Intelligent watch arrhythmia early warning method and system based on deep learning

Abstract

The invention discloses a smart watch arrhythmia early warning method and system based on deep learning. The method comprises the steps that firstly, PPG signals are collected through a photoelectric volume pulse wave sensor, and denoising and standardization processing are conducted through an adaptive filtering algorithm; then, the heart beat position is recognized through a wavelet transform peak value detection algorithm, and a heart rate variability feature sequence is obtained; converting the one-dimensional heart rate variability feature sequence into a two-dimensional heart rhythm image by using an improved Gramian angular field transformation algorithm; and training an arrhythmia classification model by adopting a ResNet-50 convolutional neural network to realize automatic identification of different types of arrhythmia. And finally, continuous monitoring and timely alarming are realized through a sliding window technology and an early warning mechanism. The key technical problems of low detection precision, poor real-time performance, high equipment cost, poor user experience and the like in the existing arrhythmia detection technology are effectively solved, and an innovative solution is provided for heart rhythm health monitoring of the intelligent wearable equipment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of arrhythmia warning, and particularly relates to an intelligent watch arrhythmia warning method and system based on deep learning. BACKGROUND

[0002] Arrhythmia refers to the abnormality of the origin site, frequency, rhythm, conduction velocity and excitation sequence of cardiac impulse, and is one of the most complex cardiovascular diseases. It has great changes, wide range and many types, and seriously threatens human health. With the acceleration of modern life rhythm and the continuous improvement of population aging, the incidence of arrhythmia shows a rising trend year by year. If dangerous arrhythmias such as atrial fibrillation and ventricular tachycardia are not discovered and treated in time, they may lead to serious consequences such as sudden cardiac death. Therefore, it has important clinical significance and social value to realize continuous monitoring and early warning of arrhythmia.

[0003] In recent years, with the rapid development of sensor technology, communication technology and artificial intelligence algorithm, intelligent wearable devices have been widely used in the field of heart rhythm monitoring. As the most representative wearable heart rhythm monitoring device, smart watches have gradually become an important tool for personal health management and disease prevention due to their convenience, continuous monitoring and real-time feedback. The smart watches on the market mainly use photoplethysmography (PPG) sensors for heart rate detection. Compared with traditional electrocardiogram monitoring methods, PPG technology has the advantages of non-invasiveness, low cost and easy integration, which lays a technical foundation for the popularization of large-scale heart rhythm monitoring.

[0004] However, the existing intelligent wearable devices still have many technical defects and application limitations in arrhythmia monitoring. Traditional arrhythmia detection methods are mainly based on simple threshold judgment or basic statistical feature analysis, and the recognition accuracy of complex arrhythmia types is low, which is prone to false positives and false negatives. Especially during the user's daily activities, factors such as motion artifacts, skin contact changes and environmental light interference can seriously affect the quality of PPG signals, leading to a significant decline in the reliability of arrhythmia detection.

[0005] The prior art represented by patent CN202411576388.X does propose certain solutions in environmental interference identification, but still has significant technical limitations. This technology mainly relies on an interference marking method based on location information, determines environmental interference factors through active positioning signals in medical sites, and marks electrocardiograms when interference is detected. However, this method has the following key defects: First, this method can only identify known interference sources in specific places, lacks adaptability to non-preset interference environments and dynamically changing interference factors, limiting its application effect in complex real environments. Second, the simple interference marking strategy cannot fundamentally improve the accuracy of arrhythmia detection, only identifies data that may be interfered with, and cannot effectively extract and utilize effective arrhythmia feature information. Finally, this technology lacks intelligent signal processing and pattern recognition algorithms, cannot achieve accurate classification and identification of different types of arrhythmias, and has limited practical value in clinical applications.

[0006] In addition, the prior art also has obvious deficiencies in feature extraction and pattern recognition. Traditional methods mostly use single-dimensional features in time or frequency domains, which cannot fully exploit the complex physiological information contained in PPG signals, and have limited feature representation capability. At the same time, classification methods based on traditional machine learning algorithms perform poorly when faced with high-dimensional complex data, making it difficult to handle non-linear feature relationships and time sequence dependency problems in arrhythmia detection. In terms of real-time performance, existing systems often require long data acquisition and processing times, which cannot meet the timeliness requirements of arrhythmia emergency warning and may miss the best opportunity for medical intervention.

