Wearable health equipment large model electrocardiogram monitoring system and method based on BSS-CNN-RF-DT-LLM fusion technology

Abstract

The invention discloses a large-model wearable electrocardiogram monitoring system and method based on a BSS-CNN-RF-DT-LLM fusion technology. The system comprises hardware equipment and a software platform. The hardware equipment integrates various core algorithm chips and sensing, transmission and interaction modules, and the software platform comprises a plurality of knowledge bases, a plurality of intelligent algorithm systems and a plurality of electrocardiosignal monitoring systems. According to the method, electrocardiosignals are collected through a wearable device, artifacts are removed through a BSS algorithm, features are extracted through a CNN algorithm, a classification model is constructed through an RF-DT algorithm, a professional knowledge report is generated in combination with an LLM algorithm and a health knowledge base, and the professional knowledge report is visually displayed in a partition mode on a display screen. All-dimensional monitoring of multiple electrocardiogram indexes and three-level evaluation of'normal-abnormity-danger 'are achieved, the anti-interference capacity is high, diagnosis is accurate, active monitoring and historical data comparison are supported, operation is convenient and fast, and a reliable solution is provided for daily health management and chronic disease tracking.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of wearable devices, and particularly relates to a wearable health device large model electrocardio monitoring system and method based on BSS-CNN-RF-DT-LLM fusion technology. BACKGROUND

[0002] With the aggravation of population aging and the increasing incidence of cardiovascular diseases, real-time electrocardio monitoring has become a key means to prevent sudden cardiac events and improve health management efficiency. Traditional electrocardio monitoring devices (such as electrocardiogram machines) have limitations such as large size, poor portability, and dependence on professional operation, and can only realize single or short-time monitoring in medical institution scenarios, which is difficult to meet the needs of daily health management, long-term tracking of chronic diseases, and emergency warning.

[0003] The development of wearable technology has promoted the application of portable electrocardio monitoring devices. Although existing wearable devices such as smart bracelets and watches can monitor basic indicators such as heart rate, they are limited by hardware performance and algorithm accuracy, and have the following technical bottlenecks: Signal interference problem is prominent: electrocardio signals are weak (usually in millivolts), easily affected by breathing, muscle tremor, motion artifacts and electromagnetic environment interference, resulting in low signal-to-noise ratio of original signals, which directly affects the accuracy of subsequent analysis; Insufficient feature extraction capability: traditional algorithms (such as wavelet transform and Fourier analysis) have limited recognition accuracy for key waveforms such as P wave, QRS complex and T wave in electrocardiogram, especially in signal distortion scenarios, which is prone to missed detection or false detection; Weak intelligent diagnosis capability: existing devices are mostly limited to data collection and simple trend display, lacking deep analysis of abnormal electrocardio signals and clinical-level diagnosis suggestions, making it difficult to assist users or medical staff in quickly judging health risks; Low knowledge integration: electrocardio monitoring needs to be combined with medical knowledge base, clinical cases and individual health data for comprehensive evaluation, while existing systems lack dynamic association with network medical resources, limiting the professionalism and reference value of diagnosis reports.

[0004] In recent years, artificial intelligence technologies such as blind source signal separation (BSS) and convolutional neural network (CNN) have shown advantages in signal processing and feature recognition, and algorithms such as random forest (RF) and decision tree (DT) have high accuracy in classification tasks, and large language models (LLM) have strong knowledge integration and natural language generation capabilities. However, existing technologies have not formed a wearable electrocardio monitoring solution that deeply integrates the above technologies, which cannot simultaneously solve the problems of signal denoising, accurate feature extraction, intelligent classification and professional report generation in the whole link.

[0005] Therefore, developing a large model wearable electrocardio monitoring system and method integrating BSS-CNN-RF-DT-LLM fusion technology to realize high-precision signal processing, multi-dimensional feature analysis, intelligent health assessment and professional report generation has become a technical problem to be solved in the field. SUMMARY

[0006] The main purpose of the present application is to overcome the shortcomings and deficiencies of the prior art, and to provide a wearable health device large model electrocardio monitoring system and method based on BSS-CNN-RF-DT-LLM fusion technology. Through the five BSS-CNN-RF-DT-LLM fusion technologies, the software and hardware system architecture and its function modules of the large model wearable electrocardio monitoring platform are designed to realize the collection, monitoring and diagnosis of the morphology, time domain, frequency domain and amplitude of the electrocardiogram, heart rate, P wave, QRS complex, T wave, U wave, PR interval, QT interval and RR interval of the wearer's electrocardio signal.

[0007] In order to achieve the above purpose, the following technical scheme is adopted in the present application: In a first aspect, the present application provides a wearable health device large model electrocardio monitoring system based on BSS-CNN-RF-DT-LLM fusion technology, comprising a hardware device and a software platform; the hardware device comprises an intelligent data integration processing chip, a blind source signal separation algorithm chip, a convolutional neural network algorithm chip, a random forest construction and traversal algorithm chip, a decision tree algorithm chip, a large model algorithm chip, a digital monitoring chip and a large model electrocardio diagnosis report storage chip; The intelligent data integration processing chip is used to receive each electrocardio signal sent by the digital trigger transmitter from the digital memory, and to perform data integration, data preprocessing and data quality filtering; The blind source signal separation algorithm chip integrates the FastICA algorithm of independent component analysis method, the SOBI algorithm based on high-order statistical method and the Infomax algorithm based on information theory method, and is used to eliminate the artifacts in the electrocardiogram source signal caused by physiological factors and environmental factors; The convolutional neural network algorithm chip integrates convolutional kernels of different sizes of convolutional neural network, pooling matrices, convolutional neural network feature recognition algorithms and LeNet architecture, AlexNet architecture, VGGNet architecture and GoogleNet architecture constructed by convolutional neural network, and is used to perform feature recognition on the statistically independent signals obtained by blind source signal separation, and accurately extract each physiological index data of the electrocardio signal; The random forest construction integrates depth-first search algorithm and breadth-first search algorithm with the algorithm chip set, which is used to construct a random forest with the statistically independent signals obtained by blind source separation as root nodes, and the root nodes include root nodes corresponding to electrocardiosignal and root nodes corresponding to artifact signal; the root nodes corresponding to electrocardiosignal are connected with the second layer nodes under the root nodes, and the second layer nodes are connected with the shape, time domain, frequency domain and amplitude sub-trees, and the second layer nodes are connected with the "normal", "abnormal" and "dangerous" sub-trees, and the root nodes corresponding to artifact signal are independent trees; The decision tree algorithm chip integrates decision tree algorithm, classification algorithm and Booting pruning algorithm, which is used to connect each tree in the random forest with "electrocardiosignal" as the root node to form "random forest-decision tree", cut off "artifact tree" by Booting pruning algorithm, and identify the leaf nodes into "normal", "abnormal" and "dangerous" three categories, and store the corresponding electrocardio indicators and collected data into the corresponding regional space; The large model algorithm chip integrates large model algorithm, which is used to perform large model deep search language understanding, information capture, semantic splicing and typical case analysis on electrocardiosignal measurement data, and form large model semantic generation text description; The large model electrocardiogram diagnosis report storage chip is used to connect the large model semantic generation storage chip, judge the health status of the wearer according to the text description generated by the large model semantic generation, and generate a large model health diagnosis report; The digital monitoring chip is used to integrate the sensing, reading, receiving, measuring and sending of electrocardiosignal by the intelligent sound and light sensor, human body sensing measurer and digital trigger sender into the large language model algorithm integrated chip for semantic generation, generate electrocardio indicator analysis report and heart health status diagnosis report, and visualize the display on the display screen; The display screen is used to display the visual real-time dynamic fluctuation graph of the shape, time domain, frequency domain and amplitude of electrocardio indicators, and is provided with "electrocardiogram", "heart rate", "P wave", "QRS wave group", "T wave", "U wave", "PR interval", "QT interval", "RR interval" and "electrocardiosignal active monitoring" icons (software icons), and clicking the first 9 software icons can display the corresponding electrocardio indicator dynamic fluctuation graph and large model analysis diagnosis report, and after clicking the "electrocardiosignal active monitoring" icon, selecting the target icon and setting the time, the corresponding electrocardio indicator dynamic fluctuation graph and large model analysis diagnosis report under active monitoring are displayed; The software platform comprises a knowledge base, an intelligent algorithm system and a large model electrocardiosignal monitoring system; the knowledge base comprises a whole network search knowledge base and a heart health knowledge base; the intelligent algorithm system comprises a blind source signal separation algorithm system, a convolutional neural network feature extraction algorithm system, a random forest construction and traversal algorithm system, a decision tree classification algorithm system, a large model search and generation algorithm system and a large model wearable electrocardiosignal monitoring algorithm system; the large model electrocardiosignal monitoring system comprises a large model electrocardiogram monitoring system, a large model heart rate monitoring system, a large model P wave monitoring system, a large model QRS wave group monitoring system, a large model T wave monitoring system, a large model U wave monitoring system, a large model PR interval monitoring system, a large model QT interval monitoring system and a large model RR interval monitoring system, each system is used for real-time monitoring of corresponding electrocardiosignals, generating corresponding decision trees and analysis reports, and displaying dynamic fluctuation graphs and analysis reports in a specified area of a display screen, and each system sets the same time monitoring granularity.

[0008] As a preferred technical solution, the working process of the algorithm in the blind source signal separation algorithm chip is as follows: When the source signals are statistically independent and non-Gaussian, the FastICA algorithm of independent component analysis is used to separate the blind source signals; When the source signals are correlated in time and non-Gaussian, the SOBI algorithm based on high-order statistics method is used to separate the blind source signals; When the mutual information between the components of the source signals is minimum and statistically independent, the Infomax algorithm based on information theory method is used to separate the blind source signals.

[0009] As a preferred technical solution, the electrocardiosignals comprise electrocardiograms, heart rates, P waves, QRS wave groups, T waves, U waves, PR intervals, QT intervals and RR intervals to form root nodes, which are respectively called the root nodes of "electrocardiogram tree", "heart rate tree", "P wave tree", "QRS wave group tree", "T wave tree", "U wave tree", "PR interval tree" and "RR interval tree"; each root node of the nine trees is connected to four sub-trees at the second layer of nodes; the four sub-trees only have root nodes, which are respectively: morphology, time domain, frequency domain and amplitude; then, the second layer of leaf nodes of the nine trees are also connected to three sub-trees respectively; the three sub-trees also only have root nodes, which are respectively: "normal", "abnormal" and "dangerous"; the root nodes of the three sub-trees are respectively made into the third layer of nodes of the nine trees, and are also the leaf nodes of the nine trees; The artifact signals are made into a tree alone, referred to as "artifact tree 1", "artifact tree 2",..., "artifact tree n", n represents that n artifact signals are separated from the mixed electrocardio signals, and all "artifact trees" only have root nodes; then, the nine electrocardio signal trees, i.e. "electrocardiogram tree", "heart rate tree", "P wave tree", "QRS complex tree", "T wave tree", "U wave tree", "PR interval tree" and "RR interval tree", and all "artifact trees" are placed together to form a "random forest".

[0010] As a preferred technical solution, the full-network search knowledge base covers medical and health websites, public hospital websites, private hospital websites, internet websites related to heart health, health newspaper websites and public websites that can publish doctor forums and online consultations from national to provincial, municipal, county and town; the heart health knowledge base is composed of "electrocardiogram-knowledge sub-base", "heart rate-knowledge sub-base", "P wave-knowledge sub-base", "QRS complex-knowledge sub-base", "T wave-knowledge sub-base", "U wave-knowledge sub-base", "PR interval-knowledge sub-base", "QT interval-knowledge sub-base" and "RR interval-knowledge sub-base", and each sub-base searches the corresponding electrocardio monitoring index in the full-network search knowledge base to establish a related knowledge base through a large model search and generation algorithm.

[0011] As a preferred technical solution, the blind source signal separation algorithm system is used to perform a blind source signal separation algorithm on the mixed electrocardio signals preprocessed and filtered by the intelligent data integration processing chip, and save the result to a BSS algorithm chip; the convolutional neural network feature extraction algorithm system is used to perform a feature extraction algorithm on the signals separated by the BSS algorithm, and save the result to a CNN algorithm chip; the random forest construction and traversal algorithm system is used to establish a random forest for the signals whose features are extracted by the CNN algorithm, traverse and mark the leaf nodes of the current heart rate condition, and save the result to a RF construction and traversal algorithm chip; the decision tree classification algorithm system is used to connect the random forest with "electrocardio monitoring" as a root node to form a "random forest-decision tree", cut off the "artifact tree" and classify the leaf nodes, and save the result to a DT classification algorithm chip; the large model search and generation algorithm system is used to generate a health diagnosis report for the data classified by the DT classification algorithm chip in combination with the knowledge base, and save the result to a large model electrocardio diagnosis report storage chip; the large model wearable electrocardio monitoring algorithm system is used to display an electrocardiogram dynamic fluctuation graph and various analysis and diagnosis reports in a specified area of a display screen.

