An intelligent distress decision method fusing multi-modal perception and life green card

Abstract

The present application relates to the field of healthcare informatics, and discloses an intelligent help-seeking decision-making method fusing multi-modal perception and life green card, comprising: acquiring physiological and environmental perception data, extracting pulse wave feature vector, signal-to-noise ratio, combined acceleration variance and attitude angle mutation rate; calling life green card data set, calculating physiological Mahalanobis residual of feature vector in deviation space; calculating perception quality factor according to combined acceleration variance and signal-to-noise ratio, determining normalized correction coefficient by comparing quality threshold, correcting physiological Mahalanobis residual and generating physiological abnormality index; solving comprehensive help-seeking decision-making index by combining disease risk weight factor and environmental risk coefficient, comparing discrete gradient threshold, and outputting help-seeking control instruction; the present application eliminates false alarm caused by baseline difference by constructing individual deviation space, blocks false image pollution by cooperating with quality threshold, and improves the accuracy and reliability of help-seeking decision-making.

Description

TECHNICAL FIELD

[0001] The application relates to an intelligent help-seeking decision method fusing multi-modal perception and a life green card, and belongs to the technical field of medical care informatics. BACKGROUND

[0002] Real-time monitoring and automatic help-seeking of pre-hospital sudden diseases are common applications in emergency medical information systems. The existing technology usually collects multiple physiological data such as body surface photoplethysmographic pulse wave signals, electrocardiogram signals and body surface acceleration signals through wearable sensing devices, and then compares the collected results with fixed physiological index thresholds or performs weighted summation on the detection results of multiple sensors to determine whether to send an emergency help-seeking signal. In daily activities, especially when the monitored subjects are older or have chronic diseases, the above method is prone to judgment deviation. The physiological indicators of some elderly people and chronic disease patients may deviate from the average level of the general population for a long time even in a resting state. Using a unified fixed threshold is prone to frequent false positives. If the threshold is relaxed to reduce false positives, it may not be able to timely detect abnormalities and miss the emergency opportunity.

[0003] On the other hand, human activities can cause motion artifacts in the body surface photoplethysmographic pulse wave signals, and fluctuations in the contact impedance of the sensor can also cause noise interference. The existing method usually uses these disturbed signals together with normal signals for judgment, and lacks a processing mechanism for dynamic weight reduction and signal recovery according to signal quality, so it is prone to false help-seeking. To alleviate the above effects, the existing technology usually uses methods such as relaxing the judgment threshold, smoothing the signal, or increasing the sampling channel and data transmission frequency. However, smoothing filtering may weaken the transient characteristics of sudden physiological abnormalities while suppressing motion artifacts. Increasing the sampling channel and transmission frequency will increase the power consumption of the wearable terminal and increase the data transmission delay. These measures still cannot effectively handle the problems of individual physiological baseline differences, sensor noise interference and lack of utilization of clinical information such as medical history. For example, the Chinese invention patent application with publication number CN120340197A discloses a wearable device alarm method and system. However, this scheme has the following limitations under actual dynamic working conditions: Its attribute correction mechanism relies on static sample statistics and discrete intersection ratio calculation, and it is difficult to effectively respond to the nonlinear changes of multi-dimensional environment and physiological dynamics coupling, leading to interval correction lag when facing sudden composite interference; its probability predictor lacks adaptive calibration capability for dynamic boundaries when dealing with individual small physiological fluctuations, making it difficult to ensure the real-time and accuracy of alarm judgment in non-ideal environments, and making it difficult to ensure the precise scheduling of emergency rescue resources.

[0004] Therefore, how to establish a multi-dimensional probability distribution model according to individual medical history characteristics, and combine sensor signal quality judgment and environmental risk analysis to realize accurate and reliable hierarchical help-seeking decision under the condition of low power consumption is still a problem to be solved in the prior art. SUMMARY

[0005] To solve the problems in the background art, the technical scheme of the present application is as follows: an intelligent help-seeking decision method fusing multi-modal perception and life green card, comprising the following steps: Step S1, acquiring the body surface photoplethysmogram signal, body surface acceleration signal, environmental air pressure data and environmental temperature data of the monitored subject, extracting the time domain feature vector of the body surface photoplethysmogram signal, calculating the signal-to-noise ratio of the body surface photoplethysmogram signal, and extracting the vector sum acceleration variance and attitude angle mutation rate of the body surface acceleration signal; Step S2, calling the life green card data set, which contains the basic physiological feature mean vector, covariance matrix inverse matrix and disease risk weight factor; mapping the time domain feature vector to the deviation space of the life green card data set to generate a physiological Mahalanobis residual; Step S3, calculating a perception quality factor according to the vector sum acceleration variance and the signal-to-noise ratio of the body surface photoplethysmogram signal; if the perception quality factor is lower than a quality threshold, truncating the normalization correction coefficient to a low confidence lower limit value, otherwise mapping it to a normalization correction coefficient; multiplying the physiological Mahalanobis residual by the normalization correction coefficient to generate a physiological abnormality index; Step S4, calculating the product of the disease risk weight factor and the physiological abnormality index and the sum of the product of the preset environmental synergy coefficient and the environmental risk coefficient according to the environmental risk coefficient constructed based on the disease risk weight factor, the attitude angle mutation rate and the time variation rate based on the environmental air pressure data, generating a comprehensive help-seeking decision index, comparing with a discrete gradient threshold, and outputting a help-seeking control instruction.

[0006] Preferably, step S3 comprises the following sub-steps: step S31, when the perception quality factor is lower than the quality threshold, truncating the normalization correction coefficient to a low confidence lower limit value to trigger signal channel self-healing compensation; step S32, updating the tap coefficient of the adaptive filter according to the frequency domain distribution of the body surface acceleration signal to reconstruct the noisy baseline of the body surface photoplethysmogram signal, generate a denoised photoplethysmogram signal, recalculate the signal-to-noise ratio based on the denoised photoplethysmogram signal, and update the perception quality factor.

[0007] Preferably, the step S1 comprises the following sub-steps: a step S11 of setting a sliding time window, extracting heart rate variability high frequency power and blood oxygen saturation extreme value of the body surface photoplethysmogram signal in the sliding time window, and combining to construct a time domain feature vector; and a step S12 of calculating a vector sum of the body surface acceleration signal in three-dimensional axial directions, solving a variance of the vector sum in the sliding time window to generate a vector sum acceleration variance, and calculating a posture angle mutation rate according to a time domain variation slope of the body surface acceleration signal.

[0008] Preferably, the step S2 comprises the following sub-steps: a step S21 of retrieving a life green card data set in the storage unit according to an identity of the monitored subject, the life green card data set further containing a drug contraindication parameter of the monitored subject; and a step S22 of constructing a Mahalanobis distance kernel based on an individual physiological state, calculating a deviation vector by using the Mahalanobis distance kernel to subtract the time domain feature vector from a basic physiological feature mean vector, performing a quadratic form matrix multiplication on the deviation vector and a covariance matrix inverse matrix, and generating a physiological Mahalanobis residual in a dimensionless scalar form by taking a square root of a product result, so as to eliminate a correlation interference between the multi-dimensional physiological parameters.