[0007] In summary, the existing intelligent wearable device arrhythmia monitoring technology has significant defects in signal processing algorithms, feature extraction methods, pattern recognition accuracy, and real-time warning capabilities, and there is an urgent need to develop more intelligent, high-precision, and robust arrhythmia warning technology to meet the growing demand for personal health monitoring and clinical application requirements. SUMMARY

[0008] Therefore, the present application provides an intelligent watch arrhythmia warning method based on deep learning, aiming to solve the technical problems of low detection accuracy, poor real-time performance, high device cost, and poor user experience in existing arrhythmia detection technology.

[0009] To achieve the above-mentioned purpose, the present application provides an intelligent watch arrhythmia warning method based on deep learning, comprising the following steps: S1: Collecting original PPG signals through the built-in photoelectric plethysmogram sensor of the intelligent watch, denoising and standardizing the collected original PPG signals, and obtaining standardized PPG signals; S2: A peak detection algorithm based on wavelet transform is used to identify the heart beat positions contained in the standardized PPG signal, and after abnormal value detection and smoothing processing, a heart rate variability feature sequence is obtained; S3: The heart rate variability feature sequence is subjected to minimum-maximum normalization processing, and then an improved Gramian angle field transformation algorithm is used to convert the one-dimensional heart rate variability feature sequence into a two-dimensional heart rhythm image matrix; the improved Gramian angle field transformation algorithm is to perform polar coordinate transformation on the heart rate variability feature sequence after minimum-maximum normalization processing, and convert the standardized sequence into a point representation on a unit circle; the improved Gramian angle field transformation algorithm calculates the Gramian angle and the angle difference field between any two sequence points in the transformed sequence, and captures the similarity and difference features of the time series data respectively; finally, the two kinds of angle field information are averaged to generate a two-dimensional heart rhythm image matrix, realizing effective conversion from one-dimensional time series to two-dimensional image; S4: A ResNet-50 convolutional neural network architecture is used to train an arrhythmia classification model through supervised learning, realizing automatic identification and classification of different types of arrhythmia; S5: Through the sliding window technology and the early warning mechanism, the continuous monitoring of the user's heart rhythm state and the timely alarm of the abnormal situation are realized.

[0010] As a further improved method of the application: Optionally, in the S1 step, the original PPG signal is collected by the photoelectric plethysmogram sensor built in the smart watch, and a self-adaptive filtering algorithm is used to denoise and standardize the collected original PPG signal to obtain a standardized PPG signal, comprising: During the PPG signal collection process, the sensor collects the PPG signal at a fixed sampling frequency The photoelectric detection is performed on the wrist part of the user to obtain the original PPG signal , wherein represents the sampling time point; The collected original PPG signal is subjected to band-pass filtering processing through a band-pass filter to filter out low-frequency drift and high-frequency noise interference to obtain a preliminary clean signal ; the cut-off frequency of the band-pass filter is designed according to the normal heart rate range of the human body to ensure that the effective heart rate related frequency components are retained; An adaptive noise cancellation algorithm is used to further process the preliminary clean signal to eliminate artifact interference and obtain a denoised PPG signal ; The denoised PPG signal is subjected to amplitude normalization to obtain a standardized PPG signal .