[0012] As a preferred technical solution, in the large model electrocardio signal monitoring system, The large model electrocardiogram monitoring system is used to monitor electrocardiogram signals in real time, generate a decision tree and an analysis report corresponding to the electrocardiogram, and display an electrocardiogram dynamic fluctuation graph and a large model analysis report in a specified area of a display screen. The large model heart rate monitoring system is used to monitor heart rate signals in real time, generate decision trees and analysis reports corresponding to heart rate, and display heart rate dynamic fluctuation graphs and large model analysis reports in a designated area of ​​the display screen. The large-scale P-wave monitoring system is used to monitor P-wave signals in real time, generate decision trees and analysis reports corresponding to P-waves, and display dynamic fluctuation diagrams of P-waves and large-scale model analysis reports in a designated area of ​​the display screen. The large-scale model QRS group monitoring system is used to monitor QRS group signals in real time, generate decision trees and analysis reports corresponding to QRS groups, and display the dynamic fluctuation diagram of QRS groups and the large-scale model analysis report in a designated area of ​​the display screen. The large-scale T-wave monitoring system is used to monitor T-wave signals in real time, generate decision trees and analysis reports corresponding to T-waves, and display T-wave dynamic fluctuation diagrams and large-scale model analysis reports in a designated area of ​​the display screen. The large-scale U-wave monitoring system is used to monitor U-wave signals in real time, generate decision trees and analysis reports corresponding to U-waves, and display the dynamic fluctuation diagram of U-waves and the large-scale model analysis report in a designated area of ​​the display screen. The large model PR interval monitoring system is used to monitor PR interval signals in real time, generate decision trees and analysis reports corresponding to PR intervals, and display the dynamic fluctuation chart of PR intervals and the large model analysis report in a designated area of ​​the display screen. The large model QT interval monitoring system is used to monitor QT interval signals in real time, generate decision trees and analysis reports corresponding to QT intervals, and display the dynamic fluctuation chart of QT intervals and the large model analysis report in a designated area of ​​the display screen. The large-scale model RR interval monitoring system is used to monitor RR interval signals in real time, generate decision trees and analysis reports corresponding to RR intervals, and display dynamic fluctuation charts of RR intervals and large-scale model analysis reports in a designated area of ​​the display screen.

[0013] Secondly, this invention provides a method for large-scale ECG monitoring of wearable health devices based on BSS-CNN-RF-DT-LLM fusion technology, comprising the following methods: S1: The wearer puts on the hardware device of the system and turns on the device to put it into normal working condition; S2: The intelligent audio-visual sensor detects the wearer's real-time health status, including electrocardiogram, heart rate, P wave, QRS complex, T wave, U wave, PR interval, QT interval, and RR interval. S3: The human body sensor measures the physiological indicators corresponding to the above-mentioned electrocardiogram signals of the wearer. S4: The digital receiver receives real-time measurements of various electrocardiogram signals and transmits them to the digital memory for storage, which is then stored in the device's backend. S5: The digital memory transmits the saved electrocardio signal measurement values to the intelligent data integration processing chip through the digital trigger transmitter, and data preprocessing and data quality filtering are performed in the chip; S6: The system starts the blind source signal separation algorithm chip, the blind source signal separation algorithm system enters the working state, and automatically executes the blind source signal separation algorithm to separate the mixed electrocardio signal containing interference to obtain a statistically independent and non-Gaussian independent signal source; S7: The system starts the convolutional neural network algorithm chip, the convolutional neural network algorithm system enters the working state, and automatically executes the convolutional neural network algorithm to identify the features of the independent signal source separated in step S6 to distinguish the features of the electrocardio signal and the interference signal; S8: The system starts the random forest construction and traversal algorithm chip, the random forest construction and traversal algorithm system enters the working state, and takes the independent signal source separated in step S6 as a root node; for the root node with the electrocardio signal feature identified in step S7, each root node is connected with a morphology, time domain, frequency domain, and amplitude sub-node as a second layer node, and each second layer node is connected with a "normal", "abnormal", and "dangerous" classification node as a third layer node, which is also a leaf node, to form an "electrocardiogram tree", a "heart rate tree", a "P wave tree", a "QRS wave group tree", a "T wave tree", a "U wave tree", a "PR interval tree", a "QT interval tree", and a "RR interval tree"; for the root node with the artifact signal feature, a "artifact tree" containing only the root node is formed; and all "electrocardio signal trees" and "artifact trees" are combined to form a random forest; S9: The system starts the decision tree classification algorithm chip, the decision tree classification algorithm system enters the working state, and automatically adds an "electrocardio signal" as a root node in the random forest established in step S8, connects the root node with all "electrocardio signal trees" and "artifact trees" in the random forest, and constructs an "electrocardio signal-decision tree"; the "electrocardio signal-decision tree" is composed of the "electrocardio signal" root node and the "random forest" constructed in step S8; therefore, when the "electrocardio signal-decision tree" is traversed, the "electrocardio signal" root node is not considered, and the forest traversal algorithm is directly used to traverse the "random forest" constructed in step S8 under the "electrocardio signal" root node; in the traversal process, when a leaf node is accessed, the actual monitoring value of the "electrocardio signal" root node corresponding to the leaf node is compared and matched with the ranges of "normal", "abnormal", and "dangerous" three classifications, and the matching result is saved to the decision tree classification algorithm chip in a ternary array (electrocardio signal, monitoring value, and judgment range); S10: The system starts the large model electrocardiogram diagnosis report storage chip, and the large model search and generation algorithm system enters a working state. The large model searches the entire network, extracts the normal range, abnormal range, dangerous range and typical cases of the corresponding electrocardiogram signal in the knowledge base, performs deep mining, understanding, comparison, information capture, semantic splicing and chart interpretation, forms a large model semantic generation text description, generates a large model electrocardiogram signal analysis report and a large model heart health status report, and saves them in the large model electrocardiogram diagnosis report storage chip. S11: The system starts the digital monitoring chip, and the large model wearable electrocardiogram monitoring algorithm system enters a working state. The real-time monitoring electrocardiogram of the wearer can be displayed left and right on the upper one-third area of the device display screen, and the large model electrocardiogram signal analysis report and the large model heart health status diagnosis report can be displayed up and down on the lower two-thirds area of the display screen.

[0014] As a preferred technical solution, in step S6, when the blind source signal separation algorithm is executed, a corresponding algorithm is selected according to the characteristics of the independent signal source: when the source signal is statistically independent and non-Gaussian distributed, the FastICA algorithm is used; when the source signal has correlation in time and non-Gaussianity, the SOBI algorithm is used; when the mutual information between the components of the source signal is minimum and statistically independent, the Infomax algorithm is used; the independent signal source separated by the above algorithm can eliminate the artifact interference caused by motion, emotion, electromagnetic environment, etc.

[0015] As a preferred technical solution, in step S8, the process of constructing the random forest includes: The morphology sub-node of the "electrocardiogram signal tree" records the waveform characteristics of the corresponding electrocardiogram signal, the time domain sub-node records the time interval parameters of the signal, the frequency domain sub-node records the frequency distribution characteristics of the signal, and the amplitude sub-node records the voltage amplitude range of the signal. The threshold range of the "normal", "abnormal" and "dangerous" classification nodes is set based on the clinical standard of the corresponding electrocardiogram index in the heart health knowledge base, wherein the "normal" range is the medically recognized healthy interval, the "abnormal" range is the interval deviating from the normal but not reaching the critical interval, and the "dangerous" range is the interval prompting the possible occurrence of acute cardiac events.

[0016] As a preferred technical solution, in step S11, the interaction logic of the display screen includes: When the wearer clicks the icon corresponding to the 9 electrocardiogram signals on the display screen, the display screen focuses on displaying the full-screen dynamic fluctuation graph of the signal, wherein the waveform segment exceeding the "normal" range is marked in red, and the special analysis report of the signal is also displayed synchronously; The display screen supports switching monitoring data of different time points through touch sliding, and the large model electrocardiosignal analysis report includes waveform feature description, index value comparison, and abnormal point labeling; the large model heart health condition diagnosis report includes risk level evaluation, recommended intervention measures, and typical case reference.

[0017] Compared with the prior art, the present application has the following advantages and beneficial effects: 1. The BSS algorithm chip integrating FastICA, SOBI and Infomax algorithms can adaptively select a separation strategy according to the characteristics of the source signal, accurately eliminate the interference of artifacts caused by physiological factors such as emotional fluctuations, breathing rhythm, motion state, body temperature changes, and electromagnetic environment, and separate the statistically independent pure electrocardiosignal source, solving the pain points of low signal-to-noise ratio and weak anti-interference ability of traditional wearable devices. Combined with the multi-architecture (LeNet, AlexNet, etc.) and multi-size convolution kernel design of the CNN algorithm chip, the morphology, time domain, frequency domain, and amplitude features of the electrocardiosignal can be accurately extracted, and the subtle distortion of key waveforms such as P wave, QRS complex, and T wave can be efficiently identified, and the feature extraction accuracy is significantly improved compared with traditional algorithms.

[0018] 2. The patent constructs a random forest containing 9 electrocardiosignal trees and artifact trees through RF traversal algorithm, and then uses Booting pruning technology and decision tree classification algorithm of DT algorithm chip to complete the pruning of artifact tree and the accurate grading of "normal- abnormal-dangerous" of "random forest-decision tree" after pruning; the classification threshold is set based on clinical standards, which can directly provide risk level reference for users, avoiding the limitation of traditional devices that can only provide raw data and lack of diagnosis interpretation. By integrating LLM algorithm and whole network medical knowledge base, the generated analysis report includes waveform feature description, abnormal point labeling, associated disease interpretation, and typical case reference, and the professionalism and readability of the diagnosis report are considered, so that non-professional users can also quickly understand their own heart health status, realizing the intelligentization of the whole link of "data collection-feature analysis-diagnosis interpretation".

[0019] 3. The 9 large model electrocardiosignal monitoring system of the present application synchronously covers 9 core electrocardiosignal indicators such as electrocardiogram, heart rate, P wave, QRS complex, T wave, U wave, PR interval, QT interval, and RR interval, and supports custom time monitoring granularity, which not only meets the normalization needs of daily health monitoring, but also provides long-term tracking data for patients with cardiovascular diseases, filling the gap of single monitoring dimension of traditional wearable devices. And it supports active monitoring and historical data comparison function, the wearer can start 1 minute high frequency sampling by long pressing the target icon, or call up the historical diagnosis report for longitudinal comparison, and intuitively master the trend of heart health changes, providing all-round support for home monitoring, chronic disease management and emergency situation early warning.

[0020] 4, The display screen of the patent adopts partition display and touch interaction design, dynamic fluctuation graph supports left and right sliding backtracking / preview, analysis report supports up and down sliding browsing, full screen switching function meets different viewing needs, abnormal waveform is marked with red color, key information is clear at a glance, and the operation threshold is reduced. Through the integration of software and hardware, the device is small and portable, and automatically enters the monitoring state after starting, without professional operation, signal acquisition and analysis can be completed, which is suitable for daily wearing scene, and takes into account practicability and portability. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0022] Figure 1 It is a plan view of the wearable electrocardio monitoring device and display screen of the large model of the present application embodiment 1. Figure 2 It is a plan view of the device and display screen in the environment of "active monitoring of electrocardio signal" of the large model of the present application embodiment 1. Figure 3 It is a plan view of the device and display screen in the environment of "large model electrocardio diagnosis report" query of the present application embodiment 1. Figure 4 It is a plan view of the built-in chip of the large model wearable electrocardio monitoring device of the present application embodiment 1. Figure 5 It is a large model electrocardio signal monitoring principle diagram based on multi-intelligent algorithm fusion technology of the present application embodiment 1. Figure 6 It is nine electrocardio signal monitoring systems of the large model wearable electrocardio monitoring platform of the present application embodiment 1. Figure 7 It is the total architecture of the software platform of the present application embodiment 1. Figure 8 It is a hardware system working principle flow chart of the large model wearable electrocardio monitoring platform of the present application embodiment 1. Figure 9 It is a display diagram on the display screen during the working of the large model of the present application embodiment 2. Figure 10 It is the first display diagram about "large model heart rate fluctuation analysis report" on the display screen after the working of the large model of the present application embodiment 2. Figure 11 It is a display diagram about heart rate on the display screen after the working of the large model of the present application embodiment 2. Figure 12The second display figure on the display screen about the "large model heart rate fluctuation analysis report" after the large model of the embodiment 2 of the present application is finished working; Figure 13 The display figure on the display screen during the working of the large model of the embodiment 3 of the present application; Figure 14 The first display figure on the display screen about the "large model QRS complex fluctuation analysis report" after the large model of the embodiment 3 of the present application is finished working; Figure 15 The display figure of the QRS complex fluctuation figure on the display screen after the large model of the embodiment 3 of the present application is finished working; Figure 16 The second display figure on the display screen about the "large model QRS complex fluctuation analysis report" after the large model of the embodiment 3 of the present application is finished working. DETAILED DESCRIPTION

[0023] In order to enable personnel in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0024] In the present application, the phrase "embodiment" means that the specific features, structures or characteristics described in combination with the embodiment can be contained in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily mean the same embodiment, nor is it an independent or alternative embodiment that is not mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described in the present application can be combined with other embodiments.

[0025] Embodiment 1 The patent proposes to develop a "large model wearable electrocardiogram monitoring device" (hardware device), and to develop a large model wearable electrocardiogram monitoring platform (software system) based on BSS-CNN-BP-DT-LLM fusion technology, referred to as "large model wearable electrocardiogram monitoring platform". Among them, BSS (Blind source separation) is an algorithm technology for separating electrocardiogram blind source signals, CNN (Convolutional Neural Network) is a convolutional neural network technology for extracting electrocardiogram signal features, RF (Random Forest) is a random forest for establishing a signal source for electrocardiogram signal feature output, referred to as electrocardiogram signal feature random forest; DT (Decision Tree) is a decision tree for electrocardiogram signal feature random forest through automatic increase of "electrocardiogram signal", referred to as "electrocardiogram signal-decision tree"; LLM is used for large model semantic analysis and logical connection of "electrocardiogram signal-decision tree" decision results, and finally outputs real-time dynamic monitoring results and accurate diagnosis results of electrocardiogram signals. Under the five fusion technologies of BSS-CNN-RF-DT-LLM, the software and hardware system architecture and function modules of the large model wearable electrocardiogram monitoring platform are designed; from hardware architecture, software platform to algorithm principle, the morphology, time domain, frequency domain and amplitude of electrocardiogram, heart rate, P wave, QRS wave group, T wave, U wave, PR interval, QT interval and RR interval of the wearer's electrocardiogram signal are collected, monitored and diagnosed.