[0009] Preferably, the step S4 comprises the following environment adaptive adjustment steps: a step S41 of retrieving an environment sensitive trigger matrix in the life green card data set; and a step S42 of monitoring a time variation rate of environment pressure data and environment temperature data, calculating an environment abnormal offset vector according to the measured environment pressure data and environment temperature data, activating an environment trigger gate when an absolute value of the time variation rate of the environment pressure data exceeds 0.05 kPa / s or the environment temperature data is lower than 0°C, calculating a transpose product of the environment sensitive trigger matrix and the environment abnormal offset vector to generate an environment correction increment, and generating a corrected environment synergy coefficient by performing a proportional multiplication correction on the preset environment synergy coefficient using the environment correction increment, and calculating the integrated distress decision index based on the corrected environment synergy coefficient.

[0010] Preferably, the step S4 comprises the following hierarchical discrimination steps: a step S43 of comparing the integrated distress decision index with a first discrete threshold and a second discrete threshold, wherein the second discrete threshold is greater than the first discrete threshold; and a step S44 of outputting a maintain rest monitoring instruction when the integrated distress decision index is less than the first discrete threshold, outputting a first-level early warning instruction when the integrated distress decision index is greater than or equal to the first discrete threshold and less than the second discrete threshold, and starting a local interactive confirmation program, and outputting a second-level emergency distress instruction when the integrated distress decision index is greater than or equal to the second discrete threshold.

[0011] Preferably, the output of the secondary emergency call instruction in step S44 also includes the following data packaging step: step S441, extract the medical history summary and drug contraindication parameters in the life green card data set, combine the real-time obtained geographic coordinate data, physiological Mahalanobis residual and time domain feature vector to generate a simplified first aid call data package; step S442, implement Huffman bit-level lossless compression on the simplified first aid call data package, and send the compressed simplified first aid call data package to the emergency command center through a wireless communication link.

[0012] Preferably, the output of the primary warning instruction in step S44 also includes the following interactive interception step: step S443, trigger the local sound-light interactive terminal to start the countdown reminder, set the countdown length according to the perception quality factor, and lengthen the countdown length when the perception quality factor decreases; step S444, if the confirmation false alarm signal input by the monitored subject through touch buttons is captured within the countdown length, cancel the primary warning instruction, reset the physiological abnormality index to zero, and terminate the output of the call control instruction; if the countdown length ends and the confirmation false alarm signal is not captured, automatically upgrade to output the secondary emergency call instruction.

[0013] Preferably, when the compressed simplified first aid call data package is sent to the emergency command center in step S442, the following multi-channel redundant transmission step is included: step S4421, detect the link connection state of the cellular mobile communication network, narrowband Internet of Things and near-field Mesh ad hoc network; step S4422, select the connected link according to the priority order of the cellular mobile communication network, narrowband Internet of Things and near-field Mesh ad hoc network; if the high-priority network link is disconnected, switch to the next high-priority network link to resend the simplified first aid call data package, ensuring the successful transmission of the simplified first aid call data package.

[0014] Compared with the prior art, the beneficial effects of the present application are: 1. In the intelligent call decision fusion of multi-modal perception and life green card, the mean vector and covariance matrix inverse matrix of the basic physiological features in the life green card data set are used, and the historical physiological state of the monitored subject is used as the judgment basis, and the deviation degree between the current physiological state and the individual baseline is reflected through the physiological Mahalanobis residual. Compared with the judgment method using a unified fixed threshold, this method can reduce the false alarm of the elderly or chronic disease patients due to different normal physiological baselines, while retaining the ability to identify early slight abnormalities, which is beneficial to reduce unnecessary emergency calls and reduce the risk of not discovering abnormal situations in time.

[0015] 2. The degree of influence of motion artifacts on the body surface photoplethysmographic pulse wave signal is determined by the quality factor, and the normalized correction coefficient is adjusted according to the quality threshold, which can reduce the influence of low-quality physiological signals in the help-seeking judgment. In the case of unstable sensor contact, intense human activity, etc., this processing can avoid misjudging noise as real physiological abnormalities; after the signal quality is restored, the effective signal can be used for judgment again, thereby improving the stability and reliability of the help-seeking decision in complex wearing state.

[0016] 3. The disease risk weight factor, physiological abnormality index, posture angle mutation rate and environmental risk coefficient are comprehensively considered, and different levels of help-seeking control instructions are output according to the comprehensive help-seeking decision index, which can make the help-seeking judgment not only reflect the current physiological abnormality degree, but also consider the risk of the monitored subject's medical history, body position change and external environment influence, thereby distinguishing between general abnormalities and emergency situations, reducing the deviation caused by judging only one physiological data, making the output warning or help-seeking instruction more consistent with the actual risk degree of the monitored subject, and providing more targeted information for pre-hospital emergency. BRIEF DESCRIPTION OF DRAWINGS

[0017] Fig. 1 The main flow chart of the intelligent help-seeking decision method of the multi-modal perception of the application; Fig. 2 The threshold comparison and hierarchical response diagram of the comprehensive help-seeking decision index of the application.

[0018] The purpose realization, functional characteristics and advantages of the application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the application will be clearly described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are part of the embodiments of the application, rather than all the embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments in the application belong to the scope of protection of the application.

[0020] An intelligent help-seeking decision method fusing multi-modal perception and life green card, comprising the following steps: Step S1, acquiring the body surface photoplethysmographic pulse wave signal, body surface acceleration signal, environmental air pressure data and environmental temperature data of the monitored subject, extracting the time domain feature vector of the body surface photoplethysmographic pulse wave signal, calculating the signal-to-noise ratio of the body surface photoplethysmographic pulse wave signal, and extracting the vector sum acceleration variance and posture angle mutation rate of the body surface acceleration signal; Step S2, call the life green card data set, the life green card data set contains the mean vector of the basic physiological characteristics, the inverse matrix of the covariance matrix and the disease risk weight factor; map the time domain feature vector to the deviation degree space of the life green card data set to generate physiological Mahalanobis residual; Step S3, calculate the perceptual quality factor according to the vector sum acceleration variance and the signal-to-noise ratio of the body surface photoplethysmogram signal; if the perceptual quality factor is lower than the quality threshold, truncate the normalization correction coefficient to the low confidence lower limit value, otherwise map it to the normalization correction coefficient; multiply the physiological Mahalanobis residual by the normalization correction coefficient to generate a physiological abnormality index; Step S4, calculate the product of the disease risk weight factor and the physiological abnormality index and the sum of the product of the preset environmental synergy coefficient and the environmental risk coefficient based on the time variation rate of the environmental air pressure data, generate a comprehensive help decision index, compare with the discrete gradient threshold, and output a help control instruction.

[0021] Preferably, step S3 includes the following sub-steps: step S31, when the perceptual quality factor is lower than the quality threshold, truncate the normalization correction coefficient to the low confidence lower limit value, trigger the signal channel self-healing compensation; step S32, update the tap coefficient of the adaptive filter according to the frequency domain distribution of the body surface acceleration signal, reconstruct the noisy baseline of the body surface photoplethysmogram signal, generate a denoised photoplethysmogram signal, and recalculate the signal-to-noise ratio based on the denoised photoplethysmogram signal to update the perceptual quality factor.

[0022] Preferably, step S1 includes the following sub-steps: step S11, set a sliding time window, extract the heart rate variability high frequency power and the blood oxygen saturation extreme value of the body surface photoplethysmogram signal within the sliding time window, and combine to construct a time domain feature vector; step S12, calculate the vector sum of the body surface acceleration signal in three-dimensional axial direction, solve the variance of the vector sum in the sliding time window to generate a vector sum acceleration variance, and calculate the posture angle mutation rate according to the time domain variation slope of the body surface acceleration signal.