[0011] Optionally, the S2 step employs a peak detection algorithm based on wavelet transform to identify the heart beat positions contained in the normalized PPG signal, including: performing continuous wavelet transform processing on the normalized PPG signal to obtain a wavelet transform result , wherein denotes a scale parameter for controlling the frequency characteristics of the wavelet, denotes a shift parameter for determining the position of the wavelet on the time axis; searching for local extreme points in the wavelet transform result , the local extreme points corresponding to the heart beat peak positions in the normalized PPG signal; by setting a threshold condition, false peaks caused by noise or abnormal fluctuations are filtered out to obtain a heart beat time sequence , wherein denotes the time point of the n-th heart beat, denotes the total number of detected heart beats; based on the detected adjacent heart beat time points, the time intervals between consecutive heart beats are calculated to form an RR interval sequence , wherein denotes the duration of the n-th RR interval; performing outlier detection and smoothing processing on the extracted RR interval sequence to eliminate extreme values caused by detection errors or physiological abnormalities, and obtaining a heart rate variability feature sequence , wherein denotes the n-th valid RR interval value in the heart rate variability feature sequence, denotes the number of valid RR interval values after outlier detection and smoothing processing. Optionally, the S3 step performs minimum-maximum normalization processing on the heart rate variability feature sequence, and then uses an improved Gramian angle field transformation algorithm to convert the one-dimensional heart rate variability feature sequence into a two-dimensional heart rhythm image; the improved Gramian angle field transformation algorithm is to map the heart rate variability feature sequence to the interval of 0 to 1 through minimum-maximum normalization processing, and then perform polar coordinate transformation to convert the normalized sequence into a point representation on the unit circle; this algorithm calculates the Gramian angle and the angle difference field between any two sequence points in the sequence, respectively capturing the similarity and difference features of the time series data; finally, the average value of the two kinds of angle field information is taken to generate a two-dimensional heart rhythm image matrix, realizing effective conversion from one-dimensional time series to two-dimensional image, including: converting the heart rate variability feature sequence

[0012] ​​​The minimum-maximum normalization processing is performed to map the numerical range to the interval [0, 1], to obtain a normalized heart rate variability feature sequence ; The polar coordinate mapping is performed on the normalized heart rate variability feature sequence to convert the normalized heart rate variability feature sequence into a point on a unit circle; The improved Gramian angle field transformation algorithm is used to calculate the angle relationship between any two normalized effective RR interval values in the normalized heart rate variability feature sequence, including the Gramian angle and field and the Gramian angle difference field; The average values of the Gramian angle and field and the Gramian angle difference field are taken as a heart rhythm image matrix Output, wherein and represent the row and column coordinates of the image, respectively.

[0013] Optionally, the ResNet-50 convolutional neural network architecture is used in the S4 step, and the arrhythmia classification model is trained by supervised learning to realize automatic identification and classification of different types of arrhythmia, including: The input layer of the ResNet-50 network is constructed, and the heart rhythm image matrix is adjusted to a standard input size of 224x224x3, and the single-channel heart rhythm image is expanded to a three-channel RGB format by an image replication method; The pre-trained ResNet-50 model parameters on the ImageNet dataset are loaded; all convolutional layer parameters of the ResNet-50 network are frozen, and only the fully connected layer is fine-tuned for training; The last fully connected layer of the ResNet-50 model is replaced, and a new classification head structure including a global average pooling layer, a Dropout regularization layer and a fully connected output layer is designed; the number of output layer neurons is set to correspond to the number of arrhythmia types, and a softmax activation function is used to output a probability distribution vector for each class , wherein represents the predicted probability of the th arrhythmia class, represents the total number of arrhythmia classes; The labeled arrhythmia image dataset is used for network fine-tuning training, the cross-entropy loss function is used to calculate the error between the predicted results and the true labels, the Adam optimizer is used for parameter update, and iterations of training are performed to obtain a trained arrhythmia classification model .

[0014] Optionally, the S5 step is realized by the sliding window technique and the early warning mechanism, continuous monitoring of the user's heart rhythm state and timely alarm of abnormal conditions, including: A real-time data stream processing mechanism is established to continuously receive new PPG signal data collected by the smart watch, and the new PPG signal data is processed according to a fixed time interval The new data is processed in segments. The newly received PPG data segment is converted in real time according to the processing flow of steps S1 to S3 to generate a corresponding heart rhythm image matrix ; The real-time generated heart rhythm image matrix is input into the trained arrhythmia classification model for inference calculation to obtain an arrhythmia prediction probability vector of the current time period , wherein represents the probability value of the detected type arrhythmia. A multi-level early warning judgment mechanism is designed to compare and judge the maximum probability value in the prediction probability vector and the preset risk threshold value; when a high-risk arrhythmia type is detected and the probability exceeds the safety threshold value, the system automatically generates an early warning signal to immediately notify the user through the vibration, sound or screen prompt of the smart watch.