[0026] The patent divides the monitored physiological index monitoring values of the nine electrocardiogram signals such as electrocardiogram, heart rate, P wave, QRS wave group, T wave, U wave, PR interval, QT interval and RR interval into three regions according to whether they are within the normal range: "normal region", "abnormal region" and "dangerous region". The standards for dividing the three regions are as follows: (1) "Normal region": the physiological index monitoring value of the electrocardiogram signal is within the normal value range; (2) "Abnormal region": the physiological index monitoring value of the electrocardiogram signal is within the abnormal value but does not reach the range of life danger; (3) "Dangerous region": the physiological index monitoring value of the electrocardiogram signal is within the abnormal value range and far exceeds the abnormal state to reach the range of life danger.

[0027] It can be understood that the present patent actually collects, monitors and diagnoses 12 basic parameters of electrocardiogram measurement, i.e. heart rate, P wave duration, PR (PQ) interval, QRS complex duration, QT (QTc) interval, P wave, QRS complex, T wave axis, P wave amplitude, QRS complex amplitude, ST segment amplitude, T wave amplitude and U wave amplitude. However, for the sake of brevity, the present patent is written as "collecting, monitoring and diagnosing the morphology, time domain, frequency domain and amplitude of electrocardiogram, heart rate, P wave, QRS complex, T wave, U wave, PR interval, QT interval and RR interval of the wearer's electrocardiogram signal" according to the large category related to the electrocardiogram parameters. The present embodiment is based on a wearable health device large model electrocardiogram monitoring system based on BSS-CNN-RF-DT-LLM fusion technology, which includes hardware devices and software platforms. Among them, the hardware device includes intelligent sound and light sensor, human body sensing measurer, digital trigger transmitter, digital receiver, digital memory, intelligent data integration processing chip, display screen, device key, large model reading key, intelligent data integration processing chip, blind source signal separation (BSS) algorithm chip, convolutional neural network (CNN) algorithm chip, random forest (RF) traversal algorithm chip, decision tree (DT) algorithm chip, large model (LLM) algorithm chip, digital monitoring chip, large model electrocardiogram diagnosis report storage chip, etc.

[0028] As shown in Figure 1 The display screen can display the visual real-time dynamic fluctuation graph of the morphology, time domain, frequency domain and amplitude of the electrocardiogram, heart rate, P wave, QRS complex, T wave, U wave, PR interval, QT interval and RR interval of the wearer's electrocardiogram index. On the display screen, the icons (all icons are software buttons) of "electrocardiogram", "heart rate", "P wave", "QRS complex", "T wave", "U wave", "PR interval", "QT interval", "RR interval" and "active monitoring of electrocardiogram signal" are displayed, among which "active monitoring of electrocardiogram signal" is a borderless software button with a background color.

[0029] Further, when the wearer uses the device, the wearer gently clicks one of the target "icon" buttons on the display screen with the other hand not wearing the device. If it is active monitoring, it needs to be continuously pressed for about 1 minute. During this period, the display screen displays "the large model is working, it needs about 1 minute. Please wait, do not release your hand……". Specifically: (1) When the "ECG" icon is clicked, the large model wearable ECG monitoring platform automatically displays the wearer's dynamic fluctuation ECG in real time on the top one-third area of the display screen. This dynamic fluctuation ECG can be gently swiped left and right with the hand. When swiping to the right, the ECG fluctuation graph before the current screen display time can be seen; when swiping to the left, the ECG fluctuation graph after the current screen display time can be seen, up to the current monitoring time. And after the wearer gently touches this one-third area with the hand, the real-time dynamic fluctuation ECG can be viewed full screen. The dynamic fluctuation ECG is equipped with dynamic text annotations, such as the P wave, T wave, QRS complex, etc. Abnormal parts of the ECG will appear red curve segments.

[0030] At the same time, the large model wearable ECG monitoring platform automatically visualizes the wearer's ECG normal and abnormal analysis and heart health status diagnosis report in the current time in the lower two-thirds area of the display screen, which is a "large model ECG fluctuation analysis report" and a "large model heart health status diagnosis report". The location of this report on the display screen can be swiped up and down with the hand. The analysis of the wearer's ECG condition, heart health status, and various related statistical chart analysis in the report. This display part can be swiped up and down for the wearer to read. And after the wearer gently touches this two-thirds area with the hand, the "large model ECG fluctuation analysis report" and "large model heart health status diagnosis report" can be viewed full screen respectively.

[0031] (2) When the "heart rate" icon is clicked, the large model wearable ECG monitoring platform automatically displays the wearer's heart rate dynamic fluctuation graph in real time on the top one-third area of the display screen. This heart rate dynamic fluctuation graph can be gently swiped left and right with the hand. When swiping to the right, the heart rate dynamic fluctuation graph before the current screen display time can be seen; when swiping to the left, the heart rate dynamic fluctuation graph after the current screen display time can be seen, up to the current monitoring time. And after the wearer gently touches this one-third area with the hand, the real-time dynamic fluctuation graph of the heart rate can be viewed full screen. The dynamic fluctuation heart rate graph is equipped with dynamic text annotations, such as the amplitude scale and time scale (seconds) of the heart rate fluctuation, etc. The display screen also displays the minimum heart rate, maximum heart rate, and average heart rate. In addition, abnormal parts of the heart rate dynamic fluctuation graph will appear red curve segments.

[0032] At the same time, the large model wearable electrocardio monitoring platform automatically visualizes the normal and abnormal analysis of the wearer's electrocardiogram at the current moment and the diagnosis report of the heart health status in the lower two-thirds area of the display screen, which is a large model heart rate fluctuation analysis report. The position of the report on the display screen can be browsed up and down by the wearer with a light touch of the hand. The report analyzes the normal and abnormal of the wearer's heart rate, displays the statistical charts related to heart rate, and prompts the normal value, abnormal value and dangerous value of heart rate in words, displays various related knowledge that affects heart rate and various heart diseases caused by abnormal heart rate. And the wearer can full-screen browse the large model heart rate fluctuation analysis report by lightly touching the two-thirds area with the hand.

[0033] (3) When the "P wave" icon is clicked, the large model wearable electrocardio monitoring platform automatically displays the wearer's dynamic fluctuation P wave graph in the upper one-third area of the display screen in real time. The dynamic fluctuation P wave graph can be swiped left and right with a light touch of the hand. When swiping to the right, the P wave graph before the current screen display time can be seen; when swiping to the left, the P wave graph after the current screen display time can be seen, and it is displayed until the current monitoring time. And the wearer can full-screen browse the real-time dynamic fluctuation P wave graph by lightly touching the one-third area with the hand. The dynamic fluctuation P wave graph is equipped with dynamic text annotations, such as the amplitude scale, frequency scale and time scale (seconds) of P wave fluctuation annotations on the graph. The abnormal part of the P wave fluctuation graph will appear a red curve segment.

[0034] At the same time, the large model wearable electrocardio monitoring platform automatically visualizes the normal and abnormal analysis of the wearer's P wave graph at the current moment and the diagnosis report of the heart health status in the lower two-thirds area of the display screen, which is a large model P wave analysis report. The position of the report on the display screen can be browsed up and down by the wearer with a light touch of the hand. The report analyzes the normal and abnormal of the wearer's P wave, displays the statistical charts related to P wave, and prompts the normal value, abnormal value and dangerous value of P wave in words, displays various related knowledge that affects P wave and various heart diseases caused by abnormal P wave. And the wearer can full-screen browse the large model P wave fluctuation analysis report by lightly touching the two-thirds area with the hand.

[0035] (4) When the "QRS complex" icon is clicked, the large model wearable ECG monitoring platform automatically displays the wearer's dynamic fluctuation QRS complex graph in real time in the upper one-third area of the display screen. This dynamic fluctuation QRS complex graph can be gently swiped left and right with the hand. When swiping to the right with the hand, the QRS complex graph before the current screen display time can be seen; when swiping to the left, the QRS complex graph after the current screen display time can be seen, and it is displayed until the current monitoring time. And after the wearer gently touches this one-third area with his hand, he can view the real-time dynamic fluctuation graph of the QRS complex on the full screen. The dynamic fluctuation QRS complex graph is equipped with dynamic text annotations, such as the amplitude scale, frequency scale, and time scale (seconds) of the P wave fluctuation marked on the graph. The abnormal part of the QRS complex fluctuation graph will appear a red curve segment.

[0036] At the same time, the large model wearable ECG monitoring platform automatically visualizes the wearer's QRS complex graph at the current time in the lower two-thirds area of the display screen, normal and abnormal analysis, and diagnosis report of the heart health status, which is a "large model QRS complex analysis report". The location of this report on the display screen can be swiped up and down with the hand. The report analyzes the normal and abnormal QRS complex conditions of the wearer, displays statistical charts related to the QRS complex, and prompts the normal value, abnormal value, and dangerous value of the QRS complex in words, displays various related knowledge that affects the QRS complex, and various heart diseases caused by QRS complex abnormalities. And after the wearer gently touches this two-thirds area with his hand, he can view the "large model QRS complex fluctuation analysis report" on the full screen.

[0037] (5) When the "T wave" icon is clicked, the large model wearable ECG monitoring platform automatically displays the wearer's dynamic fluctuation T wave graph in real time in the upper one-third area of the display screen. This dynamic fluctuation T wave graph can be gently swiped left and right with the hand. When swiping to the right with the hand, the T wave graph before the current screen display time can be seen; when swiping to the left, the T wave graph after the current screen display time can be seen, and it is displayed until the current monitoring time. And after the wearer gently touches this one-third area with his hand, he can view the real-time dynamic fluctuation graph of the T wave on the full screen. The dynamic fluctuation T wave graph is equipped with dynamic text annotations, such as the amplitude scale, frequency scale, and time scale (seconds) of the T wave fluctuation marked on the graph. The abnormal part of the T wave fluctuation graph will appear a red curve segment.

[0038] At the same time, the large model wearable electrocardiogram monitoring platform automatically visualizes the normal and abnormal analysis of the wearer's T wave graph at the current time and the diagnosis report of the heart health status in the lower two-thirds area of the display screen, which is a large model T wave analysis report. The position of the report on the display screen can be browsed up and down by the wearer with a light touch of the hand. The report analyzes the normal and abnormal T wave conditions of the wearer, displays statistical charts related to T wave, and prompts the normal, abnormal and dangerous values of T wave in words, displays various related knowledge that affects T wave and various heart diseases caused by T wave abnormalities. And the wearer can full-screen browse the large model T wave fluctuation analysis report by lightly touching the two-thirds area with his hand.

[0039] (6) When the "U wave" icon is clicked, the large model wearable electrocardiogram monitoring platform automatically displays the dynamic fluctuation U wave graph of the wearer in real time in the upper one-third area of the display screen. The dynamic fluctuation U wave graph can be swiped left and right with a light touch of the hand. When swiping to the right, the U wave graph before the current screen display time can be seen; when swiping to the left, the U wave graph after the current screen display time can be seen, and it is displayed until the current monitoring time. And the wearer can full-screen browse the real-time dynamic fluctuation U wave graph by lightly touching the one-third area. The dynamic fluctuation U wave graph is equipped with dynamic text annotations, such as the amplitude scale, frequency scale and time scale (seconds) of U wave fluctuation annotations on the graph. The abnormal part of the U wave graph will appear a red curve segment.

[0040] At the same time, the large model wearable electrocardiogram monitoring platform automatically visualizes the normal and abnormal analysis of the wearer's U wave graph at the current time and the diagnosis report of the heart health status in the lower two-thirds area of the display screen, which is a large model U wave analysis report. The position of the report on the display screen can be browsed up and down by the wearer with a light touch of the hand. The report analyzes the normal and abnormal U wave conditions of the wearer, displays statistical charts related to U wave, and prompts the normal, abnormal and dangerous values of U wave in words, displays various related knowledge that affects U wave and various heart diseases caused by U wave abnormalities. And the wearer can full-screen browse the large model U wave fluctuation analysis report by lightly touching the two-thirds area with his hand.

[0041] (7) When the "PR interval" icon is clicked, the large model wearable ECG monitoring platform automatically displays the wearer's dynamic fluctuation PR interval graph in real time in the upper one-third area of the display screen. This dynamic fluctuation PR interval graph can be gently swiped left and right with the hand. When swiping to the right with the hand, the PR interval graph before the current screen display time can be seen; when swiping to the left, the PR interval graph after the current screen display time can be seen, up to the current monitoring time. And after the wearer gently touches this one-third area with his hand, he can view the real-time dynamic fluctuation graph of the PR interval on the full screen. The dynamic fluctuation PR interval graph is equipped with dynamic text annotations, such as the amplitude scale, frequency scale, and time scale (seconds) of the PR interval fluctuation on the graph. The abnormal part of the PR interval fluctuation graph will appear a red curve segment.

[0042] At the same time, the large model wearable ECG monitoring platform automatically visualizes the wearer's PR interval graph at the current time in the lower two-thirds area of the display screen, normal and abnormal analysis, and diagnosis report of the heart health status, which is a "large model PR interval analysis report". The location of this report on the display screen can be swiped up and down with the hand. The report analyzes the normal and abnormal PR interval conditions of the wearer, displays statistical charts related to the PR interval, and prompts the normal value, abnormal value, and dangerous value of the PR interval in text, displays various related knowledge that affects the PR interval, and various heart diseases caused by PR interval abnormalities. And after the wearer gently touches this two-thirds area with his hand, he can view the "large model PR interval fluctuation analysis report" on the full screen.

[0043] (8) When the "QT interval" icon is clicked, the large model wearable ECG monitoring platform automatically displays the wearer's dynamic fluctuation QT interval graph in real time in the upper one-third area of the display screen. This dynamic fluctuation QT interval graph can be gently swiped left and right with the hand. When swiping to the right with the hand, the QT interval graph before the current screen display time can be seen; when swiping to the left, the QT interval graph after the current screen display time can be seen, up to the current monitoring time. And after the wearer gently touches this one-third area with his hand, he can view the real-time dynamic fluctuation graph of the QT interval on the full screen. The dynamic fluctuation QT interval graph is equipped with dynamic text annotations, such as the amplitude scale, frequency scale, and time scale (seconds) of the QT interval fluctuation on the graph. The abnormal part of the QT interval fluctuation graph will appear a red curve segment.

