Electrocardiogram signal recognition method and device based on artificial neural network

Abstract

The present application relates to the technical field of physiological signal analysis and recognition, in particular to an electrocardiogram signal recognition method and device based on artificial neural network. The method acquires and pre-processes electrocardiogram signal data; an artificial neural network model is trained using the pre-processed electrocardiogram signal data, and the connection weight coefficient is adjusted using the back propagation algorithm; the nodes of each layer of the artificial neural network model are connected through multiple pairs of cascaded synapse transistors; the connection weight coefficient is determined by the conductance value of each synapse transistor and the amplification multiple of each pair of synapse transistors; the trained artificial neural network model is used for testing to obtain the recognition result of the electrocardiogram signal. Compared with the prior art, the present application has the advantages of high online recognition accuracy and strong stability of electrocardiogram signal.

Description

Technical Field

[0001] This invention relates to the field of physiological signal analysis and recognition technology, and in particular to a method and apparatus for electrocardiogram signal recognition based on artificial neural networks. Background Technology

[0002] According to the "China Cardiovascular Health and Disease Report 2022" released by the National Center for Cardiovascular Diseases, cardiovascular disease (CVD) has become the leading cause of death among urban and rural residents, and its prevalence continues to rise. In the prediction and diagnosis of cardiovascular diseases, electrocardiography (ECG), as a primary harmless method for detecting cardiac electrical activity, is crucial for signal analysis and identification. On the one hand, the widespread use of ECG monitoring devices has resulted in massive amounts of ECG data; on the other hand, commonly used static ECGs may last from tens of seconds to a minute, making it difficult to detect some occasional, short-term cardiac abnormalities. Holter monitoring (also known as long-term ECG), on the other hand, typically records a patient's heart rate changes over 24 hours, combined with records of the patient's behavioral activities, to analyze the patient's overall health. Compared to static ECGs, Holter monitoring undoubtedly has advantages in capturing occasional cardiac events, but it requires processing a much larger volume of ECG data, making online processing difficult. Therefore, developing an automated and intelligent arrhythmia identification method for auxiliary diagnosis and even portable real-time monitoring is essential. The rapid development of machine learning technology in recent years has undoubtedly brought new opportunities for research on the automatic identification and classification of ECG signals, resulting in a number of studies using machine learning methods for ECG signal identification and classification. For example, Weimann K et al. disclosed a method in the paper "Transfer learning for ECG classification" that improves the performance of convolutional neural networks by performing transfer learning on static electrocardiograms, achieving a performance improvement of 6.57% under the constraint of a small dataset. Zhang P et al. disclosed a semi-supervised learning method in the paper "Semi-supervised learning for automatic atrial fibrillation detection in 24-hour Holter monitoring" for monitoring paroxysmal atrial fibrillation in 24-hour Holter monitoring. Currently, most methods for ECG signal monitoring and identification using machine learning operate on a pre-collection and post-processing offline mode, which has a certain lag. Most deep learning-based monitoring and identification methods are also based on offline analysis and processing, while directly applying digital signal processing technology for online processing has problems such as high power consumption, large size, and difficulty in wearing.

[0003] Based on the advantages of neural networks in machine learning data processing, neuromorphic chips inspired by the human brain have become one of the current development directions. One of their biggest features is that they can provide fast computing power with low power consumption while integrating storage and computation by referencing the neuronal structure of the human brain. For the connections between neurons in the brain, synapses are one of the basic structures. Correspondingly, for artificial neural networks (ANNs), artificial synapses have become a research hotspot in neuromorphic computing networks. For example, Wang R et al., in their paper "Artificial synapses based on lead-free perovskite floating-gate organic field-effect transistors for supervised and unsupervised learning," used artificial synapses based on lead-free perovskite floating-gate organic field-effect transistors to build supervised and unsupervised networks, achieving recognition accuracy of 91% and 81% for handwritten font sets, respectively. In the field of ECG signal recognition, Oh S, Lee J, and others disclosed a hardware neural network ECG signal recognition method based on photoactive synapses of MoS2 / h-BN heterojunctions in their paper "Photoelectroactive artificial synapse and its application to biosignal pattern recognition." However, the accuracy of this method is affected by factors such as the small test dataset, the device state order, and the number of device adjustment pulses, resulting in poor network testing accuracy and stability. Therefore, how to further improve the accuracy and stability of ECG signal recognition has become a problem that needs to be solved in this field. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art by providing a method and apparatus for electrocardiogram (ECG) signal recognition based on artificial neural networks, which can improve the accuracy and stability of ECG signal recognition.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] According to a first aspect of the present invention, a method for recognizing electrocardiogram (ECG) signals based on an artificial neural network is provided, the method comprising the following steps:

[0007] S1, acquire and preprocess electrocardiogram signal data;

[0008] S2, an artificial neural network model is trained using preprocessed electrocardiogram signal data, and the connection weight coefficients are adjusted using the backpropagation algorithm. The nodes of each layer of the artificial neural network model are connected to each other through multiple pairs of cascaded synaptic transistors. The connection weight coefficients are determined by the conductance of each synaptic transistor and the amplification factor of each pair of synaptic transistors.

[0009] S3 uses a trained artificial neural network model for testing to obtain the recognition results of electrocardiogram signals.

[0010] As a preferred technical solution, the process of adjusting the connection weight coefficients using the backpropagation algorithm includes obtaining a loss function based on the dimension-reduced electrocardiogram signal data, calculating the error gradient of backpropagation, and updating the conductance values ​​of each synaptic transistor.

[0011] As a preferred technical solution, the process of updating the conductance value of each synaptic transistor includes the following steps:

[0012] S201, calculate the theoretical change in conductance of the current first-level synaptic transistor based on the loss function;

[0013] S202, based on the current theoretical change in conductance and the current conductance value of the synaptic transistor, increase or decrease the conductance value of the synaptic transistor that meets the conditions in the current stage;

[0014] S203, based on the residual change value and the current conductance value of the synaptic transistor, increase or decrease the conductance value of the synaptic transistor that meets the conditions in the current stage, wherein the residual change value is the difference between the current theoretical change value and the actual change value of the conductance;

[0015] S204: When the residual change value reaches the required accuracy, the residual change value is passed to the next stage and returned to S201. When the last stage synaptic transistor is reached, the update ends.

[0016] As a preferred technical solution, the conductance of the synaptic transistor is increased by applying an external light pulse, and the conductance of the synaptic transistor is decreased by applying an external electrical pulse.

[0017] As a preferred technical solution, the method for fabricating the synaptic transistor includes the following steps:

[0018] S211, spin-coating a layer of negative photoresist onto a flexible substrate;

[0019] S212, a gate electrode is prepared using vapor deposition technology, an insulating layer is prepared on the gate electrode, and source and drain electrode patterns are photolithographically formed on the insulating layer;

[0020] S213 utilizes vapor deposition technology to deposit metal into the source and drain regions, followed by cleaning away the negative photoresist.

[0021] S214, spin-coating a layer of positive photoresist to photolithographically create a pattern template of the active region;

[0022] S215, spin-coating of the active layer, and stripping of the photoresist template using an organic polymer orthogonal solvent to obtain a patterned organic polymer.

[0023] As a preferred technical solution, the flexible substrate is one of polyimide, polyethylene naphthalate, and polyethylene terephthalate.

[0024] As a preferred technical solution, the source, drain, and gate electrodes are one of conductive metals, conductive alloys, and conductive oxides.

[0025] As a preferred technical solution, the material of the insulating layer is one of silicon dioxide, aluminum oxide, polymethyl methacrylate, polystyrene, polyvinyl alcohol, polyvinylpyrrolidone, and SU8 series photoresist.

[0026] As a preferred technical solution, the active layer comprises an organic polymer semiconductor material and a quantum dot material, which together form a bulk heterojunction structure.

[0027] According to a second aspect of the present invention, an electrocardiogram (ECG) signal recognition device based on an artificial neural network is provided, comprising a memory, a processor, and a program stored in the memory, wherein the processor executes the program to implement the method described herein, and the device comprises an ECG signal acquisition and preprocessing module, an artificial neural network model training module, and an ECG signal recognition module.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] 1. This invention uses an artificial neural network connected by multiple pairs of cascaded synaptic transistors to recognize electrocardiogram (ECG) signals. The weight coefficients of the artificial neural network are determined by the conductance of each synaptic transistor and the amplification factor of each pair of synaptic transistors, thereby improving the accuracy of online ECG signal recognition.