[0015] The application also discloses an intelligent watch arrhythmia early warning system based on deep learning, comprising: A signal acquisition and standardization module: an original PPG signal is acquired by a photoelectric plethysmogram sensor built in the smart watch, and an adaptive filtering algorithm is used to denoise and standardize the acquired original PPG signal; A heart rate variability feature extraction module: a peak detection algorithm based on wavelet transform is used to identify the heart beat position contained in the standardized PPG signal, and a heart rate variability feature sequence is obtained after abnormal value detection and smoothing processing; An image conversion module: the heart rate variability feature sequence is subjected to minimum-maximum normalization processing, and then an improved Gramian angle field transformation algorithm is used to convert the one-dimensional heart rate variability feature sequence into a two-dimensional heart rhythm image; A classification module: a ResNet-50 convolutional neural network architecture is used to train an arrhythmia classification model through supervised learning to realize automatic identification and classification of different types of arrhythmia; A monitoring module: the sliding window technique and the early warning mechanism are used to realize continuous monitoring of the user's heart rhythm state and timely alarm of abnormal conditions.

[0016] Compared with the prior art, the application has at least the following beneficial effects: This invention combines Gramian angles and fields with Gramian angle difference fields. Through polar coordinate mapping and angle relationship calculation, it simultaneously encodes the temporal dependence, periodic features, and local variation patterns of a one-dimensional heart rate variability time series into different dimensions of a two-dimensional image. Compared to traditional direct time-domain or frequency-domain feature extraction methods, this algorithm can preserve the global temporal structure of the original signal while highlighting local anomalous changes, effectively solving the feature loss problem in PPG signals caused by motion artifacts, contact changes, and other factors.

[0017] This invention employs a ResNet-50 network pre-trained on the ImageNet dataset as a feature extractor. By freezing all convolutional layer parameters and fine-tuning only the fully connected layers, it effectively utilizes the general feature representation capabilities learned from large-scale natural image datasets. This transfer learning method not only significantly reduces the amount of arrhythmia-labeled data required for model training, lowering data collection costs, but also significantly shortens training time and improves the model's generalization ability.

[0018] The multi-level early warning judgment mechanism designed in this invention intelligently compares the maximum probability value in the predicted probability vector with a preset risk threshold. This not only accurately identifies high-risk arrhythmia events but also effectively reduces the false alarm rate, avoiding the impact of frequent false alarms on user experience. When a dangerous arrhythmia is detected and its probability exceeds the safety threshold, the system can immediately issue an early warning to the user simultaneously through multiple sensory channels, ensuring timely response at critical moments. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a method for early warning of cardiac arrhythmia in a smartwatch based on deep learning, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the original PPG according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a heart rhythm image according to an embodiment of the present invention. Detailed Implementation

[0020] The present invention will be further described below with reference to the accompanying drawings, but this is not intended to limit the present invention in any way. Any modifications or substitutions made based on the teachings of the present invention shall fall within the protection scope of the present invention.

[0021] Example 1: A deep learning-based method for early warning of cardiac arrhythmia in smartwatches, such as... Figure 1 As shown, it includes the following steps: S1: The raw PPG signal is acquired through the photoplethysmography (PPG) sensor built into the smartwatch, such as... Figure 2As shown, the acquired raw PPG signal is denoised and standardized to obtain a standardized PPG signal. In this embodiment, the adaptive filtering algorithm is the minimum mean square error adaptive filtering algorithm. During the PPG signal acquisition process, the sensor operates at a fixed sampling frequency. Photoelectric detection is performed on the user's wrist to obtain the raw PPG signal. ,in Indicates the sampling time point; For the acquired raw PPG signal Bandpass filtering is performed to remove low-frequency drift and high-frequency noise interference, resulting in a preliminary clean signal. The cutoff frequency of the bandpass filter is designed according to the normal human heart rate range to ensure that effective heart rate-related frequency components are retained. In this embodiment, the cutoff frequency of the bandpass filter is 2Hz. An adaptive noise cancellation algorithm is used to further process the initial cleaning signal to eliminate artifact interference caused by factors such as user movement and changes in skin contact, thus obtaining a denoised PPG signal. In this embodiment, the adaptive noise cancellation algorithm is a recursive least squares algorithm; For the denoised PPG signal Amplitude normalization is performed to obtain a standardized PPG signal. .

