Ultrahigh heart rate anomaly detection method based on real-time signal processing

Abstract

The invention discloses an ultrahigh heart rate anomaly detection method based on real-time signal processing, and relates to the technical field of pet health monitoring. According to the ultrahigh heart rate anomaly detection method, local low-delay processing is achieved by fusing multi-modal physiological signals and utilizing edge calculation, signal preprocessing is conducted through band-pass filtering, self-adaptive Kalman filtering and the like, and the abnormal heart rate is obtained. Multi-time scale feature extraction is combined with pet state vectors to dynamically adjust feature weights, an improved attention mechanism neural network and support vector machine secondary classification and a posterior probability fusion algorithm are used to improve classification precision, and a dynamic threshold mechanism and a grading early warning mechanism are used to match response intensity according to abnormal duration, so that the classification accuracy is improved. According to the invention, accurate and real-time monitoring of the abnormal ultrahigh heart rate of the pet is realized, an efficient and reliable detection process is provided for health management of the pet, and the method has important significance for early detection of potential diseases of the pet.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of pet health monitoring, in particular to an ultra-high heart rate anomaly detection method based on real-time signal processing. BACKGROUND

[0002] With the improvement of people's living standards and the increase in the number of pets, pet health assessment is becoming increasingly important. As an effective health monitoring method, ultra-high heart rate anomaly detection of pets is gradually becoming an important part of pet medical care. Rapid heart rate may be a signal of potential illness in pets, such as heart disease, respiratory problems, or emotional anxiety. Through wearable devices or specialized detection instruments, the heart rate changes of pets can be monitored in real time, and abnormalities can be detected in time. In the case of ultra-high heart rate, the owner can take intervention measures earlier to avoid the disease from getting worse.

[0003] However, the existing technology has obvious defects in detecting ultra-high heart rate anomalies of pets. First, it relies on single heart rate value judgment, ignoring key features such as PPG waveform morphology and heart rate variability, making it difficult to distinguish between physiological (such as exercise) and pathological (such as ventricular tachycardia) ultra-high heart rate, with a high false positive rate. Second, it uses fixed threshold detection without dynamic adjustment based on pet activity state and individual baseline differences, lacking targeted capture of transient abnormal features, resulting in insufficient timeliness of detection. SUMMARY

[0004] The purpose of the present application is to provide an ultra-high heart rate anomaly detection method based on real-time signal processing to solve the problems raised in the background art.

[0005] To achieve the above purpose, the present application provides the following technical solution: an ultra-high heart rate anomaly detection method based on real-time signal processing, comprising the following steps:

[0006] Step 1, real-time acquisition of multi-modal physiological signals, including PPG heart rate signal, three-axis acceleration signal and skin temperature signal, construction of real-time data stream;

[0007] Step 2, preprocessing of the real-time data stream, including band-pass filter noise reduction, motion artifact removal and signal standardization, to obtain clean signals;

[0008] Step 3, multi-time scale feature extraction of the preprocessed PPG signal, the time scale including instantaneous feature, short-term feature and trend feature, the extracted features including peak interval, heart rate variability, waveform slope and instantaneous heart rate change rate;

[0009] Step 4, calculation of activity intensity based on three-axis acceleration signal, construction of pet state vector combined with skin temperature signal, dynamic adjustment of feature weight;

[0010] Step 5: Input the multi-scale features into the improved attention mechanism neural network, and output the preliminary classification results and confidence scores of the ultra-high heart rate.

[0011] Step 6: Use a support vector machine to perform secondary classification on the multi-scale features, and combine the confidence scores from step 5 with a posterior probability fusion algorithm to obtain the final classification result;

[0012] Step 7: Based on the pet's historical health data and current activity intensity, dynamically adjust the threshold for judging abnormally high heart rate. If the final classification result is pathological and exceeds the threshold, a real-time warning will be triggered.

[0013] Preferably, in step 1, the acquisition frequency of the multimodal physiological signals is: 100Hz for PPG signal, 50Hz for triaxial acceleration signal, and 1Hz for skin temperature signal, and local real-time processing is achieved through edge computing with a delay of ≤100ms.