[0030] 2. This invention uses a flexible substrate to fabricate synaptic transistors, which allows the synaptic transistors to be bent, reducing power consumption and facilitating their fabrication as wearable devices, thus improving the portability of electrocardiogram signal recognition devices.

[0031] 3. The synaptic transistor conductance update method used in this invention effectively ensures that the conductance of the synaptic transistor only undergoes slight changes regardless of whether it is bent or not, thereby ensuring the stability of the artificial neural network performance. Attached Figure Description

[0032] Figure 1 This is a schematic flowchart of the method of the present invention;

[0033] Figure 2 This is a schematic diagram of an artificial neural network in an embodiment of the present invention;

[0034] Figure 3 This is a schematic diagram of the fitting parameters for the three-level synaptic structure in an embodiment of the present invention;

[0035] Figure 4 This is a graph showing the variation of the conductance parameters of the flexible synaptic transistor in an embodiment of the present invention;

[0036] Figure 5 This is a graph showing the changes in ECG signal recognition rate and loss function during the training process of an embodiment of the present invention;

[0037] Figure 6 This is a schematic diagram of the confusion matrix in an embodiment of the present invention;

[0038] Figure 7 This is a schematic diagram of the structure of the device of the present invention. Detailed Implementation

[0039] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0040] like Figure 1 As shown, this embodiment provides a method for electrocardiogram (ECG) signal recognition based on artificial neural networks. The method includes the following steps:

[0041] Step S1: Acquire and preprocess ECG signal data. During the preprocessing of ECG signal data, features of the ECG signal data are extracted, continuous ECG signals are segmented into standard-length segments, and signal normalization is performed to represent complex signals in a form understandable by machine learning models. In this embodiment, the ECG signal data is obtained from the MIT-BIH arrhythmia database. As shown in Table 1, the 14 heart rate types in this database are divided into five categories: NSVFQ. Specifically, the beginning and end of all signals in the MIT-BIH arrhythmia database are truncated, and heart rate labels and their positions that match the categories shown in Table 1 are selected. For each selected label, a heart rate signal with a length of 250 points before and after its label position is extracted, and its label is updated to an NSVFQ type label according to the rules in Table 1. Signal normalization is then performed on the heart rate data to complete the preprocessing. The preprocessed ECG signal data is divided into a training set and a test set for model training and evaluation, respectively. This embodiment processed the MIT-BIH arrhythmia database to obtain an ECG dataset of 45,000 samples. Because the F and S class heart rate data were relatively scarce, leading to dataset imbalance, the F and S class datasets were oversampled by 2 to 5 times to balance the number of heart rate data types. Finally, 5,000 samples were randomly selected from the total 45,000 ECG samples as the test set, and the rest were used as the training set.

[0042] Table 1 ECG Signal Classification Table

[0043]

[0044] Step S2 involves inputting the preprocessed ECG signal training set data into the artificial neural network model and training it, using the backpropagation algorithm to adjust the connection weight coefficients. Each training process essentially utilizes synaptic transistors to achieve dimensionality reduction and classification of the preprocessed ECG signals.

[0045] like Figure 2 As shown, the artificial neural network model in this embodiment has three layers, each with several nodes. Nodes are connected via multiple pairs of cascaded synaptic transistors. Each connection weight coefficient is determined by the conductance of each synaptic transistor and the amplification factor of each pair of synaptic transistors. Specifically, each connection weight coefficient between nodes in each layer of the artificial neural network model is fitted using three pairs of cascaded organic photoelectric synaptic transistors with different amplification factors. The three-level synaptic structure is as follows: Figure 3 As shown, the amplification factor differs by a factor of 50 between each stage. Ideally, a three-stage synaptic structure can achieve approximately 0.01 / 50. 2 The fitting accuracy. The calculation rule for the connection weight coefficients of the fit is:

[0046]

[0047] in, and To magnify 50 2 The conductivity of the two flexible organic photosynaptic transistors in the first stage is twice that of the first stage. and To magnify 50 1 The conductance of the two flexible organic photosynaptic transistors in the second stage is times that of the second stage. and To magnify 50 0 The conductance of the two flexible organic photosynaptic transistors at the third stage is times that of G; max With G min γ represents the maximum and minimum conductance values ​​of the flexible organic photoelectric synaptic transistor; γ is a parameter that limits the range of values ​​for W, which ranges from [-γ, γ].