[0022] S2: A wavelet transform-based peak detection algorithm is used to identify the heartbeat positions contained in the standardized PPG signal. For standardized PPG signals Perform continuous wavelet transform processing to obtain the wavelet transform result. ,in This represents the scaling parameter, used to control the frequency characteristics of the wavelet. This represents the displacement parameter, used to determine the wavelet's position on the time axis; In wavelet transform results The search involves identifying local extrema, which correspond to the peak heartbeat position in the standardized PPG signal. Heartbeat events are then selected based on a predefined threshold condition; in this embodiment, the threshold condition is... By eliminating spurious peaks caused by noise or abnormal fluctuations, the heartbeat time series is obtained. ,in Indicates the first The timing of the second heartbeat This indicates the total number of heartbeats detected; Based on the detected adjacent heartbeat time points, the time interval between consecutive heartbeats is calculated to form an RR interval sequence. ,in denotes the duration of the i-th RR interval; outlier detection and smoothing processing are performed on the extracted RR interval sequence to eliminate extreme values caused by detection errors or physiological abnormalities, and in this embodiment, the outlier detection and smoothing processing are a four-quartile distance-based outlier monitoring algorithm and a median filter smoothing algorithm, respectively, to obtain a heart rate variability feature sequence wherein denotes the i-th valid RR interval value in the heart rate variability feature sequence, denotes the number of valid RR interval values after the outlier detection and smoothing processing.

[0023] This step realizes high-precision identification of the PPG signal heart beat position by using continuous wavelet transform technology, which has significant advantages compared to traditional differential threshold or template matching methods. Wavelet transform can analyze signal characteristics in both time and frequency domains, effectively capturing local features of heart beat peaks through multi-scale analysis, and still maintaining stable detection performance even in the presence of signal baseline drift or morphological changes. In particular, when the user is moving or the signal quality is poor, the multi-resolution characteristics of wavelet transform can identify heart beat features at different scales, greatly improving the robustness and accuracy of peak detection.

[0024] S3: performing minimum-maximum normalization processing on the heart rate variability feature sequence, and then using an improved Gramian angle field transformation algorithm to convert the one-dimensional heart rate variability feature sequence into a two-dimensional heart rhythm image matrix; the improved Gramian angle field transformation algorithm is a polar coordinate transformation performed on the minimum-maximum normalized heart rate variability feature sequence to convert the standardized sequence into a point representation on a unit circle; the improved Gramian angle field transformation algorithm calculates the Gramian angle and the angle difference field between any two sequence points in the transformed sequence, respectively capturing the similarity and difference features of the time series data; finally, the two kinds of angle field information are averaged to generate a two-dimensional heart rhythm image matrix, realizing effective conversion from one-dimensional time series to two-dimensional images; performing minimum-maximum normalization processing on the heart rate variability feature sequence wherein denotes the i-th valid RR interval value in the heart rate variability feature sequence, denotes the number of valid RR interval values after the outlier detection and smoothing processing. ​​​​wherein denotes the corresponding polar angle; The angle relationship between any two standardized effective RR interval values in the standardized heart rate variability feature sequence is calculated using the improved Gramian angle field transformation algorithm, and the angle relationship includes the Gramian angle and field and the Gramian angle difference field, specifically: wherein, and denote the index positions of the standardized heart rate variability feature sequence; is the value of the Gramian angle and field at ; is the value of the Gramian angle difference field at ; and denote the standardized th and th effective RR interval values in the heart rate variability feature sequence; The average values of the Gramian angle and field and the Gramian angle difference field are taken as the heart rhythm image matrix output, as shown in Figure 3 , wherein and denote the row and column coordinates of the image, respectively.

[0025] This step realizes intelligent conversion from one-dimensional time series data to two-dimensional images by improving the Gramian angle field transformation algorithm, providing rich feature representation for deep learning models. Compared with traditional direct time-frequency analysis or simple numerical feature extraction methods, the Gramian angle field transformation can completely maintain the time series dependence and local change patterns of the original time series, and at the same time encode these complex time series information into the spatial structure of the two-dimensional image. This conversion not only preserves the global trend information in the heart rate variability sequence, but more importantly, can highlight local abnormal changes and periodic patterns, providing an intuitive and information-rich input format for subsequent convolutional neural network recognition.