[0014] Preferably, in step 2, the motion artifact removal employs an adaptive Kalman filter algorithm, the formula of which is: in, The filtered signal, z k For the original PPG signal, K k Let H be the Kalman gain and H be the observation matrix.

[0015] Preferably, in step 3, the formula for calculating the instantaneous heart rate change rate is: Where HR(t) is the instantaneous heart rate at time t.

[0016] Preferably, in step 4, the pet state vector includes activity intensity, body temperature deviation, and resting heart rate deviation, which are standardized using the following formula: Where A(t) is the acceleration magnitude at time t, T0 is the baseline body temperature, and HR is... base This represents the pet's historical average resting heart rate.

[0017] Preferably, in step 5, the improved attention mechanism neural network includes three parallel attention heads, corresponding to time scales of 0.1s, 1s, and 5s, respectively. The attention weights are dynamically adjusted using the pet's state vector, and the calculation formula is: a s =Softmax(W s ·S+b s ), where a s W represents the weight for the s-th time scale. s b s These are the training parameters.

[0018] Preferably, in step 6, the formula for the posterior probability fusion algorithm is: P(C)=ω1·PNN (C) + ω2·P SVM (C), wherein P(C) is the final classification probability, P NN , P SVM are the output probabilities of the neural network and the support vector machine, respectively, and ω1, ω2 are the dynamic weights based on the confidence.

[0019] Preferably, in step 7, the dynamic threshold is updated by the following formula: Th(t) = Th0·(1 + β·A(t)), wherein Th0 is the pet baseline threshold, β is the activity influence coefficient, and A(t) is the current activity intensity.

[0020] Preferably, in step 7, the real-time early warning includes a grading mechanism.

[0021] First-level early warning: when the duration of pathologically high heart rate is < 5 seconds, the device locally issues a vibration reminder.

[0022] Second-level early warning: when the duration of pathologically high heart rate is ≥ 5 seconds and < 30 seconds, the vibration reminder simultaneously pushes the warning information to the user's mobile phone APP.

[0023] Third-level early warning: when the duration of pathologically high heart rate is ≥ 30 seconds, the emergency contact call is triggered + the positioning information is pushed, and the medical auxiliary data containing the heart rate waveform is synchronously sent.

[0024] The application also provides an ultra-high heart rate anomaly detection device based on real-time signal processing, which comprises a processor, a memory and a system bus, the processor and the memory are connected through the system bus, the memory is used for storing one or more programs, the one or more programs include instructions, the instructions make the processor execute the detection method of any one of the above when executed by the processor.

[0025] Compared with the prior art, the application has the beneficial effects that: through multi-modal physiological signal fusion and multi-time scale feature extraction, combined with an improved attention mechanism neural network and a support vector machine secondary classification, the application can accurately distinguish between physiological and pathological ultra-high heart rates, greatly reducing the false positive rate, at the same time, based on the pet state vector, the feature weights and the determination threshold are dynamically adjusted, individual differences and activity states are adapted, and the robustness is improved. With the help of edge computing, low-delay processing is realized, and combined with the grading early warning mechanism, the application can not only capture abnormalities in real time, but also reasonably trigger reminders according to the severity, providing efficient and reliable technical support for pet health monitoring. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 The flowchart of the ultra-high heart rate anomaly detection method based on real-time signal processing of the embodiments of the application. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0028] Please refer to Figure 1 The embodiments of the present application provide an ultra-high heart rate anomaly detection method based on real-time signal processing, including the following steps:

[0029] Step 1, real-time acquisition of multi-modal physiological signals, including PPG heart rate signals, three-axis acceleration signals and skin temperature signals, to construct a real-time data stream, wherein the acquisition frequency of the multi-modal physiological signals is: PPG signal 100 Hz, three-axis acceleration signal 50 Hz, and skin temperature signal 1 Hz, and local real-time processing is realized through edge computing, with a delay of ≤100 ms.

