the same time, parallel computing technology is used to distribute the prediction tasks of multiple decision trees to different kernels or threads for execution, further improving the inference speed and outputting a lightweight classification model.

[0109] The impedance feature vector extracted in real time is input into the classification model. The model calculates the prediction results of multiple decision trees in parallel. Each decision tree outputs a category prediction (child, adult, elderly). Then, the prediction results of all decision trees are statistically analyzed by voting. The probability distribution [P1, P2, P3] of each age group is calculated, and the maximum value max(P) in the probability distribution is found.

[0110] Set the probability threshold to δ P The classification results are then evaluated; if max(P) ≥ δ P If the value of max(P) is less than δ, then the corresponding category will be output as the final classification result; if max(P) < δ P If the result is not clear, the classification is determined to be uncertain, triggering a fast resampling mechanism. During the resampling process, the user is first prompted by an indicator light and voice to keep their finger still. Then, the electrode pressure is readjusted, the signal is acquired, and the feature is extracted again. The maximum number of resampling times is r. If a definite classification result cannot be obtained after multiple resamplings, the message "Detection abnormal, please try again later" is output and the abnormality log is recorded.

[0111] To address the physiological signal characteristics of different age groups, analog front-end parameter tables were designed for three detection modes: children, adults, and the elderly. These tables cover amplification gain, high-pass filter cutoff frequency, and sampling rate. For all three modes, the notch switch is set to be on / off at f. on To eliminate power frequency interference, the analog front-end parameter tables for each mode are stored in the device's non-volatile memory. Each analog front-end parameter table contains parameter name, value, and check code. At the same time, to facilitate quick access, a parameter index table is established to record the storage address of each mode parameter in the memory, and the parameters in the memory are periodically checked. If a check error is detected, the latest parameter table is automatically downloaded from the cloud server for updating.

[0112] According to the age classification result output by the random forest model, the corresponding detection mode is automatically switched and the sampling rate is set. The FPGA controller sends the matched sampling rate parameter to the analog front-end chip through the SPI interface, and sends the mode switching instruction. Before switching the sampling rate, the data acquisition is stopped, and after the ADC chip completes the current sampling period, the sampling rate parameter is updated to ensure the continuity and integrity of data acquisition. After the switching is completed, the sampling rate is verified. The frequency error of the sampled signal is calculated by collecting the standard test signal. If the error is greater than 0.5%, the sampling rate setting and verification are performed again.

[0113] Preferably, the impedance compensation method comprises:

[0114] After receiving the mode switching completion interrupt signal of the analog front-end chip, the FPGA controller immediately starts the fast resampling mechanism to continuously collect the dual-lead bio-wave signal and obtain real-time signal data under the new gain and filtering parameters, providing a basis for subsequent analysis.

[0115] The resampled signal is processed, and the impedance values Z left and Z right of the left and right leads are calculated, respectively, and then the dual-lead impedance mean value Z avg , impedance difference value ΔZ and pressure difference value ΔF are obtained to quantify the inconsistency of the dual-lead impedance. The expressions are as follows:

[0116]

[0117] ΔZ = |Z left -Z rigth |

[0118] ΔF = |F left -F rigth |

[0119] In the formula, F left and F rigth are the clamping forces applied to the left and right finger electrodes, respectively.

[0120] The difference threshold is set to δ ΔZ , and the pressure uniformity threshold is set to δ ΔF . The calculated impedance difference value is compared with the difference threshold. If ΔZ > δ ΔZ and ΔF > δ ΔF , it is determined that the contact is abnormal, and the abnormal processing flow is triggered. The detection device generates low-frequency vibration through the electrode vibration module, and prompts the user to adjust the finger clamping angle in the form of red flashing on the terminal. At the same time, the current pressure sensor data and contact sheet deformation data are recorded to help the user more accurately locate the contact abnormal position until the contact stable condition is met again.