[0023] Preferably, step S2 includes the following sub-steps: step S21, call the life green card data set in the storage unit according to the identity of the monitored subject, the life green card data set also contains the drug contraindication parameters of the monitored subject; step S22, construct a Mahalanobis distance kernel based on individual physiological state, calculate the difference between the time domain feature vector and the mean vector of the basic physiological characteristics by using the Mahalanobis distance kernel to obtain a deviation vector, perform quadratic matrix multiplication on the deviation vector and the inverse matrix of the covariance matrix, take the square root of the product result to generate a physiological Mahalanobis residual in the form of dimensionless scalar, and eliminate the correlation interference between multiple physiological parameters.

[0024] Preferably, the step S4 comprises the following environment adaptive adjustment steps: a step S41 of calling the environment sensitive trigger matrix in the life green card dataset; a step S42 of monitoring the time rate of change of the environment pressure data and the environment temperature data, and calculating an environment abnormal offset vector according to the measured environment pressure data and the environment temperature data; when the absolute value of the time rate of change of the environment pressure data exceeds 0.05 kPa / s or the environment temperature data is lower than 0°C, an environment trigger gate is activated, the transpose product of the environment sensitive trigger matrix and the environment abnormal offset vector is calculated, an environment correction increment is generated, the preset environment synergy coefficient is proportionally multiplied and corrected by using the environment correction increment, a corrected environment synergy coefficient is generated, and the integrated distress decision index is calculated based on the corrected environment synergy coefficient.

[0025] Preferably, the step S4 comprises the following hierarchical discrimination steps: a step S43 of comparing the integrated distress decision index with a set first discrete threshold and a second discrete threshold, wherein the second discrete threshold is greater than the first discrete threshold; a step S44 of outputting a maintain rest monitoring instruction when the integrated distress decision index is less than the first discrete threshold; outputting a first level early warning instruction when the integrated distress decision index is greater than or equal to the first discrete threshold and less than the second discrete threshold, and starting a local interactive confirmation program; and outputting a second level emergency distress instruction when the integrated distress decision index is greater than or equal to the second discrete threshold.

[0026] Preferably, the step S44 further comprises the following data packaging steps when the second level emergency distress instruction is outputted: a step S441 of extracting the medical history abstract and the drug contraindication parameter in the life green card dataset, combining the real-time acquired geographic coordinate data, the physiological Mahalanobis residual and the time domain feature vector to generate a minimalist first aid distress data packet; and a step S442 of implementing Huffman bit-level lossless compression on the minimalist first aid distress data packet, and sending the compressed minimalist first aid distress data packet to the emergency command center through a wireless communication link.

[0027] Preferably, the step S44 further comprises the following interactive interception steps when the first level early warning instruction is outputted: a step S443 of triggering the local sound-light interactive terminal to start a countdown reminder, setting a countdown duration according to the perception quality factor, and lengthening the countdown duration when the perception quality factor decreases; and a step S444 of, within the countdown duration, if a false alarm confirmation signal inputted by the monitored subject through a touch button is captured, canceling the first level early warning instruction, resetting the physiological abnormality index to zero, and terminating the output of the distress control instruction; and if the countdown duration ends and the false alarm confirmation signal is not captured, automatically upgrading the output of the second level emergency distress instruction.

[0028] Preferably, when the compressed minimal first-aid distress data packet is sent to the emergency command center in step S442, the following multi-channel redundant transmission steps are included: step S4421, detecting the link connection state of the cellular mobile communication network, narrowband Internet of Things, and near-field Mesh ad hoc network; step S4422, selecting a connected link according to the priority order of the cellular mobile communication network, narrowband Internet of Things, and near-field Mesh ad hoc network; if the high-priority network link is disconnected, switching to the next high-priority network link to resend the minimal first-aid distress data packet, thereby ensuring the successful transmission of the minimal first-aid distress data packet.

[0029] Embodiment one: the current first-aid decision-making node monitors an elderly monitored subject with a history of hypertension and coronary heart disease. When the monitored subject is in a compound working condition of severe pressure fluctuation and daily walking, the first-aid decision-making node synchronously collects the body surface photoplethysmographic pulse wave signal, the body surface acceleration signal, and the environmental air pressure data. In a continuous 10s sliding time window, the pulse wave peak time interval sequence is spline interpolated and resampled into a uniform time domain sequence of 4Hz. A 16-order autoregressive model is used to calculate the power in the 0.15Hz to 0.40Hz frequency band to obtain the heart rate variability high-frequency power. At the same time, the lowest value of blood oxygen saturation in the sliding time window is read as the blood oxygen saturation extreme value. The time domain feature vector is constructed in the order of heart rate variability high-frequency power first and blood oxygen saturation extreme value second , where the first dimension corresponds to the heart rate variability high-frequency power, and the second dimension corresponds to the blood oxygen saturation extreme value. The real-time heart rate measured in the sliding time window is 112 beats per minute, and the blood oxygen saturation extreme value is 91%. The first-aid decision-making node also extracts the vector resultant acceleration variance and the attitude angle mutation rate of the body surface acceleration signal. The vector resultant acceleration variance is 2.0m² / s 4 .

[0030] In the extraction process of the attitude angle mutation rate, the three-axis acceleration sensor collects the gravity acceleration component signals of the human body chest surface in the X, Y, and Z axis directions at a sampling frequency of 50Hz. The three-axis acceleration values of each sampling point are low-pass filtered to obtain the static gravity vector. The first-order time difference of the three-axis acceleration components of adjacent sampling points is calculated according to a time step of 20ms to obtain the time domain change slope of the body surface acceleration signal. The pitch angle and roll angle of the human body trunk relative to the horizontal reference plane are calculated using the static gravity vector and the inverse tangent function. The first-order time difference of the three-axis spatial combined attitude angle at adjacent time points is calculated according to the time domain change slope of the three-axis acceleration signal to obtain the attitude angle change rate. The maximum peak value and the root mean square value of the absolute value of the attitude angle change rate in the 10s sliding time window are taken to form the attitude angle mutation rate.

[0031] The emergency decision-making computing node retrieves the Life Green Card dataset from the storage unit based on the unique identifier of the monitored subject. The Life Green Card dataset contains mean vectors of basic physiological characteristics. Covariance matrix inverse matrix and disease risk weighting factors The emergency response decision-making computation node constructs a Mahalanobis distance kernel based on individual physiological states, and integrates time-domain feature vectors. Mapped to the deviation space based on the life green card dataset, and the dimensionless scalar physiological Mahalanobis residuals are calculated using the following formula. : ,in, for Physiological Markov residuals at time. for The temporal feature vector at time t, This is the mean vector of basic physiological characteristics in the Life Green Card dataset. The inverse of the covariance matrix is... This represents the matrix transpose, which transforms the time-domain eigenvectors. with the mean vector of basic physiological characteristics The difference is used to obtain the deviation vector, and then the deviation vector is used in conjunction with the inverse of the covariance matrix. Perform quadratic matrix multiplication and take the square root to obtain the physiological Mahalanobis residual. The value is 8.544. This calculation result represents the degree of deviation of the time-domain feature vector from the basic physiological state of the monitored subject, and eliminates the influence of the correlation between multidimensional physiological parameters on the calculation result.