[0026] S4: Adopting ResNet-50 convolutional neural network architecture, training arrhythmia classification model through supervised learning, realizing automatic recognition and classification of different types of arrhythmia: The input layer of the ResNet-50 network is constructed, and the heart rhythm image matrix is adjusted to the standard input size of 224x224x3, and the single-channel heart rhythm image is expanded to the three-channel RGB format through the image replication method; Load the pre-trained ResNet-50 model parameters on the ImageNet dataset; freeze all convolutional layer parameters of the ResNet-50 network, and only fine-tune the training of the fully connected layer; Replace the last fully connected layer of the ResNet-50 model, and design a new classification head structure including a global average pooling layer, a Dropout regularization layer and a fully connected output layer; set the number of output layer neurons to correspond to the number of arrhythmia types, and use the softmax activation function to output the probability distribution vector of each class , wherein represents the prediction probability of the th arrhythmia class, represents the total number of arrhythmia classes, and in the present embodiment, the arrhythmia classes involved include no abnormality, atrial fibrillation, ventricular fibrillation and ventricular tachycardia; Use the labeled arrhythmia image dataset for network fine-tuning training, use the cross-entropy loss function to calculate the error between the predicted result and the true label, and use the Adam optimizer to update the parameters, and perform times of iterative training to obtain the trained arrhythmia classification model , in the present embodiment ; For the training data of class imbalance, the system uses the average loss function of Focal loss and cross-entropy loss to improve the recognition accuracy of the minority classes.

[0027] This step uses the ResNet-50 deep convolutional neural network architecture to fully utilize the strong advantages of deep residual network in complex pattern recognition tasks. The residual connection structure of ResNet-50 effectively solves the gradient vanishing problem in deep network training, so that the network can learn more rich and abstract feature representations. Compared with traditional shallow machine learning methods or simple neural networks, ResNet-50 has strong non-linear modeling ability and can capture complex spatial patterns and texture features in the arrhythmia image.

[0028] S5: Through the sliding window technology and the early warning mechanism, the continuous monitoring of the user's heart rhythm state and the timely alarm of abnormal situations are realized: Establish a real-time data stream processing mechanism to continuously receive new PPG signal data collected by the smart watch, and process the data according to a fixed time interval Process the new data in segments, and in the present embodiment seconds; Convert the newly received PPG data segment in real time according to the processing procedure of steps S1 to S3 to generate the corresponding arrhythmia image matrix ; Convert the real-time generated arrhythmia image matrix Input to the trained arrhythmia classification model Inference computation is performed in the current time period to obtain an arrhythmia prediction probability vector , wherein represents the current detected arrhythmia type , and the probability value of the arrhythmia type; when the single prediction result confidence is low, the system adopts a sliding window multi-frame fusion decision mechanism, combines the historical detection results for weighted average, improves the stability and reliability of the prediction, and the multi-frame refers to the previous 5 frames of the current frame; A multi-level warning judgment mechanism is designed, and the maximum probability value in the prediction probability vector is compared with a preset risk threshold value for comparison and judgment; when a high-risk arrhythmia type is detected and the probability exceeds the safety threshold value, the system automatically generates a warning signal, immediately notifies the user through vibration, sound or screen prompt of the smart watch, and in the embodiment, the safety threshold value is 0.8.

[0029] Embodiment 2: The application also discloses an intelligent watch arrhythmia warning system based on deep learning, which comprises the following five modules: Signal acquisition and standardization module: the original PPG signal is acquired through the photoelectric plethysmogram sensor built in the intelligent watch, and the adaptive filtering algorithm is used to denoise and standardize the acquired original PPG signal; Heart rate variability feature extraction module: a peak detection algorithm based on wavelet transform is used to identify the heart beat position contained in the standardized PPG signal, and the heart rate variability feature sequence is obtained after abnormal value detection and smoothing processing; Image conversion module: the heart rate variability feature sequence is subjected to minimum-maximum normalization processing, and then an improved Gramian angle field transformation algorithm is used to convert the one-dimensional heart rate variability feature sequence into a two-dimensional heart rhythm image; Classification module: a ResNet-50 convolutional neural network architecture is used, an arrhythmia classification model is trained through supervised learning, and automatic identification and classification of different types of arrhythmia are realized; Monitoring module: through the sliding window technology and the warning mechanism, continuous monitoring of the user's heart rhythm state and timely alarm of abnormal conditions are realized.