[0044] At the same time, the large model wearable electrocardio monitoring platform automatically visualizes the normal and abnormal analysis of the wearer's QT interval graph at the current moment and the diagnosis report of the heart health status in the lower two-thirds area of the display screen, which is a large model QT interval analysis report. The position of the report on the display screen can be browsed up and down by the wearer with a light touch of the hand. The report analyzes the normal and abnormal of the wearer's QT interval status, displays statistical charts related to QT interval, and prompts the normal value, abnormal value and dangerous value of QT interval in words, displays various related knowledge that affects QT interval and various heart diseases caused by QT interval abnormalities. And after the wearer touches the three-thirds area with a light touch of the hand, the large model QT interval fluctuation analysis report can be viewed full screen.

[0045] (9) When the "RR interval" icon is clicked, the large model wearable electrocardio monitoring platform automatically displays the wearer's dynamic fluctuation RR interval graph in the upper one-third area of the display screen in real time. The dynamic fluctuation RR interval graph can be swiped left and right with a light touch of the hand. When swiping to the right, the RR interval graph before the current screen display time can be seen; when swiping to the left, the RR interval graph after the current screen display time can be seen, and it is displayed until the current monitoring time. And after the wearer touches the one-third area with a light touch of the hand, the real-time dynamic fluctuation graph of RR interval can be viewed full screen. The dynamic fluctuation RR interval graph is equipped with dynamic text annotations, such as the amplitude scale, frequency scale and time scale (seconds) of the RR interval fluctuation on the graph. The abnormal part of the RR interval fluctuation graph will appear a red curve segment.

[0046] At the same time, the large model wearable electrocardio monitoring platform automatically visualizes the normal and abnormal analysis of the wearer's QT interval graph at the current moment and the diagnosis report of the heart health status in the lower two-thirds area of the display screen, which is a large model QT interval analysis report. The position of the report on the display screen can be browsed up and down by the wearer with a light touch of the hand. The report analyzes the normal and abnormal of the wearer's QT interval status, displays statistical charts related to QT interval, and prompts the normal value, abnormal value and dangerous value of QT interval in words, displays various related knowledge that affects QT interval and various heart diseases caused by QT interval abnormalities. And after the wearer touches the three-thirds area with a light touch of the hand, the large model QT interval fluctuation analysis report can be viewed full screen.

[0047] (10) When the "electrocardio signal active monitoring" icon is clicked, the large model wearable electrocardio monitoring platform enters electrocardio signal active monitoring. At this time, the display screen displays "large model prompt: please click the target 'icon' of active monitoring" in the original "electrocardio signal active monitoring" icon position. Figure 2As shown. When the wearer clicks on the target "icon" on the display screen in the active monitoring environment of the electrocardiosignal, at this time the target "icon" can be selected. There are 9: "ECG", "heart rate", "P wave", "QRS complex", "T wave", "U wave", "PR interval", "QT interval" and "RR interval". The wearer needs to continuously press the target "icon" with his hand, which needs to be continuously pressed for about 1 minute. At this time, continuously press which target "icon", the real-time dynamic fluctuation graph of the electrocardio index corresponding to the target "icon" and the physiological index analysis, heart health status analysis and diagnosis report of the corresponding large model in the active monitoring environment will be output on the display screen in the same way as above.

[0048] Further, the hardware device includes an intelligent data integration processing chip, a blind source signal separation algorithm chip, a convolutional neural network algorithm chip, a random forest construction and traversal algorithm chip, a decision tree algorithm chip, a large model algorithm chip, a digital monitoring chip, and a large model electrocardiogram diagnosis report storage chip. The chip plan view of the large model wearable electrocardiogram monitoring device is as shown in Figure 4 The specific implementation is as follows: The intelligent data integration processing chip: by developing a small intelligent data integration processing chip, it is specially used to receive each electrocardiosignal sent out from the digital storage by the digital trigger transmitter for data integration, data preprocessing and data quality filtering.

[0049] The blind source signal separation (BSS) algorithm chip. By developing a small blind source signal separation (BSS) algorithm chip, the algorithms of blind source signal separation (BSS) such as FastICA algorithm of independent component analysis method, SOBI algorithm based on high-order statistical method and Infomax algorithm based on information theory method are integrated in the chip, and the chip is placed in the hardware structure of the large model wearable electrocardiogram monitoring device. The functions of each algorithm in the chip are as follows: (1) When the source signals are statistically independent and non-Gaussian, the FastICA algorithm of independent component analysis is used to separate the blind source signals; (2) When the source signals are correlated in time (i.e. non-white noise) and non-Gaussian, the SOBI algorithm based on high-order statistical method is used to separate the blind source signals; (3) When the mutual information between the components of the source signals is minimum and statistically independent, the Infomax algorithm based on information theory method is used to separate the blind source signals.

[0050] Then the blind source signal separation (BSS) algorithm chip is placed in the hardware structure of the large model wearable electrocardiogram monitoring device. The chip is specially used to eliminate the artifacts in the electrocardiogram source signals caused by physiological factors and environmental factors.

[0051] The convolutional neural network (CNN) algorithm chip. By developing a small convolutional neural network (CNN) algorithm chip, various sizes of convolution kernels, pooling matrices, convolutional neural network feature recognition algorithms, and convolutional neural network structures such as LeNet architecture, AlexNet architecture, VGGNet architecture, and GoogleNet architecture are integrated into the chip. The chip is placed in the hardware structure of a large model wearable electrocardiogram monitoring device. The chip is specially used for feature recognition of statistically independent signals separated by blind source separation to accurately extract physiological indicator data of electrocardiogram signals.

[0052] The random forest (RF) construction and traversal algorithm chip. By developing a small random forest (RF) construction and traversal algorithm chip, two algorithms for traversing the forest, namely depth-first traversal algorithm and breadth-first traversal algorithm, are integrated into the chip. The chip is placed in the hardware structure of a large model wearable electrocardiogram monitoring device. When constructing a random forest, statistically independent signals separated by blind source separation are used as root nodes. These root nodes contain two types of signals: (1) 9 electrocardiogram signals, namely electrocardiogram, heart rate, P wave, QRS complex, T wave, U wave, PR interval, QT interval, and RR interval, which are used to form root nodes, respectively named: "electrocardiogram tree", "heart rate tree", "P wave tree", "QRS complex tree", "T wave tree", "U wave tree", "PR interval tree", "QT interval", and "RR interval tree". The second layer nodes under each root node of the 9 trees are connected to 4 sub-trees respectively. These 4 sub-trees only have root nodes, which are: morphology, time domain, frequency domain, and amplitude. The root nodes of these 4 sub-trees are currently leaf nodes of the 9 trees. Then, the second layer leaf nodes of the 9 trees continue to connect 3 sub-trees respectively. These 3 sub-trees also only have root nodes (which are currently also leaf nodes), which are: "normal", "abnormal", and "dangerous"; (2) pseudo-signal from physiological factors and environmental factors interference separated by blind source separation. All pseudo-signal, regardless of the number, is made into a tree, called "pseudo-signal tree 1", "pseudo-signal tree 2", …, "pseudo-signal tree n" (assuming n pseudo-signal signals are separated), and all "pseudo-signal trees" only have root nodes. Then, the 9 electrocardiogram signal trees, namely "electrocardiogram tree", "heart rate tree", "P wave tree", "QRS complex tree", "T wave tree", "U wave tree", "PR interval tree", "QT interval", and "RR interval tree", and all "pseudo-signal trees" are placed together to form a "random forest".

[0053] It is explained that why it is called "random forest", because the original signal monitored by the large model wearable electrocardiogram monitoring platform is separated by the blind source signal separation algorithm after the electrocardiogram mixed signal is separated, although it is known that the separated signal has electrocardiogram signal and artifact signal, but it is not known which signal is electrocardiogram signal and which signal is artifact signal. It still needs to be further identified by the convolutional neural network algorithm to identify which signal is electrocardiogram signal and which signal is artifact signal. Because this electrocardiogram signal has a random process phenomenon in the collection process, so this electrocardiogram signal forest is called "random forest".

[0054] The decision tree (DT) algorithm chip. By developing a small decision tree (DT) algorithm chip, the current latest algorithm of decision tree, such as decision tree 5.0 or above classification algorithm and Booting pruning algorithm, are integrated in the chip, and the chip is placed in the hardware structure of the large model wearable electrocardiogram monitoring device. "Electrocardiogram signal" is taken as the root node, connecting all the electrocardiogram signal trees and artifact trees to form a "random forest-decision tree". Then the Booting pruning algorithm is used to prune the decision tree, and all the "artifact trees" are pruned. In this way, when the Booting pruning algorithm runs, the "random forest-decision tree" is left with only "electrocardiogram tree", "heart rate tree", "P wave tree", "QRS complex tree", "T wave tree", "U wave tree", "PR interval tree", "QT interval" and "RR interval tree" these 9 sub-trees, and the second layer nodes of the 9 sub-trees are accurate collected electrocardiogram data. Therefore, the chip uses the classification algorithm of decision tree 5.0 or above to identify the leaf nodes into three categories: "normal", "abnormal" or "dangerous". And the electrocardiogram indexes and their collected data that judge the leaf nodes as "normal", "abnormal" and "dangerous" are stored in one of the three region spaces of "normal region", "abnormal region" and "dangerous region".

[0055] The large model (LLM) algorithm chip. By developing a small large model (LLM) algorithm chip, the current GPT-5 or above algorithm, DeepSeek highest version and general language model algorithm such as Qianyi Qianwen are integrated in the chip, and the chip is placed in the hardware structure of the large model wearable electrocardiogram monitoring device. The chip is specially used to execute large language model generation algorithm, and the data measured by the wearer's electrocardiogram signal is subjected to large model deep search understanding, information capture, semantic splicing, typical case analysis, and forms large model semantic generation text description.

[0056] The large model electrocardiogram diagnosis report storage chip. By developing a small large model electrocardiogram diagnosis report storage chip, it is placed in the hardware structure of the large model wearable electrocardiogram monitoring device. The chip is specially used to interface the work of the large model semantic generation storage chip. The large model semantic generation text description is used to judge the health status of the wearer, and then generate a large model health diagnosis report for the wearer to finally understand their own health status. The wearer only needs to press the "large model reading key" (hardware button) on the surface of the device. At this time, the display screen will appear the prompt: "large model prompt: please click the target 'icon' of the project you want to query". The target "icon" has 9: "electrocardiogram", "heart rate", "P wave", "QRS wave group", "T wave", "U wave", "PR interval", "QT interval" and "RR interval".

[0057] (1) When the wearer clicks the "electrocardiogram" icon, the "large model electrocardiogram fluctuation analysis report" (annotated with the current monitoring date) and the "large model heart health status diagnosis report" (annotated with the current monitoring date) in a clickable state will appear in the center area of the display screen. The "large model electrocardiogram fluctuation analysis report" (annotated with the last monitoring date) and the "large model heart health status diagnosis report" (annotated with the last monitoring date) generated by the last monitoring in a clickable state also appear at the bottom of the display screen. When the wearer further clicks any one of the four diagnosis reports, the display screen will display the detailed content of the report, which can be browsed by sliding up and down.

[0058] (2) When the "heart rate" icon is clicked, the "large model heart rate fluctuation analysis report" (annotated with the current monitoring date) in a clickable state will appear in the center area of the display screen. The "large model heart rate fluctuation analysis report" (annotated with the last monitoring date) generated by the last monitoring in a clickable state also appears at the bottom of the display screen. When the wearer further clicks any one of the two diagnosis reports, the display screen will display the detailed content of the report, which can be browsed by sliding up and down.

[0059] (3) When the "P wave" icon is clicked, the "large model P wave fluctuation analysis report" (annotated with the current monitoring date) in a clickable state will appear in the center area of the display screen. The "large model P wave fluctuation analysis report" (annotated with the last monitoring date) generated by the last monitoring in a clickable state also appears at the bottom of the display screen. When the wearer further clicks any one of the two diagnosis reports, the display screen will display the detailed content of the report, which can be browsed by sliding up and down.

[0060] (4) When the "QRS complex" icon is clicked, the "Large Model QRS Complex Fluctuation Analysis Report" (annotated with the current monitoring date) in a clickable state appears in the center area of the display screen. The "Large Model QRS Complex Fluctuation Analysis Report" (annotated with the last monitoring date) generated in the last monitoring in a clickable state also appears at the bottom of the display screen. When the wearer further clicks any one of the two diagnostic reports, the detailed content of the report is displayed on the display screen, which can be browsed by scrolling up and down.

[0061] (5) When the "T wave" icon is clicked, the "Large Model T Wave Fluctuation Analysis Report" (annotated with the current monitoring date) in a clickable state appears in the center area of the display screen. The "Large Model T Wave Fluctuation Analysis Report" (annotated with the last monitoring date) generated in the last monitoring in a clickable state also appears at the bottom of the display screen. When the wearer further clicks any one of the two diagnostic reports, the detailed content of the report is displayed on the display screen, which can be browsed by scrolling up and down.

[0062] (6) When the "U wave" icon is clicked, the "Large Model U Wave Fluctuation Analysis Report" (annotated with the current monitoring date) in a clickable state appears in the center area of the display screen. The "Large Model U Wave Fluctuation Analysis Report" (annotated with the last monitoring date) generated in the last monitoring in a clickable state also appears at the bottom of the display screen. When the wearer further clicks any one of the two diagnostic reports, the detailed content of the report is displayed on the display screen, which can be browsed by scrolling up and down.

[0063] (7) When the "PR interval" icon is clicked, the "Large Model PR Interval Fluctuation Analysis Report" (annotated with the current monitoring date) in a clickable state appears in the center area of the display screen. The "Large Model PR Interval Fluctuation Analysis Report" (annotated with the last monitoring date) generated in the last monitoring in a clickable state also appears at the bottom of the display screen. When the wearer further clicks any one of the two diagnostic reports, the detailed content of the report is displayed on the display screen, which can be browsed by scrolling up and down.