[0121] If ΔZ > δ ΔZ and ΔF≤δ ΔF , the impedance difference between two electrodes exceeds the threshold, but the pressure sensor data shows that the pressure is uniform, so it is determined that the physiological difference, such as the difference in the thickness of the stratum corneum on one side of the skin, and independent compensation is enabled to ensure the quality of the signals on both sides; the expression is as follows:

[0122] I * left = k I *Z left + I0

[0123]

[0124] In the formula, I * left , I * right are the compensation currents of the left and right electrodes respectively, is the compensation coefficient;

[0125] If ΔZ≤δ ΔZ , it is determined that the contact is stable, and the impedance compensation process is entered; wherein the threshold is dynamically generated by the LSTM model trained by the user's historical detection data;

[0126] According to the current amplification gain and the parameters of the ADC chip, the best amplitude interval [V min , V max ] is calculated to ensure that the amplitude of the signal after subsequent compensation is within the effective quantization range of the ADC, avoiding signal clipping or insufficient quantization;

[0127] Using the Lagrange multiplier method, under the constraint condition that the signal amplitude is in the best amplitude interval, the optimal compensation current I * is calculated to minimize the deviation of the compensation current from the reference current, and the expression is as follows:

[0128] I * = k I × Z avg + I0

[0129] In the formula, k I is the compensation coefficient;

[0130] The humidity sensor data is read in real time, and a humidity-impedance correction curve is established. Once the environmental humidity exceeds δ RH , the compensation current is increased to compensate for the decrease in skin impedance caused by sweat; the expression is as follows:

[0131] I * new = I * × (1+τ)

[0132]

[0133] wherein δ RH is the humidity threshold, I * new is the corrected compensation current, τ is the correction coefficient, and RH is the current ambient humidity;

[0134] Meanwhile, the skin impedance measurement value is corrected according to the NTC thermistor monitored electrode temperature T to eliminate the influence of temperature change on impedance measurement, and the expression is as follows:

[0135] Z corr = Z avg ×(1+α(T-T0))

[0136] wherein Z corr is the corrected impedance measurement value, α is the skin impedance temperature coefficient, and T0 is the basic electrode temperature;

[0137] The standard signal is injected to verify the amplitude error of the compensated signal, and if the error is greater than 5%, a secondary compensation cycle is triggered, with an adjustment step of 5% of the current each time to avoid over-adjustment, and a maximum of 3 cycles; if the signal amplitude error is still greater than 5% after multiple adjustments, it still does not meet the requirements, and automatically rolls back to the default parameters, and prompts the user to check the electrode contact at the terminal.

[0138] Preferably, the evaluation method comprises:

[0139] The biological wave signal after impedance compensation is obtained, the signal is extended by 20% in length at both ends by using mirror extension combined with local mean prediction algorithm to solve the traditional EMD endpoint effect problem, and empirical mode decomposition is performed to generate a plurality of intrinsic mode functions (IMFs), each of which corresponds to a physiological signal of different time scale and contains high-frequency components, medium-frequency components and low-frequency components;

[0140] The mutual information entropy criterion is introduced to calculate the mutual information value of each IMF and the original signal, and noise modes with mutual information less than a preset value are removed to select modes containing effective physiological signals. The selected IMFs are subjected to Hilbert transform, and the frequency resolution is increased from 5 Hz to 1 Hz through wavelet interpolation, and the time resolution reaches 5 ms to generate an enhanced time-frequency energy distribution matrix;

[0141] Time domain features, frequency domain features and nonlinear features are extracted from the biological wave signal, such as the standard deviation of R-R interval of electrocardiogram, the root mean square of adjacent interval difference, the baseline drift amplitude of skin electricity, the peak latency, the low-frequency power and high-frequency power in heart rate variability and their ratio, the energy proportion of each frequency band in impedance spectrum, sample entropy, Lyapunov exponent and fractal dimension, thereby generating a physiological feature vector;

[0142] Collect environmental data such as contact area of pressure sensor, motion intensity index of IMU, temperature and humidity sensor data, fuse physiological and environmental characteristics, calculate feature weight through self-attention mechanism, construct real-time fusion feature vector to highlight key physiological indicators;