[0030] It should be noted that the above-mentioned embodiment serial numbers of the present application are only for description, not representing the advantages and disadvantages of the embodiments. Also, the terms "comprising", "containing" or any other variants thereof in this document are intended to cover the non-exclusive inclusion, so that the process, device, article or method including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, device, article or method. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0031] Through the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be realized by means of software and necessary general hardware platforms, of course, they can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a plurality of instructions for making a terminal device (which can be a mobile phone, computer, server, or network device, etc.) execute the methods described in various embodiments of the present application.

[0032] The above is only the preferred embodiment of the present application, and does not limit the patent scope of the present application, and any equivalent structure or equivalent flow transformation made by using the content of the present application specification and drawings, or directly or indirectly applied to other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A deep learning-based smart watch arrhythmia warning method, characterized in that, The method comprises the following steps: S1: collecting original PPG signals through the photoelectric plethysmogram sensor built in the smart watch, denoising and standardizing the collected original PPG signals, and obtaining standardized PPG signals; S2: identifying the heart beat positions contained in the standardized PPG signals by using a peak detection algorithm based on wavelet transform, and obtaining a heart rate variability feature sequence after abnormal value detection and smoothing processing; S3: performing minimum-maximum normalization processing on the heart rate variability feature sequence, and then using an improved Gramian angle field transformation algorithm to convert the one-dimensional heart rate variability feature sequence into a two-dimensional heart rhythm image matrix; the improved Gramian angle field transformation algorithm is to perform polar coordinate transformation on the heart rate variability feature sequence after minimum-maximum normalization processing, and convert the standardized sequence into a point representation on a unit circle; the improved Gramian angle field transformation algorithm calculates the Gramian angle and the angle difference field between any two sequence points in the transformed sequence, and captures the similarity and difference features of the time series data respectively; finally, the two kinds of angle field information are averaged and fused to generate a two-dimensional heart rhythm image matrix, realizing effective conversion from one-dimensional time series to two-dimensional image; S4: training an arrhythmia classification model by using a ResNet-50 convolutional neural network architecture through a supervised learning method, and realizing automatic identification and classification of different types of arrhythmia; S5: realizing continuous monitoring of the heart rhythm state of the user and timely alarm of abnormal conditions by using a sliding window technology and a warning mechanism. 2.The deep learning-based smart watch arrhythmia warning method of claim 1, wherein, The step S1 comprises: In the PPG signal acquisition process, the sensor acquires the PPG signal at a fixed sampling frequency The PPG signal is obtained by photoelectric detection of the wrist of the user wherein denotes a sampling time point; The original PPG signal collected The band-pass filtering process is performed by a band-pass filter to filter out low-frequency drift and high-frequency noise interference to obtain a preliminary clean signal The cut-off frequency of the band-pass filter is designed according to the normal heart rate range of human body to ensure that the effective heart rate related frequency components are retained; The adaptive noise cancellation algorithm is used to further process the preliminary cleaning signal, eliminate the artifact interference, and obtain the de-noised PPG signal ; the denoised PPG signal performing amplitude normalization to obtain a normalized PPG signal . 3.The deep learning based smart watch arrhythmia warning method of claim 2, wherein, The step S2 comprises: standardized PPG signal performing continuous wavelet transform processing to obtain a wavelet transform result wherein denotes a scale parameter for controlling the frequency characteristic of the wavelet, denotes a shift parameter for determining the position of the wavelet on the time axis; In wavelet transform results The system searches for local extrema, which correspond to the peak positions of heartbeats in the standardized PPG signal. By using predefined threshold conditions, heartbeat events are selected, and spurious peaks caused by noise or abnormal fluctuations are eliminated, resulting in a heartbeat time series. ,in Indicates the first The timing of the second heartbeat This indicates the total number of heartbeats detected; Based on the detected adjacent heart beat time points, the time intervals between consecutive heart beats are calculated, forming a sequence of RR intervals wherein denotes the duration of the i-th RR interval th RR interval The extracted RR interval sequence is subjected to outlier detection and smoothing processing to eliminate extreme values caused by detection errors or physiological abnormalities, and a heart rate variability feature sequence is obtained wherein represents the i-th valid RR interval value in the heart rate variability feature sequence, represents the number of valid RR interval values after outlier detection and smoothing processing.