[0048] Furthermore, the process of adjusting the connection weight coefficients using the backpropagation algorithm includes: obtaining a loss function based on the extreme values ​​of the dimensionality-reduced ECG signal data and corresponding label values; backpropagating the loss function to each layer of the artificial neural network; calculating the error gradient of each network parameter; multiplying the error gradient by the learning rate to obtain the theoretical change in conductance of each synaptic transistor; and then updating the conductance value of each synaptic transistor. Starting from the first-level synaptic transistor, the specific steps for updating the conductance value are as follows:

[0049] Step S201: Calculate the theoretical change in conductance of the current first-level synaptic transistor based on the loss function;

[0050] Step S202: Based on the current theoretical change in conductance and the current conductance value of the synaptic transistor, increase or decrease the conductance value of the synaptic transistor that meets the conditions in the current stage.

[0051] Step S203: Based on the residual change value and the current conductance value of the synaptic transistor, increase or decrease the conductance value of the synaptic transistor that meets the conditions in the current stage. The residual change value is the difference between the theoretical change value and the actual change value of the current conductance.

[0052] In step S204, when the residual change value reaches the required accuracy, the residual change value is passed to the next level and the process returns to S201. When the last synaptic transistor is reached, the update ends.

[0053] In steps S202 and S203, the conductance of the synaptic transistor can be increased by applying an external light pulse, and the conductance of the synaptic transistor can be decreased by applying an external electrical pulse.

[0054] The method for fabricating the synaptic transistor in this embodiment includes the following steps:

[0055] Step S211: Spin-coat a layer of negative photoresist onto the flexible substrate to smooth the surface of the flexible substrate;

[0056] Step S212: Prepare a full-surface gate electrode using vapor deposition technology, prepare an insulating layer on the gate electrode, and photolithographically print source and drain electrode patterns on the insulating layer;

[0057] Step S213: After depositing metal into the source and drain regions using vapor deposition technology, the negative photoresist is cleaned off.

[0058] Step S214: Spin-coat a layer of positive photoresist and photolithographically print the active region pattern template;

[0059] In step S215, an active layer is spin-coated, and the photoresist template is stripped using an organic polymer orthogonal solvent to obtain a patterned organic polymer.

[0060] In steps S211 to S215, the flexible substrate used in the fabrication method is polyimide (PI) or polyethylene terephthalate (PET); the negative photoresist used to smooth the surface of the flexible substrate is SU8-2005; the source, drain, and gate electrodes are 10nm chromium and 35nm gold. The insulating layer is made of negative photoresist SU8-2000.5; the active layer is a composite of organic polymer DPPDTT and CsPbBr3 quantum dots, forming a bulk heterojunction structure. Because a flexible substrate is used to fabricate the synaptic transistor, the synaptic transistor can be bent. After bending, the conductivity value changes only slightly compared to the unbent array device. Combined with the aforementioned method for updating the synaptic transistor conductivity value, the accuracy of ECG signal recognition and the stability of artificial neural network performance can be fully guaranteed. Figure 4 As shown, the conductivity parameters of the actual test device change under external light / electric pulse stimulation. The first half is the conductivity value increase curve under light pulse stimulation in the case of no bending / bending, and the second half is the conductivity value decrease curve under electrical pulse stimulation in the case of no bending / bending.

[0061] After steps S1 to S2, a network parameter update is completed. Through multiple iterations until convergence, the artificial neural network training is completed, which can improve the accuracy of ECG signal recognition.

[0062] Step S3: Using 5000 test set data to simulate the patient's heart examination, the trained artificial neural network model is used for testing to obtain the recognition result of ECG signal.

[0063] Under three scenarios—pure software training, training with unbent device parameters, and training with bent parameters—and with random initial values, 40,000 training data points were used for 80 iterations in each scenario. The final results for the recognition rate and loss function changes on 5,000 test data points are shown below. Figure 5 As shown. Among them, Figure 5 (a) shows the curves of the test set recognition rate as a function of the number of iterations under three conditions, yielding recognition accuracy rates of 0.9442, 0.9196, and 0.9172, respectively. Figure 5 (b) shows the curves of the loss function changing with the number of iterations in three cases. The final confusion matrix is ​​as follows. Figure 6 As shown, where Figure 6 (a) is the result obtained from pure software training. Figure 6 (b) Results obtained from training under conditions where the device is not bent. Figure 6 (c) Results obtained from training under device bending conditions. In summary, the ECG signal recognition method provided in this embodiment can be analyzed and processed online, and can be worn in a flexible manner. Once the network is trained, it can be integrated with ECG signal acquisition equipment to process dynamic ECG signals online.