[0030] In step 1, the PPG heart rate signal adopts the photoplethysmography method, a green LED light source is used to irradiate the pet skin (preferably the neck, chest and other areas with thin hair), the absorption change of hemoglobin in blood to light is used to capture the blood vessel pulsation, and the heart rate change is directly reflected, a 100 Hz acquisition frequency is selected, covering the common heart rate range of pets (such as the resting heart rate of small dogs 60-140 times / min, up to 240 times / min when exercising, corresponding to a frequency of 1-4 Hz), to ensure that the waveform details of each heartbeat are captured.

[0031] The three-axis acceleration signal acquires X, Y and Z three-axis motion data through a MEMS acceleration sensor, and a frequency of 50 Hz is sufficient to depict the dynamic behaviors of pets such as running, jumping and rolling, and is used to distinguish between motion state and resting state.

[0032] Skin temperature signal: the core body temperature of the pet is related to the heart rate, and a 1 Hz acquisition frequency can reflect the temperature trend, which helps to exclude heart rate abnormalities caused by infectious diseases.

[0033] Step 2, pre-processing of the real-time data stream, including band-pass filtering and noise reduction, motion artifact removal and signal standardization, to obtain clean signals.

[0034] Specifically, the motion artifact removal adopts an adaptive Kalman filter algorithm, and the formula is:

[0035]

[0036] wherein, is the filtered signal, z k is the original PPG signal, K kis the Kalman gain, H is the observation matrix.

[0037] Further, since the original PPG signal is easily affected by artifacts during pet movement, Kalman filtering is used to correct the signal, and the specific formula and algorithm are as follows:

[0038]

[0039] wherein, represents the estimated value, P(k) is the estimated error covariance, A and B are system matrices, Q is the process noise covariance, the motion intensity is calculated in real time through the three-axis acceleration signal, and the process noise covariance Q is dynamically adjusted (Q increases when the motion intensity increases, allowing greater state changes), ensuring the artifact removal effect.

[0040] Step 3, multi-time scale feature extraction is performed on the preprocessed PPG signal, the time scales include instantaneous features, short-term features and trend features, the extracted features include peak interval, heart rate variability, waveform slope and instantaneous heart rate change rate, and the calculation formula of the instantaneous heart rate change rate is:

[0041]

[0042] wherein, HR(t) is the instantaneous heart rate at time t.

[0043] Specifically, the extraction of multi-time scale features covers instantaneous features, short-term features and trend features, the time scales, types and physiological meanings of each feature are shown in the following table:

[0044]

[0045]

[0046] Through the combination of the above multi-time scale features, the differences of super high heart rate in instantaneous fluctuation, short-term stability and long-term change trend can be fully described.

[0047] Step 4, the activity intensity is calculated based on the three-axis acceleration signal, the pet state vector is constructed combining with the skin temperature signal, and the feature weight is dynamically adjusted, the pet state vector includes activity intensity, body temperature deviation and resting heart rate deviation, and is standardized through the following formula:

[0048]

[0049] wherein, A(t) is the acceleration module value at time t, T0 is the baseline body temperature, HR base is the mean value of the pet's historical resting heart rate.

[0050] Step 5, input the multi-scale features into the improved attention mechanism neural network, output the preliminary classification results and confidence of the super high heart rate.

[0051] Wherein, the improved attention mechanism neural network contains 3 parallel attention heads, corresponding to 0.1s, 1s, 5s time scales respectively, and the attention weight is dynamically adjusted by the pet state vector, and the calculation formula is:

[0052] s=Softmax(W s ·S+b s ) (Formula 5);

[0053] Wherein, a s is the weight of the s th time scale, W s , b s are training parameters, Softmax activation function is adopted, and the probability and confidence of "physiological" and "pathological" two categories are output.

[0054] Step 6, using support vector machine to classify the multi-scale features again, combining the confidence of step 5 to get the final classification result through the posterior probability fusion algorithm, and the formula of the posterior probability fusion algorithm is:

[0055] P(C) = ω1·P NN (C) + ω2·P SVM (C) (Formula 6);

[0056] Wherein, P(C) is the final classification probability, and C is "pathological" or "physiological", when P(C) > 0.5, it is determined as "pathological", P NN , P SVM are the output probability of neural network and support vector machine respectively, ω1, ω2 are the dynamic weight based on confidence.