[0032] The signal-to-noise ratio of the photoplethysmography (PPG) wave signal on the body surface during the shaking caused by the subject's walking. The variance of the vector resultant acceleration is 8.0. 2.0 m² / s 4 The emergency response decision calculation node combines a preset motion artifact attenuation coefficient. Calculate the perceived quality factor using the following formula. : ,in, for Perceived quality factor at any given moment for Signal-to-noise ratio of the photoplethysmography (PPG) signal at any given time. The variance of the vector resultant acceleration. The motion artifact attenuation coefficient is denoted as . Substituting these parameters into the above parameters yields the perceived quality factor. The perceived quality factor is 6.154. With quality threshold Compare and determine Greater than or equal to The sensor signal quality meets the calibration requirements. The sensor calibration mapping table uses the sensing quality factor as the lookup value and records the correspondence between the sensing quality factor and the normalized correction coefficient. When the sensing quality factor is between adjacent calibration points, linear interpolation is used to obtain the normalized correction coefficient. Based on this sensor calibration mapping table, the... Mapped to normalized correction coefficients Physiological Markov residuals Multiply by the normalization correction factor Physiological abnormality index obtained .

[0033] In the above calculation method, the variance of the vector resultant acceleration is expressed in m² / s². 4 In units, motion artifact attenuation coefficient With s 4 The unit is / m², and the product of the two is a dimensionless quantity. The constant 1 in the formula is a dimensionless reference term. Another engineering implementation method is to divide the variance of the vector resultant acceleration by the square of the standard gravitational acceleration 9.8m / s², first convert the variance of the vector resultant acceleration into a dimensionless quantity, and then add it to the dimensionless motion artifact attenuation coefficient and the constant 1.

[0034] When the absolute value of the time-varying rate of change of ambient air pressure data exceeds 0.05 kPa / s, the system activates environmental trigger gating, retrieves the environmental sensitive trigger matrix from the Life Green Card dataset, and proportionally corrects the preset environmental synergy coefficient. The environmental synergy coefficient used in this calculation... The environmental risk coefficient is 0.35. The emergency response decision calculation node combines the attitude angle abrupt change rate and the time change rate of environmental air pressure data to obtain the environmental risk coefficient. And calculate the comprehensive emergency call decision index according to the following formula. : ,in, for A comprehensive emergency call decision index at any moment. As a disease risk weighting factor, for Physiological abnormality index at any time The environmental synergy coefficient used in this calculation is after being gated by environmental factors. for The environmental risk coefficient at any given time, after being substituted into the parameters, forms the comprehensive emergency call decision index. The comprehensive emergency call decision index is 5.760. With the first discrete threshold and the second discrete threshold Compare and determine Greater than or equal to the second discrete threshold , output secondary emergency help instruction, emergency decision calculation node extracts medical history summary and drug contraindication parameters in life green card data set, combined with real-time geographic coordinate data, physiological Mahalanobis residual and time domain feature vector , combined to generate a minimalist emergency help data packet, and the minimalist emergency help data packet is subjected to Huffman bit-level lossless compression, and then sent to the emergency command center through a wireless communication link.

[0035] In the calculation process of the environmental risk coefficient, the absolute value of the time variation rate of the environmental air pressure data is divided by 0.05 kPa / s to obtain the dimensionless air pressure stress factor; the posture angle mutation rate is divided by 180° / s to obtain the dimensionless body position mutation factor, and the weight coefficients of the air pressure stress factor and the body position mutation factor are respectively set to 0.6 and 0.4. The weighted square sum of the two is square rooted, when the absolute value of the time variation rate of the environmental air pressure data does not exceed 0.05 kPa / s, the environmental risk coefficient takes the base value 1.0; when the absolute value exceeds 0.05 kPa / s, the square root of the weighted sum is added to the base value to obtain the dimensionless environmental risk coefficient 1.2.

[0036] Example two: this embodiment constructs a multi-modal physiological and environmental signal test platform, tests the emergency decision process under the conditions of sensing motion artifact interference, individual physiological baseline deviation, environmental change and physiological stress, the test platform includes a multi-channel physiological signal generation module, a three-axis acceleration excitation module and a pressure control cavity. The sampling frequency of the multi-channel physiological signal generation module is set to 250 Hz, the dynamic output range is 0.05 mV to 5.0 mV, and the measurement accuracy is 0.01 mV; the pressure control range of the pressure control cavity is 30 kPa to 120 kPa, the pressure resolution is 0.01 kPa, the test data is taken from the multi-modal physiological time series database, and Gaussian white noise and low-frequency motion interference of different amplitudes are superimposed in the original body surface photoplethysmogram signal and electrocardiogram signal, so that the signal-to-noise ratio of the body surface photoplethysmogram signal covers 1.1 to 12.0, and the vector sum acceleration variance covers 0.1 m² / s 4 to 46.0 m² / s 4 .

[0037] When the sliding time window is shorter than 5s, the frequency resolution of the high frequency power of heart rate variability is lower than 0.05Hz, and it is difficult to separate the respiratory frequency band signal; when the sliding time window is longer than 30s, the time domain response delay of the calculation node increases, and the response to the sudden physiological abnormalities is correspondingly lagged. Based on the above test conditions, the sliding time window is set to 10s. For the pulse wave peak time interval sequence in the 10s sliding time window, the three times spline interpolation is carried out first, then the 4Hz resampling is carried out to obtain the uniform time domain sequence, and the 16-order autoregressive model is used to estimate the power spectrum. After the model parameters are calculated by using the forward and backward least square algorithm, the frequency domain power spectrum density is calculated. In the short sequence composed of 40 resampling data points, the respiratory high frequency power spectrum peak in the range of 0.15Hz to 0.40Hz can be distinguished.

[0038] Test setting test group, control group one, control group two and control group three, the test group adopts Mahalanobis distance kernel mapping, perception quality factor arbitration cutoff, life green card data set coupling and environmental risk collaborative processing; the control group one adopts fixed general threshold determination method, sets the upper limit and lower limit of the group for heart rate and blood oxygen saturation; the control group two adopts linear weighted accumulation method without quality threshold cutoff, carries out linear operation on the original residual error and signal to noise ratio; the control group three adopts Mahalanobis distance method without individual life green card data set, replaces the individualized baseline parameters with the mean vector and unit covariance matrix of the general population, and synchronously processes the time sequence data of 10 groups of working conditions in four groups, and records the physiological Mahalanobis residual error , perception quality factor , normalized correction coefficient , physiological abnormality index , comprehensive help decision index and corresponding output instructions.

[0039] In the resting normal state, the signal to noise ratio of the body surface photoelectric plethysmogram signal is 11.5, the vector sum acceleration variance is 0.05m² / s 4 , the physiological Mahalanobis residual error calculated by the test group according to the life green card data set is 0.640, and the perception quality factor obtained by combining the above signal to noise ratio and vector sum acceleration variance is 11.414; the normalized correction coefficient obtained according to the sensor calibration mapping table is 0.90, and then the physiological abnormality index is 0.576, the comprehensive help decision index is 0.755, which is lower than the first discrete threshold, and the test group outputs the maintenance resting monitoring instruction; the control group one, the control group two and the control group three also output the maintenance resting monitoring instruction.

[0040] Under baseline deviation conditions in elderly individuals, the monitored subject's resting intrinsic heart rate was 88 beats / min, blood oxygen saturation was 95%, the signal-to-noise ratio of the surface photoplethysmography (PPG) signal was 9.8, and the variance of the vector summation acceleration was 0.12 m² / s². 4 The experimental group used the mean vector and inverse covariance matrix of the basic physiological characteristics in the Life Green Card dataset to obtain the physiological Mahalanobis residuals. The value is 0.640; the perceived quality factor is calculated based on the input signal quality. The value is 9.627, and the normalized correction coefficient is obtained by mapping. The value is 0.75, from which the physiological abnormality index is obtained. The comprehensive emergency call decision index is 0.480. The value was 0.670. The experimental group output the command to maintain resting monitoring, and the control group 2 also output the command to maintain resting monitoring. The control groups 1 and 3 were judged using the general population baseline, and both output the first-level warning command.