[0064] (8) When the "QT interval" icon is clicked, the "Large Model QT Interval Fluctuation Analysis Report" (annotated with the current monitoring date) in a clickable state appears in the center area of the display screen. The "Large Model QT Interval Fluctuation Analysis Report" (annotated with the last monitoring date) generated in the last monitoring in a clickable state also appears at the bottom of the display screen. When the wearer further clicks any one of the two diagnostic reports, the detailed content of the report is displayed on the display screen, which can be browsed by scrolling up and down.

[0065] (9) When the "RR interval" icon is clicked, the "Large Model RR Interval Fluctuation Analysis Report" (annotated with the current monitoring date) in a clickable state will appear in the center area of the display screen. At the same time, the "Large Model RR Interval Fluctuation Analysis Report" generated by the last monitoring (annotated with the last monitoring date) in a clickable state will also appear at the bottom of the display screen. When the wearer further clicks on any one of the two diagnostic reports, the display screen will display the detailed content of the report, which can be scrolled up and down for browsing.

[0066] The plan view of the device and display screen in the "Large Model ECG Diagnosis Report" query environment is shown in Figure 3

[0067] When the wearer clicks on the target icon of the item he wants to query, the display screen will simultaneously display the large model ECG diagnosis report generated last time, which is intended to allow the wearer to compare it with the large model ECG diagnosis report generated by the current monitoring, so as to know whether the heart problem is developing in a good direction or a worse direction. For a single ECG indicator, is it getting better or worse? By comparing this diagnosis report with the previous diagnosis report, especially for the wearer of the heart patient, the most suitable suggestion can be given: whether to continue self-monitoring at home or to seek emergency medical treatment? In this way, treatment will not be delayed, and the heart patient can maximize life safety.

[0068] The digital monitoring chip. By developing a small digital monitoring chip, it is placed in the hardware structure of the large model wearable ECG monitoring device. The main work of the digital monitoring chip is: through wearing the large model wearable ECG monitoring device (in the working state), the intelligent sound and light sensor, human body sensing measurer and digital trigger sender and other hardware devices sense, read, receive, measure and send the wearer's ECG signal, i.e. ECG, heart rate, P wave, QRS complex, T wave, U wave, PR interval, QT interval and RR interval, etc., and integrate them into the large language model algorithm integrated chip for semantic generation. Finally, the "Large Model ECG Indicator Analysis Report" and "Large Model Heart Health Diagnosis Report" are generated, which are visualized on the display screen of the large model wearable ECG monitoring device through the digital monitoring chip, for the wearer to read and browse.

[0069] Further, the software platform includes 2 knowledge bases, 6 intelligent algorithm systems and 9 large model ECG signal monitoring systems, as follows: The 2 knowledge bases include a whole network search knowledge base and a heart health knowledge base. Among them: (1) Whole network search knowledge base: the whole network search knowledge base is defined as at least containing the following 6 categories of websites: ① From the medical and health websites of the whole country to the provinces, cities, counties and towns; ​② From the national to the provincial, municipal, county, town, etc. Public hospital website; ③ All private hospitals (including small clinics) in the country, including districts (counties), towns, etc. Website; ④ Baidu website, NetEase website and Sina website and other Internet websites related to heart health; ⑤ From the national to the provincial, municipal, county, etc. Health report website; ⑥ All kinds of public websites that can publish doctor forums and doctor consultations, etc.

[0070] (2) Heart health knowledge base: using large model search and generation algorithm to search the knowledge base on the network to search the relevant knowledge of heart health in depth, width, recursion, cycle and hyperlink cross-network search, search for 9 ECG monitoring indicators such as ECG, heart rate, P wave, QRS complex, T wave, U wave, PR interval, QT interval and RR interval. The relevant knowledge searched is respectively established: "ECG-knowledge base", "heart rate-knowledge base", "P-wave-knowledge base", "QRS complex-knowledge base", "T-wave-knowledge base", "U-wave-knowledge base", "PR interval-knowledge base", "QT interval-knowledge base" and "RR interval-knowledge base" 9 knowledge bases. Therefore, the heart health knowledge base is composed of the union of the 9 knowledge bases.

[0071] Further, the 6 intelligent algorithm systems include blind source signal separation algorithm system, convolutional neural network feature extraction algorithm system, random forest construction and traversal algorithm system, decision tree classification algorithm system, large model search and generation algorithm system and large model wearable ECG monitoring algorithm system. Specifically as follows: The blind source signal separation algorithm system: this system is specially responsible for the mixed ECG signals that have been preprocessed and filtered by the data quality integration processing chip through the digital trigger transmitter, and performs blind source signal separation algorithm on the mixed ECG signals, and saves the blind source signal separation result to the blind source signal separation (BSS) algorithm chip; The convolutional neural network feature extraction algorithm system: this system is specially responsible for the ECG signals that have been separated from the mixed ECG signals by the blind source signal separation (BSS) algorithm chip through the digital trigger transmitter, and performs convolutional neural network feature extraction algorithm on the ECG signals, and saves the ECG signal feature extraction result to the convolutional neural network feature extraction (CNN) algorithm chip; The random forest construction and traversal algorithm system: this system is responsible for establishing a random forest for each electrocardiogram signal and interference signal that has been extracted by the convolutional neural network feature extraction (CNN) algorithm chip through the digital trigger transmitter, and each electrocardiogram signal and interference signal is a separate tree in the random forest. The tree made of electrocardiogram signals is called "electrocardiogram signal tree", and the tree made of interference signals is called "artifact tree". The system automatically executes the random forest construction and traversal algorithm, and marks the leaf nodes of the current heart rate condition of the electrocardiogram signal tree with symbols during the traversal process, and saves the results of the traversal and the leaf node marking to the random forest (RF) construction and traversal algorithm chip. The decision tree classification algorithm system: this system is responsible for establishing a random forest for each electrocardiogram signal and interference signal that has been extracted by the convolutional neural network feature extraction (CNN) algorithm chip through the digital trigger transmitter, and each electrocardiogram signal and interference signal is a separate tree in the random forest. The tree made of electrocardiogram signals is called "electrocardiogram signal tree", and the tree made of interference signals is called "artifact tree". The system automatically executes the random forest construction and traversal algorithm, and marks the leaf nodes of the current heart rate condition of the electrocardiogram signal tree with symbols during the traversal process, and saves the results of the traversal and the leaf node marking to the random forest (RF) construction and traversal algorithm chip.

[0072] The large model search and generation algorithm system: this system is responsible for establishing a random forest for each electrocardiogram signal and interference signal that has been extracted by the convolutional neural network feature extraction (CNN) algorithm chip through the digital trigger transmitter, and each electrocardiogram signal and interference signal is a separate tree in the random forest. The tree made of electrocardiogram signals is called "electrocardiogram signal tree", and the tree made of interference signals is called "artifact tree". The system automatically executes the random forest construction and traversal algorithm, and marks the leaf nodes of the current heart rate condition of the electrocardiogram signal tree with symbols during the traversal process, and saves the results of the traversal and the leaf node marking to the random forest (RF) construction and traversal algorithm chip.

[0073] The large model wearable electrocardiogram monitoring algorithm system is responsible for displaying the fluctuation graph (a total of 9 fluctuation graphs) of the wearer's real-time dynamic electrocardiogram monitoring on the upper one-third area of the display screen of the large model wearable electrocardiogram monitoring device, so that the wearer can browse the dynamic fluctuation graph of the electrocardiogram of the past date on the display screen by randomly swiping left and right with the hand. At the same time, the display screen displays each large model electrocardiogram fluctuation analysis report (a total of 9 analysis reports) and a large model heart health status diagnosis report in the lower two-thirds area.

[0074] Further, as shown in Figure 5 The large model electrocardiogram signal monitoring principle algorithm (algorithm I, referred to as "large model electrocardiogram signal monitoring algorithm") based on the fusion of seven kinds of multi-intelligent algorithm technologies, including blind source signal separation algorithm, convolutional neural network feature extraction algorithm, random forest construction and traversal algorithm, Booting pruning algorithm, decision tree classification algorithm, large model search and generation algorithm, and large model wearable electrocardiogram monitoring algorithm, is as follows: Step 1: Perform blind source signal separation algorithm to separate the mixed signal of electrocardiogram sensed and measured from the wearer's wrist to the skin by the intelligent sound and light sensor and the human body sensor, to obtain statistically independent and non-Gaussian electrocardiogram signals (such as electrocardiogram, heart rate, P wave, QRS complex, T wave, U wave, PR interval, QT interval, and RR interval) and emotional, motion and other artifacts; Step 2: Perform convolutional neural network feature extraction algorithm to extract features from the statistically independent and non-Gaussian electrocardiogram signals and artifact interference signals separated by the blind source signal separation algorithm in step 1, and independently identify and distinguish all electrocardiogram signals such as electrocardiogram, heart rate, P wave, QRS complex, T wave, U wave, PR interval, QT interval, and RR interval, and emotional, motion and other artifact signals; Step 3: Perform random forest construction and traversal algorithm to establish an electrocardiogram signal "random forest-decision tree" according to the implementation steps of the random forest (RF) traversal algorithm chip working principle for all independent signals (electrocardiogram signals and artifact signals) identified and distinguished after convolutional neural network feature extraction; Step 4: For the constructed electrocardiogram signal "random forest-decision tree", first perform the Booting pruning algorithm to cut off the "pseudo tree" and the unmarked leaf nodes in the "electrocardiogram tree" in the decision tree; then perform the decision tree classification algorithm (such as C5,0 algorithm) to classify the measurement values of the leaf nodes (i.e. "electrocardiogram-decision tree" leaf nodes, "heart rate-decision tree" leaf nodes, "P wave-decision tree" leaf nodes, "QRS complex-decision tree" leaf nodes, "T wave-decision tree", "U wave-decision tree" leaf nodes, "PR interval-decision tree" leaf nodes, "QT interval-decision tree" leaf nodes, and "RR interval-decision tree" sub-tree leaf nodes, etc.) in the electrocardiogram signal "random forest-decision tree" into "normal", "abnormal", and "dangerous" three categories; Step 5: Perform the large model search and generation algorithm to search and access the established heart health knowledge base. At this time, the "electrocardiogram knowledge sub-base" can be accessed immediately. The large model extracts, understands, captures information, splices semantics, logically interprets charts, and forms a large model logical semantic textual description, chart display, and graphic explanation of the wearer's electrocardiogram. Finally, it generates a "large model electrocardiogram fluctuation analysis report" and a "large model heart health status diagnosis report"; Step 6: Perform the large model wearable electrocardiogram monitoring algorithm to visualize the electrocardiogram on the top one-third area of the display screen of the large model wearable electrocardiogram monitoring device, allowing the wearer to browse the dynamic fluctuation graph of the past date electrocardiogram by swiping left and right on the display screen. At the same time, the "large model electrocardiogram fluctuation analysis report" and the "large model heart health status diagnosis report" are displayed in the lower two-thirds area of the display screen. Moreover, whether it is the top one-third area of the display screen or the lower two-thirds area of the display screen, the wearer can display the corresponding fluctuation graph and the corresponding large model analysis report on the full screen by gently touching the screen.

[0075] Further, as shown in Figure 6 The 9 large model electrocardiogram signal monitoring systems are: a large model electrocardiogram monitoring system, a large model heart rate monitoring system, a large model P wave monitoring system, a large model QRS complex monitoring system, a large model T wave monitoring system, a large model U wave monitoring system, a large model PR interval monitoring system, a large model QT interval monitoring system, and a large model RR interval monitoring system. The 9 electrocardiogram signals (parameters) of the wearer, i.e. electrocardiogram, heart rate, P wave, QRS complex, T wave, U wave, PR interval, QT interval, and RR interval, are set with the same time monitoring granularity. For example, the time monitoring granularity of the 9 electrocardiogram parameters is set to monitor every 5 minutes, or every 10 minutes. And so on! Details are as follows: The big model ECG monitoring system is specially used to realize the real-time monitoring function of the big model wearable ECG monitoring platform to the wearer's ECG. The wearer can click the "ECG" icon (software icon, see Figure 1 ) on the display screen with his hand, and the system will immediately work to visualize the real-time dynamic fluctuation graph of the ECG in the upper one-third area of the display screen. The wearer can swipe left and right on the display screen at will to browse the dynamic fluctuation monitoring graph of the ECG of any past date. The big model ECG signal monitoring working principle algorithm (algorithm II) is implemented as follows: Step 1: Execute steps (1)-(4) of the big model ECG signal monitoring algorithm to generate an "ECG-decision tree". The measurement values of the leaf nodes of the "ECG-decision tree" have been classified into "normal", "abnormal" and "dangerous". That is, the "ECG-decision tree" has only three leaf nodes, which are the three judgment results of "normal", "abnormal" and "dangerous".

[0076] Step 2: Execute step (5) of the big model ECG signal monitoring algorithm to search and access the established heart health knowledge base, and immediately access the "ECG-knowledge sub-base". The big model extracts, understands, captures information, splices semantics, logically interprets the charts in the sub-base to form the big model logical semantic textual description, chart display and graphic explanation of the wearer's ECG, and finally generates a big model ECG fluctuation analysis report; Step 3: Execute step (6) of the big model ECG signal monitoring algorithm to visualize the ECG in the upper one-third area of the display screen of the big model wearable ECG monitoring device, and the wearer can swipe left and right on the display screen at will to browse the dynamic fluctuation graph of the ECG of past dates. At the same time, the "big model ECG fluctuation analysis report" is displayed in the lower two-thirds area of the display screen. Whether it is the upper one-third area of the display screen or the lower two-thirds area of the display screen, the wearer can display the corresponding fluctuation graph or the corresponding big model analysis report on the full screen by touching it gently with his hand.