[0143] Obtain the fusion feature sequence of several recent detections, construct a spatio-temporal attention LSTM network, in the time dimension, capture the time dependence of the feature sequence, such as time series features such as abnormal low frequency power and high frequency power ratio in consecutive multiple detections, in the spatial dimension, focus on the dual-lead cooperative features through the attention mechanism, such as left and right phase difference, and analyze through reconstruction error, when the Euclidean distance between the real-time fusion feature vector and the historical data exceeds the preset distance value, trigger early abnormal warning;

[0144] Calculate the instantaneous phase difference of each IMF of the left and right leads, quantify the synchronization through the phase-locked value, generate a PLV time-varying atlas, and combine wavelet transform to detect synchronization mutation points to identify abnormal conditions;

[0145] Build a three-dimensional coordinate system, x-axis represents time, y-axis represents normalized amplitude, and z-axis represents center frequency, encode frequency high and low through color gradient, visualize left and right lead signals as red / blue spiral lines, spiral spacing reflects phase difference, and line width maps signal energy, automatically identify and label feature event points such as ECG R-wave peak and skin electric response peak, support gesture operation to realize time axis scaling, long press to view detailed parameters of feature points, and thus generate a three-dimensional visualization atlas;

[0146] Based on impedance spectrum characteristics, HRV parameters, and skin electric response entropy values, set different weight coefficients, calculate physiological age, and calculate health risk index according to multiple indicators such as stress index, abnormal score, and phase-locked value, divide into low, medium, and high risk levels, and according to different health states and risk levels, call corresponding health suggestions and intervention schemes from a preset scheme library, such as recommending cervical spine stretching movements for fatigue state, integrate the above generated data, perform encryption operation using AES-256 encryption algorithm, store detection records using blockchain technology, ensure that the data meets the HIPAA data security standard, reduce the risk of data leakage, and thus generate an intervention measure recommendation list.

[0147] Embodiment 2:

[0148] Please refer to Figure 4 The present application provides another embodiment: a dual-lead finger clip electrode biological wave detection device, comprising: a mechanical contact terminal, an intelligent adaptation module, and a display terminal.

[0149] The mechanical contact terminal comprises two left and right finger clamping electrodes, each of which is provided with a conductive material contact piece on the inner side for clamping the left and right fingers of the user. Through the composite structure of the stainless steel main spring and the shape memory alloy auxiliary spring, the basic clamping force is applied to the skin surface of the fingertips, and the micro piezoresistive sensor and the laser displacement sensor are combined to monitor and self-adaptively adjust the clamping force in real time, so that the electrodes are tightly attached to fingers of different thicknesses, ensuring that the contact area is stable in the optimal interval. At the same time, a double protection mechanism is provided to avoid compression of the digital artery, solving the problems of unstable contact, complex operation and blood flow blockage of traditional electrodes. After pressure adjustment, the dual-lead finger clamping electrode outputs an alternating current excitation signal, and real-time biological wave signals are received;

[0150] The intelligent adaptive module is used for preprocessing the received biological wave signals, dividing them into low, medium and high frequency bands to extract key features, calculating the cross-correlation coefficient and baseline drift amplitude, realizing noise suppression and frequency band physiological feature analysis of wide frequency signals, improving signal analysis accuracy, and based on the random forest algorithm, an age classification model is constructed to identify the user's age range and automatically switch the analog front-end parameters. Through the fast resampling mechanism and dynamic threshold verification of contact stability, impedance linear compensation is realized by combining digital potentiometers and temperature and humidity compensation, solving the signal adaptation and distortion problems caused by physiological differences of different groups of people;

[0151] The display terminal is used for decomposing biological wave signals, extracting multi-domain features and modeling through a spatiotemporal attention LSTM network to generate a three-dimensional vector diagram to visualize signal differences, calculate physiological age and health risk index, and generate an intervention recommendation list to realize deep physiological state analysis and active health intervention.