[0041] Under mild walking motions, the signal-to-noise ratio of the surface photoplethysmography (PPG) signal was 6.2, and the variance of the vector resultant acceleration was 1.50 m² / s². 4 The experimental group calculated the physiological Markov residuals. The perceived quality factor is 0.816. The value is 5.061, and the normalized correction coefficient is obtained from the sensor calibration mapping table. The physiological abnormality index is obtained by multiplying the physiological Markov residual by a normalization correction factor of 0.45. The value is 0.367; after combining the disease risk weighting factor and the environmental risk coefficient, the comprehensive emergency call decision index is... The value was 0.494. The experimental group, control group 1, control group 2 and control group 3 all output the command to maintain resting monitoring.

[0042] Under conditions of intense running artifacts, motion interference reduces the signal-to-noise ratio of the surface photoplethysmography (PPG) pulse wave signal to 1.1, while increasing the variance of the vector sum acceleration to 46.00 m² / s². 4 The experimental group calculated the physiological Mahalanobis residuals based on the time-domain feature vectors and the life green card dataset. The value is 8.544, but the perceived quality factor obtained from the signal-to-noise ratio and the variance of the vector sum acceleration is... The value is only 0.139, which is below the quality threshold, thus the normalization correction factor is... The threshold value was truncated to a low confidence level of 0.05 to obtain the physiological abnormality index. The comprehensive emergency call decision index is 0.427. The index was 0.569, which was below the first discrete threshold, and the experimental group output a command to maintain resting monitoring; control groups one, two, and three all output a level two emergency call command.

[0043] Under sensor contact vibration conditions, the signal-to-noise ratio of the photoplethysmography (PPG) wave signal from the body surface is 0.8, and the variance of the vector resultant acceleration is 32.00 m² / s². 4 The experimental group obtained physiological Markovian residuals. The perceived quality factor is 14.215. The value is 0.138. Since this perceived quality factor is below the quality threshold, the normalized correction coefficient is... Similarly, the threshold value was truncated to 0.05 to obtain the physiological abnormality index. The comprehensive emergency call decision index is 0.711. The value was 0.924, and the experimental group output a resting monitoring command accordingly; control groups 1, 2, and 3 all output a level 2 emergency call command.

[0044] Under normal sudden stress conditions, the signal-to-noise ratio of the photoplethysmography (PPG) signal on the body surface is 8.0, and the variance of the vector resultant acceleration is 2.00 m² / s². 4 The experimental group calculated the physiological Markov residuals. The perceived quality factor is 8.544. The perceived quality factor is 6.154, which is not lower than the quality threshold. The normalized correction coefficient is obtained according to the sensor calibration mapping table. The value is 0.50, which yields the physiological abnormality index. The value is 4.272; after combining disease risk weighting factors and environmental risk coefficients, the comprehensive emergency call decision index is... The index is 5.760, which is greater than or equal to the second discrete threshold. The experimental group outputs a level 2 emergency call command, and control groups 1 and 2 also output level 2 emergency call commands. Control group 3 outputs a level 1 warning command.

[0045] Under severe low-pressure induced conditions, the signal-to-noise ratio of the surface photoplethysmography (PPG) signal was 7.5, and the variance of the vector resultant acceleration was 1.80 m² / s². 4 The experimental group obtained physiological Markovian residuals. The perceived quality factor is 7.058. The value is 5.906, and the normalized correction coefficient is obtained by mapping. The score was 0.50. After correction, the physiological abnormality index was... The value is 3.529; this index, together with the environmental risk coefficient, is used to calculate the comprehensive emergency call decision index, resulting in... The index is 5.391, which is greater than or equal to the second discrete threshold. The experimental group outputs a level-two emergency call command; control groups one, two, and three all output level-one warning commands.

[0046] Under early, weak stress conditions, the signal-to-noise ratio of the surface photoplethysmography (PPG) signal was 8.5, and the variance of the vector resultant acceleration was 0.80 m² / s². 4 The experimental group calculated the physiological Markov residuals. The perceived quality factor is 3.820. The value is 7.589, and the normalized correction coefficient is obtained by mapping. The value is 0.60, from which the physiological abnormality index is obtained. The comprehensive emergency call decision index is 2.292. The index is 2.900. This index is greater than or equal to the first discrete threshold and less than the second discrete threshold. The experimental group outputs a first-level warning command, the control group 2 also outputs a first-level warning command, and both control group 1 and control group 3 output a command to maintain rest monitoring.

[0047] Under low-temperature conditions, the signal-to-noise ratio of the surface photoplethysmography (PPG) pulse wave signal is 8.2, and the variance of the vector resultant acceleration is 0.90 m² / s². 4 The experimental group obtained physiological Markovian residuals. The perceived quality factor is 4.150. The value is 7.225, and the normalized correction coefficient is obtained by mapping. The physiological abnormality index was 0.60 after correction. The value is 2.490; the comprehensive emergency call decision index is calculated by combining the environmental risk coefficient. The value is 3.325, which is between the first and second discrete thresholds. The experimental group outputs a first-level warning command, and the control group 2 also outputs a first-level warning command. Both control group 1 and control group 3 output a command to maintain rest monitoring.

[0048] Under the artifact recovery and self-healing condition, the signal-to-noise ratio of the surface photoplethysmography pulse wave signal is 7.8, and the variance of the vector resultant acceleration is 0.30 m² / s². 4 The experimental group calculated the physiological Markov residuals. The perceived quality factor is 1.105. The value is 6.812, and the normalized correction coefficient is obtained by mapping. The physiological abnormality index is obtained by multiplying the physiological Markov residual by a normalization correction factor of 0.55. The comprehensive emergency call decision index is 0.608. The value was 0.795. The experimental group, control group 1, control group 2 and control group 3 all output the command to maintain resting monitoring.

[0049] In the above 10 groups of working conditions, the high physiological Mahalanobis residual is generated by the running artifact and the sensor contact shaking, but the corresponding perception quality factor is 0.139 and 0.138 respectively, and the normalized correction coefficient is truncated to 0.05, and the comprehensive help decision index is maintained at 0.569 and 0.924 respectively, and the test group maintains the resting monitoring, and in the high baseline deviation working condition, the test group obtains the physiological Mahalanobis residual of 0.640 and the comprehensive help decision index of 0.670 according to the individual life green card data set, and the individual resting baseline deviation is not judged as physiological abnormality, and in the conventional sudden stress and low pressure induced working conditions, the comprehensive help decision index of the test group is 5.760 and 5.391 respectively, and the secondary emergency help instruction is output; in the early weak stress and low temperature environment superposition working condition, the comprehensive help decision index is 2.900 and 3.325 respectively, and the first level warning instruction is output.

[0050] In the third embodiment, the emergency decision calculation node synchronously collects the body surface photoplethysmogram signal, the body surface acceleration signal, the environmental pressure data and the environmental temperature data, when the absolute value of the time variation rate of the environmental pressure data reaches 0.08 kPa / s, the absolute value of the time variation rate of the environmental pressure data exceeds 0.05 kPa / s when the environmental temperature data is-5℃, and the environmental temperature is lower than 0℃, the system activates the environmental inducement gate, and the environmental sensitive inducement matrix is called from the life green card data set .