[0057] Step 7, based on the pet historical health data and the current activity intensity, the threshold of super high heart rate anomaly is dynamically adjusted, if the final classification result is pathological and exceeds the threshold, real-time warning is triggered.

[0058] Further, in step 7, the dynamic threshold is updated by the following formula:

[0059] Th(t) = Th0·(1+β·A(t)) (Formula 7);

[0060] Wherein, Th0 is the baseline threshold of pet, β is the activity influence coefficient, and A(t) is the current activity intensity.

[0061] According to the duration of heart rate anomaly, different levels of warning are set, wherein, the real-time warning contains grading mechanism:

[0062] Primary warning: When the pathological high heart rate duration is less than 5 seconds, the device locally issues a vibration reminder;

[0063] Secondary warning: When the pathological high heart rate duration is greater than or equal to 5 seconds and less than 30 seconds, the vibration reminder pushes the warning information to the user's mobile phone APP at the same time;

[0064] Tertiary warning: When the pathological high heart rate duration is greater than or equal to 30 seconds, the emergency contact call is triggered + the positioning information is pushed, and the medical auxiliary data containing the heart rate waveform is sent synchronously, through real-time monitoring of the physiological signal changes of the pet, combining historical health data and current activity intensity, dynamically adjusting the heart rate threshold and judgment standard, to ensure timely response to abnormal heart rate conditions.

[0065] In summary: the abnormal detection method of high heart rate based on real-time signal processing proposed in the application, through the use of multi-modal physiological signal fusion strategy, synchronously collects PPG heart rate signal, three-axis acceleration signal and skin temperature signal, realizes local low-delay processing combined with edge computing, lays a data foundation for real-time detection, in the preprocessing stage, through band-pass filtering, adaptive Kalman filtering and standardization processing, effectively purifies the signal, ensures the accuracy of subsequent feature extraction; multi-time scale feature extraction covers instantaneous, short-term and trend features, fully captures the details and rules of heart rate fluctuations.

[0066] In addition, by introducing the pet state vector (including activity intensity, body temperature deviation degree, etc.), the feature weight is dynamically adjusted, so that the model is more suitable for the individual state difference of the pet, the improved attention mechanism neural network and support vector machine are used in the classification link, the posterior probability fusion algorithm is combined to improve the classification accuracy, effectively distinguishing between physiological and pathological high heart rate, the dynamic threshold mechanism is based on historical data and current activity intensity to update the judgment standard in real time, avoiding the limitations of fixed threshold, and the hierarchical warning mechanism matches the response intensity according to the duration of the abnormality, taking into account timeliness and rationality.

[0067] The whole process optimization from signal processing, feature extraction to classification decision in the application not only solves the problems of high false alarm rate and insufficient timeliness in the prior art, but also provides an efficient and reliable detection process for pet health monitoring through personalized dynamic adjustment and low-delay processing, which has important practical significance for early detection of potential diseases in pets and reduction of health risks.

[0068] The parts not involved in the application are the same as or can be realized by the prior art. Although embodiments of the application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, replacements and variations can be made to these embodiments without departing from the principles and spirit of the application, and the scope of the application is defined by the appended claims and their equivalents.