[0152] In summary, the application discloses a dual-lead finger clip electrode biological wave detection method and device. A user places a finger in the finger clip electrode. The device applies a basic clamping force through a composite elastic structure. A micro piezoresistive sensor and a laser displacement sensor monitor the pressure distribution and contact sheet deformation in real time. The contact area is estimated based on a skin elasticity mechanics model. The clamping force is dynamically adjusted by a PID controller to ensure that the electrode closely adheres to fingers of different thicknesses and avoids blood flow compression. After physical contact is stable, a constant current source injects wideband alternating current to form a dual-lead loop. Biological wave signals are synchronously collected. After anti-aliasing filtering and FFT transformation, low, medium and high frequency bands are divided to extract impedance modulus, phase angle and other characteristics. An impedance feature vector is constructed, and the age range is identified by a random forest algorithm. The analog front-end parameters are automatically switched. After mode switching, fast resampling is triggered. The dual-lead impedance difference and pressure uniformity are verified. The contact state is determined, and abnormal prompts or independent compensation are implemented. After verification, the best amplitude interval is calculated based on the amplification gain and ADC parameters. Impedance linear compensation is realized by using the Lagrange multiplier method combined with temperature and humidity correction. After compensation is completed, Hilbert-Huang transformation is performed on the signal to extract multi-domain features and fuse environmental data. Abnormalities are monitored by a spatiotemporal attention LSTM. A three-dimensional vector diagram is constructed to visualize signal differences. Physiological age and health risk index are calculated. An encrypted report and personalized intervention suggestions are generated to complete the detection process.

[0153] The above only describes the preferred embodiments of the application, and the protection scope of the application is not limited to the above-mentioned embodiments. Any technical solutions falling within the concept of the application shall fall within the protection scope of the application. It should be noted that, for ordinary skilled persons in the art, some improvements and refinements without departing from the principles of the application shall also be considered as falling within the protection scope of the application.

Claims

1. A two-lead finger clip electrode biowave detection device, characterized by, The application relates to a dual-lead finger electrode bio-wave detection device. The mechanical contact terminal comprises left and right finger electrode for receiving bio-wave signals in real time. The intelligent adaptive module is used for preprocessing the received bio-wave signals, dividing the bio-wave signals into low, medium and high frequency bands to extract key features, calculating the cross-correlation coefficient and baseline drift amplitude, constructing a classification model, identifying the user age range and automatically adjusting the mode parameters, checking the contact stability through a fast resampling mechanism and a dynamic threshold, and performing impedance compensation. The display terminal is used for decomposing bio-wave signals, extracting multi-domain features and modeling through a space-time attention LSTM network, calculating physiological age and health risk index, and generating an intervention measure recommendation list.

2. The dual-lead finger electrode bio-wave detection device according to claim 1, wherein: the intelligent adaptive module is configured with a pattern recognition method; the pattern recognition method comprises: obtaining the bio-wave signals, performing anti-aliasing filtering, and then performing fast Fourier transform to calculate the impedance modulus and phase angle; dividing the frequency range into low, medium and high frequency bands, and extracting features in different frequency band ranges, calculating the mean and standard deviation of the impedance modulus in the low frequency band, extracting the median of the phase angle in the medium frequency band, calculating the change rate and coefficient of variation of the impedance modulus in the high frequency band, and using the least square method to linearly fit the impedance modulus in the high frequency band to obtain the fitting slope; and in the full frequency band range, the cross-correlation coefficient of the left and right electrodes is calculated, and a sliding window is set to calculate the signal mean value in each window to obtain the baseline drift amplitude.

3. The dual-lead finger electrode bio-wave detection device according to claim 2, wherein: the pattern recognition method further comprises: integrating the calculated features to generate a real-time impedance feature vector, including frequency band features, cross-correlation coefficients and baseline drift information.

4. The dual-lead finger electrode bio-wave detection device according to claim 3, wherein: the pattern recognition method further comprises: obtaining test samples and dividing them by age, extracting the features of each test sample, associating them with the corresponding age range label, and generating a multi-dimensional feature database; calculating the contribution of each feature to the age range classification through the Gini coefficient, screening out key features with a contribution higher than a preset contribution threshold, and generating a key feature library; based on a random forest classifier, configuring the number of trees, the maximum number of features and the minimum number of leaf node samples, training the random forest classifier using the key feature library and the age range label, removing branches with a contribution lower than the contribution threshold through pruning operation, compressing the model parameters using bit quantization technology, and outputting the classification model.