[0051] The environmental abnormality deviation vector is calculated according to the environmental pressure data and the environmental temperature data , wherein the first dimension takes the value 0.08 corresponding to the absolute value 0.08 kPa / s of the time variation rate of the environmental pressure data, and the second dimension is normalized to 0.50 according to the preset temperature calibration relationship from the temperature deviation of-5℃ relative to 0℃, the environmental sensitive inducement matrix is stored in the form of one row and two columns, and the environmental abnormality deviation vector is transposed to form a column vector of two rows and one column, and the product of and is calculated, wherein the first dimension value 1.50 is multiplied by 0.08 to obtain 0.12, and the second dimension value 0.20 is multiplied by 0.50 to obtain 0.10, and the sum of the two products is 0.22, which is the environmental correction increment, and the environmental correction increment is used as the proportional increment relative to 1, and the preset environmental coordination coefficient is multiplied to obtain the corrected environmental coordination coefficient .

[0052] When the body surface motion interference is enhanced, the vector resultant acceleration variance is increased to 38.5 m² / s 4 , and the signal-to-noise ratio of the body surface photoplethysmogram signal decreases to 1.2, thus the perceived quality factor is calculated is 0.177, since is below the quality threshold , the system will normalize the correction factor to the low confidence limit value 0.05, while triggering the signal channel self-healing compensation.

[0053] The signal channel self-healing compensation extracts the main peak frequency component in the frequency domain distribution of the body surface acceleration signal, and updates the tap coefficient of the adaptive filter according to the frequency domain distribution, to reconstruct the noisy baseline of the body surface photoplethysmogram signal, filter out the motion frequency band interference, and generate a denoised photoplethysmogram signal, based on which the signal-to-noise ratio is recalculated, increases to 7.5, the perceived quality factor is updated to 5.882, and the system maps to the normalized correction factor according to the calibration interval corresponding to the perceived quality factor in the sensor calibration mapping table, multiplies the physiological Mahalanobis residual by , to obtain the physiological abnormality index .

[0054] In the signal channel self-healing compensation process, the system uses a 32-order transversal finite impulse response filter, takes the body surface acceleration signal as the reference noise input, and takes the noisy body surface photoplethysmogram signal as the original input. First, the body surface acceleration signal is subjected to fast Fourier transform to extract the center frequency and bandwidth corresponding to the energy main peak, and the initial center stopband frequency band of the filter is determined according to the center frequency. Then, using the least mean square error algorithm, the weight coefficients of each tap are iteratively updated according to the cross-correlation between the output residual and the reference noise, with a step factor of 0.01. After 50 iterations of convergence, the tap coefficients approximate the transfer function of the motion artifact, and the baseline drift and high-frequency oscillation noise caused by human motion are deducted from the original input.

[0055] When the monitored subject is in a resting state, the vector sum acceleration variance is less than 0.01 m² / s 4 , and the perceived quality factor is greater than or equal to 0.90, the system writes the time domain feature vector of the real-time collected body surface photoplethysmogram signal into the sliding memory buffer, and after the sliding memory buffer accumulates 1024 groups of samples, calculates the mean value of the time domain feature vectors in the buffer, and retrieves the updated basic physiological feature mean vector from the Green Card data set, combines the incremental update factor , and calculates the updated basic physiological feature mean vector according to the following formula : wherein, is the updated mean vector of the base physiological features, is the mean vector of the base physiological features before the update, is the incremental update factor, is the mean of the time domain feature vector in the sliding memory buffer, after substituting the above parameters, we get , and update the mean vector of the base physiological features in the life green card dataset with this vector.

[0056] When the integrated distress decision index is greater than or equal to the first discrete threshold and less than the second discrete threshold , the system outputs a first-level warning instruction, triggers the local sound-light interactive terminal to start a countdown reminder, and sets the countdown duration according to the real-time perception quality factor . If the confirmation false alarm signal input by the monitored subject through the touch button is captured within the countdown, the system cancels the first-level warning instruction, resets the physiological abnormality index to zero, and terminates the output of the distress control instruction; if the confirmation false alarm signal is not captured when the countdown ends, the second-level emergency distress instruction is output.

[0057] After outputting the second-level emergency distress instruction, the system extracts the medical history summary and drug contraindication parameters from the life green card dataset, and combines the real-time geographic coordinate data, physiological Mahalanobis residual and time domain feature vector to generate a minimalist first-aid distress data packet after de-identification processing. After lossless compression of the data packet by Huffman bit level, the link connection state of the cellular mobile communication network, narrowband Internet of Things and near-field Mesh ad hoc network is detected, and only in the detected connected link, the sending link is selected according to the priority order of the cellular mobile communication network, narrowband Internet of Things and near-field Mesh ad hoc network. When the high-priority cellular mobile communication network link is disconnected, switch to the next high-priority narrowband Internet of Things link, and resend the minimalist first-aid distress data packet.

[0058] The reference value of the countdown duration is set to 15s. When the perception quality factor is greater than or equal to the quality threshold 1.0, the system maintains the countdown of 15s; when the perception quality factor is lower than the quality threshold 1.0, the countdown duration is extended by multiplying 15s by the ratio of the quality threshold to the current perception quality factor. When the perception quality factor drops to 0.5, the countdown duration is extended to 30s. The upper limit of the countdown duration is set to 60s.

[0059] Embodiment four: This embodiment combines Figs. 1-2 , to describe an intelligent distress decision-making method that integrates multi-modal perception and life green card, such asFig. 1 As shown, step S1 acquires the body surface photoplethysmographic pulse wave signal, the body surface acceleration signal, the ambient air pressure data and the ambient temperature data of the monitored subject, extracts the time domain feature vector of the body surface photoplethysmographic pulse wave signal, calculates the signal-to-noise ratio of the body surface photoplethysmographic pulse wave signal, extracts the vector sum acceleration variance and the attitude angle mutation rate of the body surface acceleration signal; step S1 flows to step S2, and the life green card data set is called, the life green card data set includes the basic physiological feature mean vector, the covariance matrix inverse matrix and the disease risk weight factor, the time domain feature vector is mapped to the deviation space of the life green card data set, and the physiological Mahalanobis residual error is generated; step S2 flows to step S3, and the perceptual quality factor is calculated according to the vector sum acceleration variance and the signal-to-noise ratio of the body surface photoplethysmographic pulse wave signal; if the perceptual quality factor is lower than the quality threshold, the normalized correction coefficient is truncated to the lower limit value of the low confidence, otherwise the normalized correction coefficient is mapped; the physiological Mahalanobis residual error is multiplied by the normalized correction coefficient to generate the physiological abnormality index; step S3 flows to step S4, and the product of the disease risk weight factor and the physiological abnormality index and the sum of the product of the preset environmental synergy coefficient and the environmental risk coefficient are calculated according to the environmental risk coefficient constructed based on the disease risk weight factor, the attitude angle mutation rate and the time variation rate based on the ambient air pressure data, to generate the comprehensive help decision index, and the discrete gradient threshold is compared to output the help control instruction.