[0077] The big model heart rate monitoring system is specially used to realize the real-time monitoring function of the big model wearable ECG monitoring platform to the wearer's heart rate. The wearer can click the "heart rate" icon (software icon, see Figure 1 ) on the display screen with his hand, and the system will immediately work to visualize the real-time dynamic fluctuation graph of the heart rate in the upper one-third area of the display screen. The wearer can swipe left and right on the display screen at will to browse the dynamic fluctuation monitoring graph of the heart rate of any past date. The big model heart rate monitoring working principle algorithm (algorithm III) is implemented as follows: Step 1: Perform steps (1)-(4) of the large model ECG signal monitoring algorithm, which can generate a "heart rate-decision tree". And the measurement values of the leaf nodes of the heart rate-decision tree have been classified into three categories: "normal", "abnormal" and "dangerous". That is, the "heart rate-decision tree" has only three leaf nodes, which are the three judgment results, respectively "normal", "abnormal" and "dangerous".

[0078] Step 2: Perform step (5) of the large model ECG signal monitoring algorithm, search and access the established heart health knowledge base, and immediately access the "heart rate-knowledge base". The large model extracts, understands, captures information, and semantically splices the normal range, abnormal range and dangerous range of the heart rate signal in this sub-base, forms a large model logical semantic textual description of the wearer's heart rate, displays charts and provides graphic explanations, and finally generates a large model heart rate fluctuation analysis report; Step 3: Perform step (6) of the large model wearable ECG signal monitoring algorithm, visualize the heart rate in the upper one-third area of the display screen of the large model wearable ECG monitoring device, and display it for the wearer to swipe left and right on the display screen at will, browse the dynamic fluctuation graph of the heart rate of the past date. At the same time, display the "large model heart rate fluctuation analysis report" in the lower two-thirds area of the display screen. And no matter whether it is the upper one-third area of the display screen or the lower two-thirds area of the display screen, the wearer can display the corresponding fluctuation graph or the corresponding large model analysis report on the full screen by gently touching it with his hand.

[0079] The large model P wave monitoring system is specially used to realize the real-time monitoring function of the large model wearable ECG monitoring platform to the wearer's P wave. The wearer can click the "P wave" icon (software icon, see Figure 1 ) on the display screen with his hand, and the system will immediately work to visualize the real-time dynamic fluctuation graph of the P wave in the upper one-third area of the display screen. For the wearer to swipe left and right on the display screen at will, browse the dynamic fluctuation monitoring graph of the P wave of any past date. The large model P wave monitoring working principle algorithm (algorithm IV) is implemented as follows: Step 1: Perform steps (1)-(4) of the large model ECG signal monitoring algorithm, which can generate a "P wave-decision tree". And the measurement values of the leaf nodes of the P wave-decision tree have been classified into three categories: "normal", "abnormal" and "dangerous". That is, the "P wave-decision tree" has only three leaf nodes, which are the three judgment results, respectively "normal", "abnormal" and "dangerous".

[0080] Step 2: Perform the large model ECG signal monitoring algorithm Step (5), search and access the established heart health knowledge base, and immediately access the "P-wave-knowledge sub-library". The large model extracts, understands, information captures, semantic splices, and logically interprets the normal range, abnormal range, and dangerous range of P-wave signals in this sub-library, forms a large model logical semantic textual description of the wearer's P-wave, displays charts and texts, and finally generates a large model P-wave fluctuation analysis report; Step 3: Perform the large model wearable ECG signal monitoring algorithm Step (6), visualize the P-wave in the top one-third area of the display screen of the large model wearable ECG monitoring device, allowing the wearer to swipe left and right on the display screen with their hand at will to view the dynamic fluctuation graph of the P-wave on past dates. At the same time, display the "large model P-wave fluctuation analysis report" in the lower two-thirds area of the display screen. Whether it is the top one-third area of the display screen or the lower two-thirds area of the display screen, the wearer can display the corresponding fluctuation graph or the corresponding large model analysis report on the full screen by gently touching it with their hand.

[0081] The large model QRS complex monitoring system is specially designed to realize the real-time monitoring function of the large model wearable ECG monitoring platform for the wearer's QRS complex. The wearer can click the "QRS complex" icon (software icon, see Figure 1 ) on the display screen with their hand, and the system will immediately work to visualize the real-time dynamic fluctuation graph of the QRS complex in the top one-third area of the display screen. The wearer can swipe left and right on the display screen with their hand at will to view the dynamic fluctuation monitoring graph of the QRS complex on any past date. The large model QRS complex monitoring working principle algorithm (Algorithm V) is implemented as follows: Step 1: Perform steps (1)-(4) of the large model ECG signal monitoring algorithm to generate a "QRS complex-decision tree". The measurement values of the leaf nodes of the QRS complex-decision tree have been classified into "normal", "abnormal", and "dangerous". This means that the "QRS complex-decision tree" has only three leaf nodes, which are the three judgment results, "normal", "abnormal", and "dangerous".

[0082] Step 2: Perform the large model ECG signal monitoring algorithm Step (5), search and access the established heart health knowledge base, and immediately access the "QRS complex-knowledge sub-library". The large model extracts, understands, information captures, semantic splices, and logically interprets the normal range, abnormal range, and dangerous range of QRS complex signals in this sub-library, forms a large model logical semantic textual description of the wearer's QRS complex, displays charts and texts, and finally generates a large model QRS complex fluctuation analysis report; Step 3: Perform the wearable ECG signal monitoring algorithm of the large model Step (6), visualize the QRS complex in the upper third of the display screen of the large model wearable ECG monitoring device, and display it for the wearer to swipe left and right on the display screen with his hand at will, and browse the dynamic fluctuation graph of the QRS complex of the past date. At the same time, display the "large model QRS complex fluctuation analysis report" in the lower two-thirds of the display screen. And no matter whether it is the upper third of the display screen or the lower two-thirds of the display screen, the wearer can display the corresponding fluctuation graph or the corresponding large model analysis report on the full screen by gently touching it with his hand.

[0083] The large model T wave monitoring system is specially used to realize the real-time monitoring function of the large model wearable ECG monitoring platform to the wearer's T wave. The wearer can click the "T wave" icon (software icon, see Figure 1 ) on the display screen with his hand, and the system will immediately work to visualize the real-time dynamic fluctuation graph of the T wave in the upper third of the display screen. For the wearer to swipe left and right on the display screen with his hand at will, and browse the dynamic fluctuation monitoring graph of the T wave of any past date. The working principle algorithm of the large model T wave monitoring (algorithm VI) is as follows: Step 1: Perform steps (1)-(4) of the large model ECG signal monitoring algorithm to generate a "T wave-decision tree". And the measurement values of the leaf nodes of the T wave-decision tree have been classified into "normal", "abnormal" and "dangerous". That is, the "T wave-decision tree" has only three leaf nodes, which are three judgment results, "normal", "abnormal" and "dangerous".

[0084] Step 2: Perform step (5) of the large model ECG signal monitoring algorithm to search and access the established heart health knowledge base, and immediately access the "T wave-knowledge base". The large model extracts, understands, captures information, splices semantics, logically interprets charts, forms a large model logical semantic textual description of the wearer's T wave, chart display and graphic explanation, and finally generates a large model T wave fluctuation analysis report in the sub-base; Step 3: Perform the wearable ECG signal monitoring algorithm of the large model Step (6), visualize the T wave in the upper third of the display screen of the large model wearable ECG monitoring device, and display it for the wearer to swipe left and right on the display screen with his hand at will, and browse the dynamic fluctuation graph of the T wave of the past date. At the same time, display the "large model T wave fluctuation analysis report" in the lower two-thirds of the display screen. And no matter whether it is the upper third of the display screen or the lower two-thirds of the display screen, the wearer can display the corresponding fluctuation graph or the corresponding large model analysis report on the full screen by gently touching it with his hand.

[0085] The big model U wave monitoring system is specially used to realize the real-time monitoring function of the big model wearable electrocardio monitoring platform to the U wave of the wearer. The wearer can click the "U wave" icon (software icon, see Figure 1 ) on the display screen with his hand, and the system will immediately work to visualize the real-time dynamic fluctuation graph of the U wave on the upper one-third area of the display screen. The wearer can swipe left and right on the display screen at will to browse the dynamic fluctuation monitoring graph of the U wave of any past date. The working principle algorithm (algorithm VII) of the big model U wave monitoring is implemented as follows: Step 1: execute steps (1)-(4) of the big model electrocardio signal monitoring algorithm to generate a "U wave-decision tree". The measurement values of the leaf nodes of the U wave-decision tree have been classified into three categories: "normal", "abnormal" and "dangerous". That is, the "U wave-decision tree" has only three leaf nodes, which are three judgment results, respectively "normal", "abnormal" and "dangerous".

[0086] Step 2: execute step (5) of the big model electrocardio signal monitoring algorithm to search and access the established heart health knowledge base, and immediately access the "U wave-knowledge base". The big model extracts, understands, captures information, splices semantics, logically interprets charts, forms the big model logical semantic textual description of the U wave of the wearer, chart display and graphic explanation, and finally generates the fluctuation analysis report of the big model U wave; Step 3: execute step (6) of the big model wearable electrocardio signal monitoring algorithm to visualize the U wave on the upper one-third area of the display screen of the big model wearable electrocardio monitoring device, so that the wearer can swipe left and right on the display screen at will to browse the dynamic fluctuation graph of the U wave of the past date. At the same time, the "fluctuation analysis report of the big model U wave" is displayed on the lower two-thirds area of the display screen. No matter the upper one-third area of the display screen or the lower two-thirds area of the display screen, the wearer can display the corresponding fluctuation graph or the corresponding big model analysis report on the full screen by touching it gently with his hand.

[0087] The big model PR interval monitoring system is specially used to realize the real-time monitoring function of the big model wearable electrocardio monitoring platform to the PR interval of the wearer. The wearer can click the "PR interval" icon (software icon, see Figure 1 ) on the display screen with his hand, and the system will immediately work to visualize the real-time dynamic fluctuation graph of the PR interval on the upper one-third area of the display screen. The wearer can swipe left and right on the display screen at will to browse the dynamic fluctuation monitoring graph of the PR interval of any past date. The working principle algorithm (algorithm VIII) of the big model PR interval monitoring is implemented as follows: Step 1: Perform steps (1)-(4) of the large model ECG signal monitoring algorithm, which can generate a "PR interval-decision tree". And the measurement values of the leaf nodes of the PR interval-decision tree have been classified into three categories: "normal", "abnormal" and "dangerous". That is, the "PR interval-decision tree" has only three leaf nodes, which are three judgment results, respectively "normal", "abnormal" and "dangerous".

[0088] Step 2: Perform step (5) of the large model ECG signal monitoring algorithm to search and access the established heart health knowledge base, and immediately access the "PR interval-knowledge base". The large model extracts and understands the normal range, abnormal range and dangerous range of the PR interval signal in this sub-base, captures information, splices semantics, interprets charts and logic, forms a large model logical semantic textual description of the wearer's PR interval, chart display and graphic explanation, and finally generates a large model PR interval fluctuation analysis report; Step 3: Perform step (6) of the large model wearable ECG signal monitoring algorithm to visualize the PR interval in the upper one-third area of the display screen of the large model wearable ECG monitoring device, allowing the wearer to swipe left and right on the display screen with their hand to view the dynamic fluctuation graph of the PR interval on past dates. At the same time, the "large model PR interval fluctuation analysis report" is displayed in the lower two-thirds area of the display screen. And whether it is the upper one-third area of the display screen or the lower two-thirds area of the display screen, the wearer can display the corresponding fluctuation graph or the corresponding large model analysis report on the full screen by gently touching it with their hand.

[0089] The large model QT interval monitoring system is specially designed to realize the real-time monitoring function of the large model wearable ECG monitoring platform for the wearer's QT interval. The wearer can click the "QT interval" icon (software icon, see Figure 1 ) on the display screen with their hand, and the system will immediately work to visualize the real-time dynamic fluctuation graph of the QT interval in the upper one-third area of the display screen. The wearer can swipe left and right on the display screen with their hand to view the dynamic fluctuation monitoring graph of the QT interval on any past date. The large model QT interval monitoring working principle algorithm (algorithm IX) is implemented as follows: Step 1: Perform steps (1)-(4) of the large model ECG signal monitoring algorithm, which can generate a "QT interval-decision tree". And the measurement values of the leaf nodes of the QT interval-decision tree have been classified into three categories: "normal", "abnormal" and "dangerous". That is, the "QT interval-decision tree" has only three leaf nodes, which are three judgment results, respectively "normal", "abnormal" and "dangerous".

[0090] Step 2: Perform the large model ECG signal monitoring algorithm Step (5), search and access the established heart health knowledge base, and immediately access the "QT interval-knowledge sub-library". The large model extracts, understands, information captures, semantic splices, and logically interprets the normal range, abnormal range, and dangerous range of the QT interval signal in this sub-library, forms a large model logical semantic textual description of the wearer's QT interval, displays charts and text instructions, and finally generates a large model QT interval fluctuation analysis report; Step 3: Perform the large model wearable ECG signal monitoring algorithm Step (6), visualize the QT interval in the top one-third area of the display screen of the large model wearable ECG monitoring device, allowing the wearer to swipe left and right on the display screen with their hand at will to view the dynamic fluctuation graph of the QT interval on past dates. At the same time, display the "large model QT interval fluctuation analysis report" in the lower two-thirds area of the display screen. Whether it is the top one-third area of the display screen or the lower two-thirds area of the display screen, the wearer can display the corresponding fluctuation graph or the corresponding large model analysis report on the full screen by gently touching it with their hand.