5. The dual-lead finger electrode bio-wave detection device according to claim 4, wherein: the pattern recognition method further comprises: calling corresponding mode parameters from an analog front-end parameter table according to the classification result, and automatically switching the corresponding detection mode.

6. The dual-lead finger electrode bio-wave detection device according to claim 5, wherein: the intelligent adaptive module is configured with an impedance compensation method. ​ ​ ​ ​ Linear regression is performed on the mean values of the continuous windows to obtain the slope and intercept of the regression line to determine whether there is baseline drift; if , it is determined that there is baseline drift, the time points of the start and end of the drift are recorded, and a polynomial fitting method is used for baseline correction; otherwise, it is determined that there is no baseline drift; wherein, is a preset drift slope threshold value. ​​ ​ ​ ​ ​ ​ ​ ​ ​ inputting the impedance feature vector into the classification model, outputting a probability distribution for each age group, and finding a maximum value in the probability distribution ; Set the probability threshold to The classification results are then judged; if If the maximum value corresponds to a category, then the category will be output as the classification result; if If the classification is uncertain, the signal fast resampling mechanism is triggered. ​ ​ ​ The impedance compensation method comprises: Start the fast resampling mechanism to obtain the dual-lead bio-wave signal under the new benefit and filter parameters; The impedance values of left and right leads are calculated to obtain the mean value of the dual-lead impedance , impedance difference value , and pressure difference value ; The difference threshold is set as The pressure uniformity threshold is set as The calculated impedance difference value and the pressure difference value are compared with the threshold value; wherein the threshold value is dynamically generated by an LSTM model trained by user historical detection data; If and then determine that the contact is abnormal, trigger an abnormal handling process, and until the contact stable condition is met again. If and then determine physiological difference, enable independent compensation, calculate left and right electrode compensation current; If then the contact is determined to be stable and the impedance compensation procedure is entered.

7. The dual-lead finger clip electrode bio-wave detection device according to claim 6, characterized in that: An adaptive adjustment method is configured in the mechanical contact terminal; The adaptive adjustment method comprises: initializing the pressure on the detection device, so that the finger electrode is in a standby clamping state, and the user places the finger into the electrode clamping port and applies a basic clamping force ; acquiring normal pressure and the deformation amount of the contact sheet due to the pressure , calculating the contact area ; The optimal contact area interval is set as to determine whether to perform adaptive adjustment of pressure; If , the contact area does not exceed the optimal contact area interval, the original pressure is maintained; like The pressure regulation is calculated using a PID controller. Based on the pressure adjustment amount, the spring is driven to... Increase the clamping force by increasing the step size until the contact area reaches the target; If , the back drive spring relaxes, and the limit of each adjustment step is ; In the adjustment process, if the pressure and continues to exceed the preset period, it is determined that there is a risk of digital artery compression, triggering a two-stage protection mechanism; wherein, is the clamping warning threshold.

8. The dual-lead finger clip electrode bio-wave detection device according to claim 7, characterized in that: An evaluation method is configured in the display terminal; The evaluation method comprises: Obtain the bio-wave signal after impedance compensation, and perform mirror continuation and empirical mode decomposition to generate a plurality of intrinsic mode functions; Calculate the mutual information value of each intrinsic mode function and the original signal, screen the effective intrinsic mode function, generate an enhanced time-frequency energy matrix by using Hilbert transform and wavelet interpolation; Extract the time domain, frequency domain and nonlinear features of the bio-wave signal, fuse the environmental data, and construct a fusion feature vector through a self-attention mechanism.

9. The dual-lead finger clip electrode bio-wave detection device according to claim 8, characterized in that: An evaluation method is configured in the display terminal; The evaluation method comprises: Obtain the fusion feature sequence of the latest multiple detections, analyze by using a spatiotemporal attention LSTM network, capture the time dependence and dual-lead collaborative features, combine the reconstruction error analysis and phase-locked value detection to detect the synchronism mutation point, so as to realize early abnormal warning; Construct a three-dimensional visualization atlas to support gesture interaction and feature point labeling; Based on multi-feature weighted calculation, physiological age and health risk index are calculated, and according to the risk level, a preset scheme is called to generate individualized intervention measures.

Images