​ 4.The deep learning-based smart watch arrhythmia warning method of claim 3, wherein, The step S3 comprises: a sequence of heart rate variability features a sequence of normalized heart rate variability features ; performing polar coordinate mapping on the standardized heart rate variability feature sequence, and converting the standardized heart rate variability feature sequence into a point on a unit circle; using an improved Gramian angle field transformation algorithm to calculate the angle relationship between any two standardized effective RR interval values in the standardized heart rate variability feature sequence, the angle relationship comprising a Gramian angle and field and a Gramian angle difference field; averaging the gramian angle and field and the gramian angle difference field as the electrocardiogram image matrix output, wherein and denote the row and column coordinates of the image, respectively. 5.The deep learning-based smart watch arrhythmia warning method of claim 4, wherein, The step S4 comprises: The input layer of the ResNet-50 network is constructed, and a heart rhythm image matrix is input into the input layer The single-channel heart rhythm image is expanded into a three-channel RGB format through an image replication method, and is adjusted to a standard input size of 224x224x3. loading the pre-trained ResNet-50 model parameters on the ImageNet dataset; freezing all convolutional layer parameters of the ResNet-50 network, and only fine-tuning the training of the fully connected layer; The last fully connected layer of the ResNet-50 model is replaced, and a new classification head structure is designed, including a global average pooling layer, a Dropout regularization layer and a fully connected output layer; the number of neurons in the output layer is set to correspond to the number of arrhythmia types, and a softmax activation function is used to output a probability distribution vector of each category wherein represents the prediction probability of the i-th arrhythmia, represents the total number of arrhythmia categories The network fine-tuning training is performed on the labeled arrhythmia image dataset, an error between a prediction result and a true label is calculated using a cross-entropy loss function, and parameter updating is performed through an Adam optimizer The second iteration training is performed, and an arrhythmia classification model trained is obtained . 6.The deep learning-based smart watch arrhythmia warning method of claim 5, wherein, The step S5 comprises: A real-time data stream processing mechanism is established to continuously receive new PPG signal data collected by the smart watch, and the new PPG signal data is segmented according to a fixed time interval The new data is segmented for processing. The newly received PPG data segment is converted in real time according to the processing flow of steps S1 to S3 to generate a corresponding heart rate image matrix ; a real-time generated heart rhythm image matrix to a trained arrhythmia classification model a current time period wherein represents a probability value of a current detected type of arrhythmia A multi-stage early warning judgment mechanism is designed, and the maximum probability value in the prediction probability vector is compared with a preset risk threshold value for judgment; when a high-risk arrhythmia type is detected and the probability exceeds the safety threshold value, the system automatically generates an early warning signal, which immediately notifies the user in the form of vibration, sound or screen prompt of the smart watch.

7. A deep learning based smart watch arrhythmia warning system, characterized in that, It comprises: a signal acquisition and standardization module: collecting original PPG signals through the photoelectric plethysmogram sensor built in the smart watch, and denoising and standardizing the collected original PPG signals by using an adaptive filtering algorithm; a heart rate variability feature extraction module: identifying the heart beat positions contained in the standardized PPG signals by using a peak detection algorithm based on wavelet transform, and obtaining a heart rate variability feature sequence after abnormal value detection and smoothing processing; an image conversion module: performing minimum-maximum normalization processing on the heart rate variability feature sequence, and then using an improved Gramian angle field transformation algorithm to convert the one-dimensional heart rate variability feature sequence into a two-dimensional heart rhythm image. The classification module adopts a ResNet-50 convolutional neural network architecture, and a heart arrhythmia classification model is trained through supervised learning, so as to realize automatic identification and classification of different types of heart arrhythmia; The monitoring module realizes continuous monitoring of the heart rhythm state of the user and timely alarm of abnormal conditions through a sliding window technology and an early warning mechanism. To realize the intelligent watch heart arrhythmia early warning method based on deep learning in any one of claims 1-6.

Images