[0091] The large model RR interval monitoring system is specifically designed to realize the real-time monitoring function of the large model wearable ECG monitoring platform for the wearer's RR interval. The wearer can click the "RR interval" icon (software icon, see Figure 1 ) on the display screen with their hand, and the system will immediately work to visualize the real-time dynamic fluctuation graph of the RR interval in the top one-third area of the display screen. The wearer can swipe left and right on the display screen with their hand at will to view the dynamic fluctuation monitoring graph of the RR interval on any past date. The large model RR interval monitoring working principle algorithm (Algorithm X) is implemented as follows: Step 1: Perform steps (1)-(4) of the large model ECG signal monitoring algorithm to generate an "RR interval-decision tree". The measurement values of the leaf nodes of the RR interval-decision tree have been classified into "normal", "abnormal", and "dangerous" categories. This means that the "RR interval-decision tree" has only three leaf nodes, which are the three judgment results, "normal", "abnormal", and "dangerous".

[0092] Step 2: Perform the large model ECG signal monitoring algorithm Step (5), search and access the established heart health knowledge base, and immediately access the "RR interval-knowledge sub-library". The large model extracts, understands, information captures, semantic splices, and logically interprets the normal range, abnormal range, and dangerous range of the QT interval signal in this sub-library, forms a large model logical semantic textual description of the wearer's QT interval, displays charts and text instructions, and finally generates a large model QT interval fluctuation analysis report; Step 3: Execute Step (6) of the large-model wearable ECG signal monitoring algorithm. Visualize the RR interval in the upper third of the large-model wearable ECG monitoring device's display screen, allowing the wearer to freely swipe left and right on the screen to browse the dynamic fluctuation graph of the RR interval over past dates. At the same time, display the "Large-Model RR Interval Fluctuation Analysis Report" in the lower two-thirds of the display screen. Furthermore, whether it is the upper third of the display screen or the lower two-thirds of the display screen, the wearer can simply touch it to display the corresponding fluctuation graph or the corresponding large-model analysis report in full screen.

[0093] like Figure 7 , Figure 8 As shown, in another embodiment of this application, a method for large-scale ECG monitoring of wearable health devices based on BSS-CNN-RF-DT-LLM fusion technology is provided, including the following method: Step 1: The wearer puts on the large-model wearable ECG monitoring device, turns on the device, and puts the device into normal working condition. Step 2: The intelligent sound and light sensor detects the wearer's real-time health status by measuring various ECG signals, including ECG, heart rate, P wave, QRS complex, T wave, U wave, PR interval, QT interval, and RR interval. Step 3: The human body sensor measures various physiological indicators of the wearer; Step 4: After receiving the real-time measurement values ​​of various electrocardiogram signals, the digital receiver transmits them to the digital memory, stores the data, and saves it in the device's backend. Step 5: The digital memory transmits the stored ECG signal measurements (including mixed signal measurements with interference) to the intelligent data integration processing chip via a digital trigger transmitter, where data preprocessing and data quality filtering are performed. Step 6: The system activates the Blind Source Separation (BSS) algorithm chip, and the BSS algorithm system enters working mode, automatically executing the BSS algorithm. It performs blind source separation on the ECG signal (a mixed ECG signal containing interference), separating statistically independent and non-Gaussian distributed independent signal sources. Step 7: The system starts the Convolutional Neural Network (CNN) algorithm chip, and the CNN algorithm system enters the working state, automatically executing the CNN algorithm to perform feature recognition on the independent signal sources separated in Step 6, and identify the characteristics of ECG signals and interference signals; Step 8: The system starts the random forest (RF) construction and traversal algorithm chip, so that the random forest construction and traversal algorithm system enters the working state, and the independent signal source separated in step 6 (with electrocardio signals and motion and emotion and other artifact signals) is made into a root node. How many independent signals are there, how many root nodes are there. Further identify the root nodes with electrocardio signal characteristics in step 7, and each root node is respectively connected to 4 sub-nodes: morphology, time domain, frequency domain and amplitude, as the second layer nodes of the electrocardio signal tree. For the second layer nodes of each electrocardio signal tree, each node is respectively connected to three classification nodes of the electrocardio signal monitoring results: "normal", "abnormal" and "dangerous". In this way, "electrocardiogram tree", "heart rate tree", "P wave tree", "QRS complex tree", "T wave tree", "U wave tree", "PR interval tree", "QT interval tree" and "RR interval tree" are obtained. The root nodes further identified in step 7 with artifact signal characteristics are an independent tree (or multiple trees), called "artifact tree". How many interference signals are there, how many "artifact trees" are there. "Artifact tree" only has a root node. Further, all "electrocardio signal trees" and all "artifact trees" are put together to form a random forest.

[0094] It can be understood that the reason why this forest is called "random forest" is that after the original signals monitored by the large model wearable electrocardio monitoring platform are separated by the blind source signal separation algorithm in step 6, although it is known that the separated signals have electrocardio signals and artifact signals, it is not known which signal is an electrocardio signal and which signal is an artifact signal. This still needs to be further identified by the convolutional neural network algorithm in step 7. In this way, it can be identified which signal is an electrocardio signal and which signal is an artifact signal. Because this electrocardio signal acquisition process has the phenomenon of random process, the obtained forest is called "random forest".

[0095] Step 9: The system starts the decision tree (DT) classification algorithm chip, and the decision tree classification algorithm system enters the working state. First, a "ECG signal" is automatically added as a root node in the random forest established in step 8, and then this root node is connected to each tree in the already built random forest, i.e., "ECG tree", "heart rate tree", "P wave tree", "QRS complex tree", "T wave tree", "U wave tree", "PR interval tree", "QT interval tree", and "RR interval tree", and all "artifact trees". In this way, an "ECG signal-decision tree" is constructed. This "ECG signal-decision tree" is obviously composed of the "ECG signal" root node and the "random forest" constructed in step 8 of the algorithm. Therefore, when traversing the "ECG signal-decision tree", the "ECG signal" root node is not considered, and the forest traversal algorithm is directly used to traverse the "random forest" constructed in step 8 under the "ECG signal" root node. During the traversal process, when a leaf node is accessed, the actual monitoring value of the "ECG signal" root node corresponding to the leaf node is compared and matched with the "normal", "abnormal", and "dangerous" three classification ranges. The leaf node of the current heart rate condition is symbolically labeled, and then the matching result is saved as a ternary array (ECG signal, monitoring value, judgment range) to the decision tree (DT) classification algorithm chip. Step 10: The system starts the large model ECG diagnosis report storage chip, and the large model search and generation algorithm system enters the working state. For the monitored ECG signal, the large model searches the entire network, extracts the normal range, abnormal range, dangerous range, and typical cases of the corresponding ECG signal in the knowledge base for deep mining, understanding, comparison, information capture, semantic splicing, chart interpretation, forms a large model semantic generation text description, and finally generates a large model ECG signal analysis report and a large model heart health status report, which are saved in the large model ECG diagnosis report storage chip.

[0096] Step 11: The system starts the digital monitoring chip, and the large model wearable ECG monitoring algorithm system enters the working state. The real-time monitored ECG of the wearer is displayed on the upper one-third area of the large model wearable ECG monitoring device display screen, and the "large model ECG signal analysis report" and "large model heart health status diagnosis report" are displayed on the lower two-thirds area of the large model wearable ECG monitoring device display screen.

[0097] Example 2 This example 2 applies the large model wearable ECG monitoring platform to monitor and display heart rate, including the following steps: (1) Wear the large model wearable ECG monitoring device with one hand, and turn on the device to be in the working state; (2) Use the other hand to swipe the display screen (1) of the device Figure 1) gently click on the "heart rate" icon (software button), if it is active monitoring needs to continuously press this "heart rate" icon for about 1 minute. At this time, the display screen displays "the big model is working, it takes about 1 minute. Please wait, do not loose hand……”, as shown in Figure 9 .

[0098] (3) when the big model work is finished, the big model wearable ECG monitoring platform automatically displays the wearer's heart rate dynamic fluctuation graph in the upper one-third area of the display screen in real time, which can be moved left and right by hand; In the lower two-thirds area of the display screen, the normal and abnormal analysis of the wearer's heart rate at the current time and the diagnosis report of the heart health status are visualized and output, which is a "big model heart rate fluctuation analysis report", which can be moved up and down by hand.

[0099] For example, the heart rate fluctuation graph of a wearer is as shown in Figure 10 , because this heart rate fluctuation curve is abnormal, this curve segment is red.

[0100] (4) if the wearer clicks on the upper one-third position of the display screen with his hand, the wearer's heart rate fluctuation graph will occupy the screen on the display screen. For example, the minimum heart rate of this wearer is 99 times / min, the maximum heart rate is 150 times / min, and the average heart rate is 120 times / min, which are displayed on the display screen. Because this heart rate curve is abnormal, the heart rate fluctuation graph is red, as shown in Figure 11 .

[0101] (5) if the wearer clicks on the lower two-thirds position of the display screen with his hand, the text content of the "big model heart rate fluctuation analysis report" will occupy the entire screen on the display screen, as shown in Figure 12 .

[0102] Example 3 This embodiment 3 utilizes the big model wearable ECG monitoring platform to monitor and display QRS complex, as follows: (1) wear the big model wearable ECG monitoring device with one hand, turn on the device to work; (2) gently click on the "QRS complex" icon (software button) on the display screen of the device Figure 1 with the other hand, if it is active monitoring needs to continuously press this "QRS complex" icon for about 1 minute. At this time, the display screen displays "the big model is working, it takes about 1 minute. Please wait, do not loose hand……”, as shown in Figure 13 .

[0103] (3) When the large model work is finished, the large model wearable ECG monitoring platform automatically displays the wearer's QRS complex fluctuation graph in real time in the upper one-third area of the display screen, which can be swiped left and right by hand; in the lower two-thirds area of the display screen, the normal and abnormal analysis of the wearer's ECG at the current time and the diagnosis report of the heart health status are visualized and output, which is a large model QRS complex fluctuation analysis report, which can be swiped up and down by hand.

[0104] For example, the QRS complex fluctuation graph of a wearer is as shown in Figure 14 Because this QRS complex curve has an abnormal state, the abnormal part of this curve segment is red.

[0105] (4) If the wearer clicks the upper one-third position of the display screen with his hand, the wearer's QRS complex fluctuation graph will occupy the entire screen on the display screen. For example, the normal cycle time of QRS complex is 0.04s. Then, in a 5-cycle monitoring of this wearer, it is found that the 3rd-5th cycle is abnormal, so the QRS complex fluctuation graph of the wearer in the 3rd-5th cycle is red, as shown in Figure 15

[0106] (5) If the wearer clicks the lower two-thirds position of the display screen with his hand, the text content of the large model QRS complex fluctuation analysis report will occupy the entire screen on the display screen, as shown in Figure 16

[0107] ​​Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The program can be stored in a non-volatile computer readable storage medium, and when the program is executed, the processes of the above-mentioned embodiment methods can be included. Any reference to memory, storage, database or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0108] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

[0109] The above embodiments are the preferred embodiments of the present application, but the embodiments of the present application are not limited to the above embodiments, and any changes, modifications, substitutions, combinations and simplifications made without departing from the spirit and principles of the present application should be equivalent replacement methods, and are included in the protection scope of the present application.

Claims

1. A wearable health device large model electrocardio monitoring system based on BSS-CNN-RF-DT-LLM fusion technology, characterized in that, The hardware device comprises an intelligent data integration processing chip, a blind source signal separation (BSS) algorithm chip, a convolutional neural network (CNN) algorithm chip, a random forest (RF) construction and traversal algorithm chip, a decision tree (DT) algorithm chip, a large language model (LLM) algorithm chip, a digital monitoring chip, and a large model electrocardiogram diagnosis report storage chip. The intelligent data integration processing chip is configured to receive each electrocardiogram signal sent by the digital trigger transmitter from the digital storage, and perform data integration, data preprocessing, and data quality filtering. The BSS algorithm chip integrates a FastICA algorithm of an independent component analysis method, a SOBI algorithm based on a high-order statistical method, and an Infomax algorithm based on an information theory method, and is configured to eliminate artifacts in electrocardiogram source signals caused by physiological factors and environmental factors. The CNN algorithm chip integrates different sizes of convolution kernels, pooling matrices, a CNN feature recognition algorithm, and LeNet, AlexNet, VGGNet, and GoogleNet architectures constructed by the CNN, and is configured to perform feature recognition on the statistically independent signals obtained by the BSS, and accurately extract physiological indicator data of the electrocardiogram signals. The RF construction and traversal algorithm chip integrates a depth-first traversal algorithm and a width-first traversal algorithm, and is configured to construct a random forest with the statistically independent signals obtained by the BSS as root nodes, wherein the root nodes include root nodes corresponding to the electrocardiogram signals and root nodes corresponding to artifact signals, the second layer nodes under the root nodes corresponding to the electrocardiogram signals are connected to morphology, time domain, frequency domain, and amplitude sub-trees, the second layer nodes are connected to actual monitoring values of morphology, time domain, frequency domain, and amplitude, respectively, as third layer nodes, the third layer nodes are connected to "normal", "abnormal", and "dangerous" sub-trees as fourth layer nodes, and the fourth layer nodes are leaf nodes. The DT algorithm chip integrates a DT algorithm, a classification algorithm, and a Booting pruning algorithm, and is configured to connect each tree in the random forest to form a "random forest-DT" with the electrocardiogram signals as the root node, prune the "artifact tree" by the Booting pruning algorithm, identify the leaf nodes as "normal", "abnormal", and "dangerous", and store the corresponding electrocardiogram indicators and collected data in corresponding regional spaces. The LLM algorithm chip integrates an LLM algorithm, and is configured to perform LLM deep search language understanding, information capture, semantic splicing, and typical case analysis on the electrocardiogram measurement data, and form a large model semantic generation text description. The large model electrocardiogram diagnosis report storage chip is configured to interface with the large model semantic generation storage chip, determine the health status of the wearer according to the text description generated by the large model semantic generation, and generate a large model health diagnosis report. The digital monitoring chip is used for the sensing, reading, receiving, measuring and sending of the electrocardio signal by the intelligent sound and light sensor, the human body sensing measurer and the digital trigger sender, semantic generation by the large language model algorithm integrated chip, generation of the electrocardio index analysis report and the heart health condition diagnosis report, and visual display on the display screen; The display screen is used for displaying the visual real-time dynamic fluctuation graph of the morphology, time domain, frequency domain and amplitude of the electrocardio index, and is provided with "electrocardiogram", "heart rate", "P wave", "QRS wave group", "T wave", "U wave", "PR interval", "QT interval", "RR interval" and "electrocardio signal active monitoring" icons; clicking the first 9 software icons can display the corresponding electrocardio index dynamic fluctuation graph and large model analysis diagnosis report; after clicking the "electrocardio signal active monitoring" icon, selecting the target icon and setting the time, the corresponding electrocardio index dynamic fluctuation graph and large model analysis diagnosis report under active monitoring are displayed; The software platform comprises a knowledge base, an intelligent algorithm system and a large model electrocardio signal monitoring system; the knowledge base comprises a whole network search knowledge base and a heart health knowledge base; the intelligent algorithm system comprises a blind source signal separation algorithm system, a convolutional neural network feature extraction algorithm system, a random forest construction and traversal algorithm system, a decision tree classification algorithm system, a large model search and generation algorithm system and a large model wearable electrocardio monitoring algorithm system; the large model electrocardio signal monitoring system comprises a large model electrocardiogram monitoring system, a large model heart rate monitoring system, a large model P wave monitoring system, a large model QRS wave group monitoring system, a large model T wave monitoring system, a large model U wave monitoring system, a large model PR interval monitoring system, a large model QT interval monitoring system and a large model RR interval monitoring system, each system is used for real-time monitoring of the corresponding electrocardio signal, generating the corresponding decision tree and analysis report, and displaying the dynamic fluctuation graph and analysis report in the specified area of the display screen, and each system is set with the same time monitoring granularity.