[0060] As shown in the figure, Fig. 2 The comprehensive help decision index compares the discrete gradient threshold, when it is determined to be less than the first discrete threshold, the resting and normal state determination is output, and the continuous monitoring state is maintained; when it is determined to be greater than or equal to the first and less than the second threshold, the first level early warning and local interaction are triggered, respectively, the local sound and light interaction terminal is set according to the perceptual quality factor, the countdown time length is set, and the false alarm signal is captured through the touch button input, the alarm is cancelled and the index is reset; when it is determined to be greater than or equal to the second discrete threshold, the second level emergency help trigger is generated, the simplified first aid help data packet is generated to extract the medical history abstract, the parameters and geographic coordinate data, and the multi-channel redundant transmission is selected according to the priority order.

[0061] Embodiment five: before the deployment of the emergency decision calculation node, a multi-axis vibration table and a multi-wavelength light emitting diode detection module are used to apply a discrete step excitation to the body surface photoplethysmographic pulse wave sensor and the body surface acceleration sensor. During the calibration process, the vector sum acceleration variance is set to be within the range of 0.01 m² / s 4 to 50.0 m² / s 4 , and the signal-to-noise ratio of the body surface photoplethysmographic pulse wave is set in the range of 0.5 to 15.0, forming a cross gradient working condition; collect the sensor output under each working condition to form a discrete point array data, and calculate the corresponding perception quality factor according to the variance of each group of vector resultant acceleration and the signal-to-noise ratio .

[0062] with a quality threshold as the demarcation point of the normalized mapping interval, when the perception quality factor is greater than or equal to 1.0, a continuous numerical mapping relationship between the perception quality factor and the normalized correction coefficient is established according to the discrete point array data; the generated sensor calibration mapping table is written into the non-volatile memory of the emergency decision calculation node for real-time calculation.

[0063] The sensor calibration mapping table adopts a double-column one-dimensional discrete point array lookup table and is stored in the read-only area of the non-volatile memory, the first column is the perception quality factor node, the node value starts from 1.0 and increases by 1.0 to 10.0, a total of 10 discrete nodes; the second column is the normalized correction coefficient node, the calibration candidate values are 0.10, 0.20, 0.35, 0.50, 0.65, 0.78, 0.88, 0.94, 0.98 and 1.00 in turn, after completing the cross gradient calibration, according to the actual measured perception quality factor and its corresponding correction result under each working condition, the two columns of nodes are paired and solidified as effective calibration records in the order of the size of the perception quality factor node.

[0064] In the real-time calculation process, first, the adjacent effective calibration records located on both sides of the current perception quality factor are retrieved from the sensor calibration mapping table, then the relative position of the current perception quality factor between the two adjacent nodes is calculated; the relative position is multiplied by the difference between the two normalized correction coefficient nodes, and added to the normalized correction coefficient corresponding to the lower node to obtain the current normalized correction coefficient, when the perception quality factor is 6.154, the effective calibration records corresponding to the 6th node and the 7th node are retrieved, and after first-order linear interpolation, the normalized correction coefficient 0.50 is obtained.

[0065] In the initial wearing stage of the emergency decision calculation node, a pre-resting sampling program of 24 consecutive hours is started to determine the initial parameters of the life green card data set of the monitored subject, under the physiological resting working condition that the vector resultant acceleration variance is less than 0.01 m² / s 4 and the perception quality factor is greater than or equal to 0.90, the real-time heart rate sequence and the blood oxygen saturation sequence are continuously collected; in each sliding time window, the heart rate variability high frequency power is extracted according to the real-time heart rate sequence, and the blood oxygen saturation extreme value is extracted from the blood oxygen saturation sequence, and the time domain feature vector is constructed in accordance with the fixed dimension order .

[0066] resting sampling procedure extracts group time domain feature vector , the data of each dimension in the sample set is calculated respectively to obtain the basic physiological feature mean vector ; at the same time, the sample covariance matrix is calculated according to the deviation of each dimension data relative to the corresponding mean value, and the Gauss elimination method is used to inverse the sample covariance matrix to obtain the covariance matrix inverse matrix , the preset disease risk weight factor is read according to the clinical history level of the monitored subject , the basic physiological feature mean vector , the covariance matrix inverse matrix and the disease risk weight factor are written into the life green card data storage area of the emergency decision calculation node to form the life green card data set of the monitored subject, and the initial parameter calibration of the deviation space is completed.

[0067] Example six: when the emergency decision calculation node determines that the comprehensive emergency decision index is greater than or equal to the second discrete threshold , outputting a secondary emergency call instruction, the microcontroller retrieves the life green card data set from the static storage area, extracts the medical history abstract and drug contraindication parameters, and masks the personal identity field through bit masking, combines the de-identified medical history abstract, drug contraindication parameters, real-time geographic coordinate data, physiological Mahalanobis residual and time domain feature vector to generate a minimalist emergency call data packet with a length of 128 bytes, and the data packet carries the above data in the original emergency call data structure before compression.

[0068] The microcontroller calls the preset static Huffman coding tree to implement Huffman bit-level lossless compression on the minimalist emergency call data packet. The static Huffman coding tree is constructed according to the frequency of characters in the emergency clinical history data. In the original emergency call data structure of 128 bytes, the resting and normal physiological state mask, the prefix reference number of geographic coordinates and other high-frequency characters account for more than 75% of the total character frequency. The depth of the coding tree is 8 layers, wherein the high-frequency static mask and zero value field correspond to a short prefix code with a length of 2 bits to 3 bits, including 00 or 010; the physiological Mahalanobis residual and the change value corresponding to the medium-frequency appearance correspond to a code with a length of 5 bits to 7 bits; the special medical history abnormal character corresponding to the low-frequency appearance corresponds to a code with a length of 12 bits.

[0069] The microcontroller queries the static Huffman coding tree according to the byte sequence of the minimal emergency distress data packet, replaces each field with the corresponding variable-length bit stream, and losslessly compresses the 128-byte minimal emergency distress data packet into 34 bytes. The compressed minimal emergency distress data packet is written into the direct memory access buffer of the wireless communication controller and waits for the wireless communication link to send.

[0070] The wireless communication controller detects the link connection state of the cellular mobile communication network, the narrowband Internet of Things, and the near-field mesh ad hoc network, and selects a sending link from the links in the connected state according to the priority order of the cellular mobile communication network, the narrowband Internet of Things, and the near-field mesh Ad hoc network. For the cellular mobile communication network, the wireless communication controller polls the link state register of the cellular mobile communication network transceiver at a period of 1 s. When the received signal strength indication is not lower than -110 dBm and the network registration state is registered, the network is determined to be connected, and the compressed minimal emergency distress data packet is sent through the network. The wireless communication controller also reads the link state flag of the narrowband Internet of Things and the near-field mesh ad hoc network to determine whether it has the data sending condition.

[0071] When the received signal strength indication of the cellular mobile communication network is lower than -110 dBm, and the network registration state returns to unregistered, the network is determined to be disconnected, the control unit turns off the power supply of the cellular mobile communication network transceiver, turns on the narrowband Internet of Things communication module, and completes the initialization within 120 ms. The 34-byte compressed minimal emergency distress data packet in the direct memory access buffer is sent through the narrowband Internet of Things link.

[0072] After the narrowband Internet of Things link is sent, the control unit starts a 3-s response timer. When the response timer times out and no acknowledgment frame returned by the emergency command center is received, if the near-field mesh ad hoc network is in a connected state, the control unit starts the 2.4 GHz radio frequency physical layer, delivers the compressed minimal emergency distress data packet to the near-field mesh ad hoc network, and relays and forwards it by the adjacent wearable computing nodes. When the narrowband Internet of Things link is detected to be disconnected before sending, the control unit directly selects the connected near-field mesh ad hoc network to send the compressed minimal emergency distress data packet.