Claims

1. A method for detecting abnormal ultra-high heart rate based on real-time signal processing, characterized in that, The method comprises the following steps: Step 1, collecting multi-modal physiological signals in real time, including PPG heart rate signal, three-axis acceleration signal and skin temperature signal, and constructing real-time data stream; Step 2, pre-processing the real-time data stream, including band-pass filter noise reduction, motion artifact removal and signal standardization, to obtain clean signals; Step 3, multi-time scale feature extraction on the pre-processed PPG signal, the time scale including instantaneous feature, short-term feature and trend feature, the extracted features including peak interval, heart rate variability, waveform slope and instantaneous heart rate change rate; Step 4, calculating activity intensity based on three-axis acceleration signal, combining skin temperature signal to construct pet state vector, and dynamically adjusting feature weight; Step 5, inputting multi-scale features into an improved attention mechanism neural network, and outputting preliminary classification results and confidence of hyper-high heart rate; Step 6, using support vector machine to perform secondary classification on the multi-scale features, combining the confidence of step 5 to obtain the final classification result through posterior probability fusion algorithm; Step 7, dynamically adjusting the hyper-high heart rate abnormality judgment threshold based on pet historical health data and current activity intensity, if the final classification result is pathological and exceeds the threshold, triggering real-time early warning.

2. The real-time signal processing based ultra-high heart rate anomaly detection method according to claim 1, characterized in that: In step 1, the collection frequency of the multi-modal physiological signals is: PPG signal 100Hz, three-axis acceleration signal 50Hz, and skin temperature signal 1Hz, and local real-time processing is realized through edge computing with a delay of ≤100ms.

3. The real-time signal processing based ultra-high heart rate anomaly detection method according to claim 1, characterized in that: In step 2, the motion artifact removal adopts an adaptive Kalman filtering algorithm, and the formula is as follows: Wherein, is the filtered signal, z k is the original PPG signal, K k is the Kalman gain, and H is an observation matrix.

4. The real-time signal processing based ultra-high heart rate anomaly detection method according to claim 1, characterized in that: In step 3, the formula for calculating the instantaneous heart rate change rate is: wherein HR(t) is the instantaneous heart rate at time t.

5. The real-time signal processing based super high heart rate anomaly detection method according to claim 1, characterized in that: In step 4, the pet state vector includes activity intensity, body temperature deviation, and resting heart rate deviation, which are normalized by the following formulas: where A(t) is the acceleration module value at time t, T0 is the baseline body temperature, HR base is the mean value of the pet's historical resting heart rate.

6. The real-time signal processing based super high heart rate anomaly detection method according to claim 1, characterized in that: In step 5, the improved attention mechanism neural network comprises 3 parallel attention heads corresponding to 0.1s, 1s and 5s time scales respectively, and the attention weight is dynamically adjusted through the pet state vector, and the calculation formula is: a s =Softmax(W s ·S+b s ), wherein a s is the weight of the s-th time scale, W s and b s are training parameters.

7. The real-time signal processing based ultra-high heart rate anomaly detection method according to claim 1, characterized in that: The formula of the posterior probability fusion algorithm in step 6 is: P(C) = ω1·P NN (C) + ω2·P SVM (C), wherein P(C) is a final classification probability, P NN , P SVM are output probabilities of the neural network and the support vector machine respectively, and ω1, ω2 are dynamic weights based on confidence.

8. The real-time signal processing based super high heart rate anomaly detection method according to claim 1, characterized in that: In step 7, the dynamic threshold is updated by the following formula: Th(t)=Th0·(1+β·A(t)), where Th0 is the pet baseline threshold, β is the activity influence coefficient, and A(t) is the current activity intensity.

9. The real-time signal processing based super high heart rate anomaly detection method according to claim 1, characterized in that: In step 7, the real-time warning includes a grading mechanism: First-level warning: when the duration of pathological hyper-high heart rate is <5 seconds, the device locally issues a vibration reminder; Second-level warning: when the duration of pathological hyper-high heart rate is ≥5 seconds and <30 seconds, the vibration reminder pushes warning information to the user's mobile phone APP at the same time; Third-level warning: when the duration of pathological hyper-high heart rate is ≥30 seconds, triggering emergency contact call + location information push, and synchronously sending medical auxiliary data containing heart rate waveform.

10. The real-time signal processing based super high heart rate anomaly detection device of claim 1, wherein, A device comprising a processor, a memory and a system bus, the processor and the memory being connected through the system bus, the memory being used to store one or more programs, the one or more programs including instructions, the instructions being executed by the processor to make the processor execute the detection method of any one of claims 1-9.

Images