2. The wearable health device large model electrocardio monitoring system based on the BSS-CNN-RF-DT-LLM fusion technology according to claim 1, wherein, The electrocardio signal comprises an electrocardiogram, a heart rate, a P wave, a QRS wave group, a T wave, a U wave, a PR interval, a QT interval and an RR interval, and is respectively made into a tree with only a root node, and is respectively called an "electrocardiogram tree", a "heart rate tree", a "P wave tree", a "QRS wave group tree", a "T wave tree", a "U wave tree", a "PR interval tree", a "QT interval tree" and an "RR interval tree"; the 9 trees currently only have root nodes; then, the root nodes of each tree are respectively connected to 4 sub-trees; the measurement values of the morphology, time domain, frequency domain and amplitude of the electrocardio signal; the 4 sub-trees also currently only have root nodes, and the root nodes are respectively made into the second layer nodes of the 9 trees, which are currently leaf nodes; then, the second layer leaf nodes of the 9 trees are respectively connected to 3 sub-trees, which are "normal", "abnormal" and "dangerous"; the 3 sub-trees also only have root nodes, and are respectively made into the third layer nodes of the 9 trees, which are leaf nodes; The artifact signals are made into a tree alone, called "artifact tree 1", "artifact tree 2", …, "artifact tree n", n represents that there are n independent artifact signals, and all "artifact trees" only have root nodes; then, the nine electrocardiosignal trees, i.e. "electrocardiogram tree", "heart rate tree", "P wave tree", "QRS complex tree", "T wave tree", "U wave tree", "PR interval tree" and "RR interval tree", and all "artifact trees" are put together to form a "random forest".

3. The wearable health device large model electrocardio monitoring system based on the BSS-CNN-RF-DT-LLM fusion technology according to claim 1, wherein, The full-network search knowledge base covers medical and health websites, public hospital websites, private hospital websites, internet websites related to heart health, health newspaper websites and public websites that can publish doctor forums and online consultations nationwide to province, city, county and town; the heart health knowledge base is composed of "electrocardiogram-knowledge sub-base", "heart rate-knowledge sub-base", "P wave-knowledge sub-base", "QRS complex-knowledge sub-base", "T wave-knowledge sub-base", "U wave-knowledge sub-base", "PR interval-knowledge sub-base", "QT interval-knowledge sub-base" and "RR interval-knowledge sub-base", and each sub-base searches for the corresponding electrocardiosignal monitoring index in the full-network search knowledge base to establish the related knowledge base through the large model search and generation algorithm.

4. The wearable health device large model electrocardio monitoring system based on the BSS-CNN-RF-DT-LLM fusion technology according to claim 1, wherein, The blind source signal separation algorithm system is used for executing a blind source signal separation algorithm on the mixed electrocardiosignal preprocessed and filtered by the intelligent data integration processing chip, and saving the result to a BSS algorithm chip; the convolutional neural network feature extraction algorithm system is used for executing a convolutional neural network feature extraction algorithm on the signal separated by the BSS algorithm, and saving the result to a CNN algorithm chip; the random forest construction and traversal algorithm system is used for establishing a random forest for the signal whose features are extracted by the CNN algorithm, traversing and marking the leaf node of the current heart rate condition, and saving the result to a RF construction and traversal algorithm chip; the decision tree classification algorithm system is used for connecting the random forest with "electrocardiosignal monitoring" as a root node to form a "random forest-decision tree", cutting off the "artifact tree" and classifying the leaf nodes, and saving the result to a DT classification algorithm chip; the large model search and generation algorithm system is used for generating a health diagnosis report for the data classified by the DT classification algorithm chip in combination with the health knowledge base, and saving the result to a large model electrocardiosignal diagnosis report storage chip; and the large model wearable electrocardiosignal monitoring algorithm system is used for displaying an electrocardiogram dynamic fluctuation graph and various large model analysis diagnosis reports in a specified area of a display screen.

5. The wearable health device large model electrocardio monitoring system based on the BSS-CNN-RF-DT-LLM fusion technology according to claim 1, wherein, In the large model electrocardiosignal monitoring system, The large model electrocardiogram monitoring system is used for monitoring an electrocardiogram signal in real time, generating a decision tree and an analysis report corresponding to the electrocardiogram, and displaying an electrocardiogram dynamic fluctuation graph and a large model analysis report in a specified area of a display screen; The large model heart rate monitoring system is used for monitoring a heart rate signal in real time, generating a decision tree and an analysis report corresponding to the heart rate, and displaying a heart rate dynamic fluctuation graph and a large model analysis report in a specified area of a display screen; The large model P wave monitoring system is used for monitoring a P wave signal in real time, generating a decision tree and an analysis report corresponding to the P wave, and displaying a P wave dynamic fluctuation graph and a large model analysis report in a specified area of a display screen; The large model QRS complex monitoring system is used for monitoring a QRS complex signal in real time, generating a decision tree and an analysis report corresponding to the QRS complex, and displaying a QRS complex dynamic fluctuation graph and a large model analysis report in a specified area of a display screen. The large model T wave monitoring system is used for monitoring a T wave signal in real time, generating a decision tree and an analysis report corresponding to the T wave, and displaying a T wave dynamic fluctuation graph and a large model analysis report in a specified area of a display screen. The large model U wave monitoring system is used for monitoring a U wave signal in real time, generating a decision tree and an analysis report corresponding to the U wave, and displaying a U wave dynamic fluctuation graph and a large model analysis report in a specified area of a display screen. The large model PR interval monitoring system is used for monitoring a PR interval signal in real time, generating a decision tree and an analysis report corresponding to the PR interval, and displaying a PR interval dynamic fluctuation graph and a large model analysis report in a specified area of a display screen. The large model QT interval monitoring system is used for monitoring a QT interval signal in real time, generating a decision tree and an analysis report corresponding to the QT interval, and displaying a QT interval dynamic fluctuation graph and a large model analysis report in a specified area of a display screen. The large model RR interval monitoring system is used for monitoring an RR interval signal in real time, generating a decision tree and an analysis report corresponding to the RR interval, and displaying an RR interval dynamic fluctuation graph and a large model analysis report in a specified area of a display screen.

6. The wearable health device large model electrocardio monitoring method based on BSS-CNN-RF-DT-LLM fusion technology, characterized in that, The method comprises the following steps: S1: a wearer wears the hardware device of the system according to any one of claims 1-5, and turns on the device to make it in a normal working state; S2: the intelligent sound and light sensor detects the real-time health status of electrocardiogram, heart rate, P wave, QRS complex, T wave, U wave, PR interval, QT interval and RR interval of the wearer; S3: the human body sensing measurer senses and measures the physiological indexes corresponding to the above-mentioned electrocardio signals of the wearer; S4: the digital receiver receives the real-time measurement values of the electrocardio signals, and transmits them to the digital storage for storage in the background of the device; S5: the digital storage transmits the saved measurement values of the electrocardio signals to the intelligent data integration processing chip through the digital trigger transmitter, and performs data preprocessing and data quality filtering in the chip; S6: the system starts the blind source signal separation algorithm chip, and the blind source signal separation algorithm system enters a working state to automatically execute the blind source signal separation algorithm to separate the mixed electrocardio signals containing interference to obtain statistically independent and non-Gaussian distributed independent signal sources; S7: the system starts the convolutional neural network algorithm chip, and the convolutional neural network algorithm system enters a working state to automatically execute the convolutional neural network algorithm to identify the features of the independent signal sources separated in step S6, and distinguish the features of the electrocardio signals and the interference signals. S8: The system starts the random forest construction and traversal algorithm chip, and the random forest construction and traversal algorithm system enters the working state. The independent signal source separated in step S6 is taken as a root node. For the root node with the electrocardiogram signal characteristics identified in step S7, each root node is connected with the shape, time domain, frequency domain, and amplitude sub-nodes as the second layer nodes, respectively. Each second layer node is further connected with the "normal", "abnormal", and "dangerous" classification nodes, respectively, to form an "electrocardiogram tree", a "heart rate tree", a "P wave tree", a "QRS wave group tree", a "T wave tree", a "U wave tree", a "PR interval tree", a "QT interval tree", and a "RR interval tree". For the root node with the artifact signal characteristics identified, a "artifact tree" containing only the root node is constructed. All "electrocardiogram signal trees" and "artifact trees" are combined to form a random forest. S9: The system starts the decision tree classification algorithm chip, and the decision tree classification algorithm system enters the working state. In the random forest established in step S8, "electrocardiogram signal" is added as a root node. The root node is connected with all "electrocardiogram signal trees" and "artifact trees" in the random forest to construct an "electrocardiogram signal-decision tree". The "electrocardiogram signal-decision tree" is composed of the "electrocardiogram signal" root node and the "random forest" constructed in step S8. Therefore, when traversing the "electrocardiogram signal-decision tree", the "electrocardiogram signal" root node is not considered, and the traversal algorithm of the forest is directly used to traverse the "random forest" under the "electrocardiogram signal" root node constructed in step S8, that is, the tree traversal is performed on each sub-tree under the "electrocardiogram signal" root node. In the traversal process, when a leaf node is accessed, the actual monitoring value of the leaf node corresponding to the root node of the sub-tree is compared and matched with the ranges of the three classifications of "normal", "abnormal", and "dangerous". The matching result is saved to the decision tree classification algorithm chip in the form of a ternary array (electrocardiogram signal, monitoring value, and judgment range). S10: The system starts the large model electrocardiogram diagnosis report storage chip, and the large model search and generation algorithm system enters the working state. The normal range, abnormal range, dangerous range, and typical cases of the corresponding electrocardiogram signal in the knowledge base are extracted through the large model search on the whole network. Deep mining, understanding, comparison, information capture, semantic splicing, and chart interpretation are performed to form a large model semantic generated text description. The "large model electrocardiogram signal analysis report" and "large model heart health status report" are generated and saved in the large model electrocardiogram diagnosis report storage chip. S11: The system starts the digital monitoring chip, and the large model wearable electrocardiogram monitoring algorithm system enters the working state. The real-time monitoring electrocardiogram of the wearer can be displayed left and right in the upper one-third area of the device display screen. The "large model electrocardiogram signal analysis report" and "large model heart health status diagnosis report" can be displayed up and down in the lower two-thirds area of the display screen.

7. The wearable health device large model electrocardio monitoring method based on the BSS-CNN-RF-DT-LLM fusion technology according to claim 7, characterized in that, In step S6, when the blind source signal separation algorithm is executed, the corresponding algorithm is selected according to the characteristics of the independent signal source. The corresponding algorithms include FastICA algorithm, SOBI algorithm, and Infomax algorithm.

8. The wearable health device large model electrocardio monitoring method based on the BSS-CNN-RF-DT-LLM fusion technology according to claim 7, characterized in that, The process of constructing the random forest in step S8 includes: The morphology sub-node of the "ECG signal tree" records the waveform characteristics of the corresponding ECG signal, the time domain sub-node records the time interval parameters of the corresponding ECG signal, the frequency domain sub-node records the frequency distribution characteristics of the corresponding ECG signal, and the amplitude sub-node records the voltage amplitude range of the corresponding ECG signal; The threshold range of the "normal", "abnormal" and "dangerous" classification nodes is set based on the clinical standards of the corresponding ECG indicators in the heart health knowledge base, wherein the "normal" range is the medically recognized healthy interval, the "abnormal" range is the interval deviating from the normal but not reaching the critical interval, and the "dangerous" range is the interval prompting the possible occurrence of acute cardiac events.

9. The wearable health device large model electrocardio monitoring method based on the BSS-CNN-RF-DT-LLM fusion technology according to claim 7, characterized in that, The interaction logic of the display screen in step S11 includes: When the wearer clicks on the icons corresponding to the 9 ECG signals on the display screen, the display screen focuses on displaying the full-screen dynamic fluctuation graph of the signal, wherein the waveform segment exceeding the "normal" range is marked in red, and the large model analysis report of the signal is synchronously displayed; The display screen supports switching the monitoring data of different time points through touch sliding, and the "large model ECG signal analysis report" includes waveform characteristic description, indicator value comparison and abnormal point labeling; the "large model heart health status diagnosis report" includes risk level assessment, recommended intervention measures and typical case reference.

Images