[0073] It is obvious to those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application.

[0074] Finally, it should be noted that the above examples are merely intended to illustrate the technical solutions of the present application and not to limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced without departing from the spirit and scope of the present application.

Claims

1. An intelligent emergency call decision-making method integrating multimodal perception and life-saving green card, characterized in that, Includes the following steps: Step S1: Acquire the surface photoplethysmography (PPG) signal, surface acceleration signal, ambient air pressure data, and ambient temperature data of the monitored subject; extract the time-domain feature vector of the surface PPG signal; calculate the signal-to-noise ratio of the surface PPG signal; and extract the vector sum acceleration variance and attitude angle change rate of the surface acceleration signal. Step S2: Retrieve the Life Green Card dataset, which includes the mean vector of basic physiological characteristics, the inverse of the covariance matrix, and disease risk weight factors. The temporal feature vectors are mapped to the deviation space of the life green card dataset to generate physiological Mahalanobis residuals; Step S3: Calculate the sensing quality factor based on the variance of the vector sum acceleration and the signal-to-noise ratio of the photoplethysmography pulse wave signal on the body surface. If the perceived quality factor is below the quality threshold, the normalized correction coefficient is truncated to the lower confidence limit; otherwise, it is mapped to the normalized correction coefficient. The physiological Markov residuals are multiplied by a normalization correction factor to generate the physiological abnormality index. Step S4: Based on the disease risk weight factor, attitude angle mutation rate, and environmental risk coefficient constructed based on the time change rate of environmental air pressure data, calculate the product of the disease risk weight factor and the physiological abnormality index, and the sum of the product of the preset environmental synergy coefficient and the environmental risk coefficient, generate a comprehensive emergency call decision index, compare it with the discrete gradient threshold, and output the emergency call control command.

2. The intelligent emergency call decision-making method integrating multimodal perception and life green card as described in claim 1, characterized in that, Step S3 includes the following sub-steps: Step S31, when the perceived quality factor is lower than the quality threshold, the normalized correction coefficient is truncated to the low confidence lower limit value, triggering the signal channel self-healing compensation; Step S32, the tap coefficients of the adaptive filter are updated according to the frequency domain distribution of the body surface acceleration signal, the noisy baseline of the body surface photoplethysmography (PPG) signal is reconstructed, the denoised PPG signal is generated, the signal-to-noise ratio is recalculated based on the denoised PPG signal, and the perceived quality factor is updated.

3. The intelligent emergency call decision-making method integrating multimodal perception and life green card as described in claim 1, characterized in that, Step S1 includes the following sub-steps: Step S11, set a sliding time window, extract the high-frequency power of heart rate variability and the extreme value of blood oxygen saturation of the body surface photoplethysmography pulse wave signal within the sliding time window, and combine them to construct a time-domain feature vector; Step S12, calculate the vector sum of the body surface acceleration signal in the three-dimensional axis, solve the variance of the vector sum within the sliding time window, generate the variance of the vector sum acceleration, and calculate the attitude angle change rate based on the slope of the time-domain change of the body surface acceleration signal.

4. The intelligent emergency call decision-making method integrating multimodal perception and life green card as described in claim 1, characterized in that, Step S2 includes the following sub-steps: Step S21, retrieve the Life Green Card dataset from the storage unit according to the identity identifier of the monitored subject. The Life Green Card dataset also contains the drug contraindication parameters of the monitored subject; Step S22, construct a Mahalanobis distance kernel based on the individual's physiological state, use the Mahalanobis distance kernel to calculate the difference between the time-domain feature vector and the mean vector of the basic physiological features to obtain the deviation vector, perform a quadratic matrix multiplication calculation on the deviation vector and the inverse of the covariance matrix, take the square root of the product result to generate a dimensionless scalar physiological Mahalanobis residual, and eliminate the correlation interference between multidimensional physiological parameters.

5. The intelligent emergency call decision-making method integrating multimodal perception and life green card as described in claim 1, characterized in that, Step S4 includes the following environmental adaptive adjustment steps: Step S41, retrieve the environmental sensitive trigger matrix from the Life Green Card dataset; Step S42, monitor the time change rate of environmental air pressure data and environmental temperature data, and calculate the environmental anomaly offset vector based on the measured environmental air pressure data and environmental temperature data; when the absolute value of the time change rate of environmental air pressure data exceeds 0.05 kPa / s or the environmental temperature data is below 0°C, activate the environmental trigger gating, calculate the transpose product of the environmental sensitive trigger matrix and the environmental anomaly offset vector, generate the environmental correction increment, use the environmental correction increment to perform proportional multiplication correction on the preset environmental coordination coefficient, generate the corrected environmental coordination coefficient, and calculate the comprehensive emergency call decision index based on the corrected environmental coordination coefficient.

6. The intelligent emergency call decision-making method integrating multimodal perception and life green card as described in claim 1, characterized in that, Step S4 includes the following hierarchical discrimination steps: Step S43, compare the comprehensive emergency call decision index with the set first discrete threshold and second discrete threshold, wherein the second discrete threshold is greater than the first discrete threshold; Step S44, when the comprehensive emergency call decision index is less than the first discrete threshold, output the command to maintain rest monitoring. When the comprehensive emergency call decision index is greater than or equal to the first discrete threshold and less than the second discrete threshold, a level one warning command is output and the local interactive confirmation procedure is initiated; when the comprehensive emergency call decision index is greater than or equal to the second discrete threshold, a level two emergency call command is output.

7. The intelligent emergency call decision-making method integrating multimodal perception and life green card as described in claim 6, characterized in that, When outputting the Level 2 emergency call command in step S44, the following data packaging steps are also included: Step S441, extract the medical history summary and drug contraindication parameters from the Life Green Card dataset, and combine them with the real-time acquired geographic coordinate data, physiological Mahalanobis residuals and time-domain feature vectors to generate a simplified emergency call data packet; Step S442, perform Huffman bit-level lossless compression on the simplified emergency call data packet, and send the compressed simplified emergency call data packet to the emergency command center through a wireless communication link.

8. The intelligent emergency call decision-making method integrating multimodal perception and life green card as described in claim 6, characterized in that, When outputting the first-level warning command in step S44, the following interactive interception steps are also included: Step S443, trigger the local audio-visual interactive terminal to start the countdown reminder, set the countdown duration according to the perception quality factor, and extend the countdown duration when the perception quality factor decreases; In step S444, if a false alarm confirmation signal is detected by the monitored subject via touch button within the countdown time, the first-level warning command is canceled, the physiological abnormality index is reset to zero, and the output of the distress call control command is terminated; if the countdown time ends and no false alarm confirmation signal is detected, the output of the second-level emergency distress call command is automatically upgraded.

9. The intelligent emergency call decision-making method integrating multimodal perception and life green card as described in claim 7, characterized in that, When sending the compressed simplified emergency call data packet to the emergency command center in step S442, the following multi-channel redundant transmission steps are included: Step S4421, detecting the link connectivity status of the cellular mobile communication network, narrowband IoT, and near-field Mesh ad hoc network; Step S4422, selecting a connected link according to the priority order of the cellular mobile communication network, narrowband IoT, and near-field Mesh ad hoc network; if the high-priority network link is disconnected, switching to the second-highest priority network link to retransmit the simplified emergency call data packet, ensuring successful transmission of the simplified emergency call data packet.

Images