[0064] Furthermore, this embodiment also provides an electrocardiogram signal recognition device based on an artificial neural network, including a memory, a processor, and a program stored in the memory, wherein the processor executes the program to implement the aforementioned method. Figure 7 As shown, the electrocardiogram signal recognition device in this embodiment includes the following modules:

[0065] The electrocardiogram signal acquisition and preprocessing module 101 is used to acquire and preprocess ECG signal data;

[0066] The artificial neural network model training module 102 includes an error backpropagation unit 112 and a synaptic conductance adjustment unit 122. This module trains the artificial neural network model using preprocessed ECG signal data. The nodes in each layer of the artificial neural network model are connected to each other through multiple pairs of cascaded synaptic transistors. The error backpropagation unit 112 is used to adjust the connection weight coefficients through a backpropagation algorithm. The connection weight coefficients are jointly determined by the conductance value of each synaptic transistor and the amplification factor of each pair of synaptic transistors. The synaptic conductance adjustment unit 122 is used to update the conductance value of each synaptic transistor.

[0067] The electrocardiogram signal recognition module 103 is used to perform tests using a trained artificial neural network model to obtain the recognition results of the ECG signal.

[0068] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the aforementioned modules can be referred to the corresponding process in the aforementioned method embodiments, and will not be repeated here.

[0069] The aforementioned method and apparatus employ an artificial neural network based on multi-level flexible organic photoelectric synaptic transistors to achieve high-speed and accurate recognition of ECG signals. The hardware implementation method of the flexible organic photoelectric synaptic transistor for the artificial neural network and the efficient adjustment algorithm of the flexible organic photoelectric synaptic transistor parameters during the gradient descent process are presented. The results show that the artificial neural network based on multi-level flexible organic photoelectric synaptic transistors can effectively identify different types of ECG signals under both bending and non-bending conditions.

[0070] The ECG signal recognition device provided in this embodiment can perform various appropriate actions and processes according to computer program instructions stored in read-only memory (ROM) or loaded from storage units into random access memory (RAM). The RAM can also store various programs and data required for device operation. The CPU, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus. Multiple components in the device are connected to the I / O interfaces, including: input units, such as keyboards, mice, etc.; output units, such as various types of displays, speakers, etc.; storage units, such as disks, optical disks, etc.; and communication units, such as network cards, modems, wireless transceivers, etc. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks. The processing unit executes the various methods and processes described above, such as methods S1 to S3. For example, in some embodiments, methods S1 to S3 can be implemented as computer software programs tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program can be loaded and / or installed on the device via ROM and / or the communication unit. When a computer program is loaded into RAM and executed by the CPU, one or more steps of methods S1-S3 described above can be performed. Alternatively, in other embodiments, the CPU can be configured to execute methods S1-S3 by any other suitable means (e.g., by means of firmware). The functions described above can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0071] The program code used to implement the aforementioned methods can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on a machine, partially on a machine, partially on a machine and partially on a remote machine as a standalone software package, or entirely on a remote machine or server. In this embodiment, the machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0072] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. An electrocardiogram signal recognition method based on artificial neural networks, characterized by, The method comprises the following steps: S1, acquiring and preprocessing electrocardiogram signal data; S2, training an artificial neural network model using the preprocessed electrocardiogram signal data, and adjusting the connection weight coefficient using a back propagation algorithm, wherein the nodes of each layer of the artificial neural network model are connected through multiple pairs of cascaded synaptic transistors, and the connection weight coefficient is determined by the conductance value of each synaptic transistor and the amplification factor of each pair of synaptic transistors; The process of adjusting the connection weight coefficient using the back propagation algorithm comprises obtaining a loss function according to the reduced electrocardiogram signal data, calculating the error gradient of back propagation, and updating the conductance value of each synaptic transistor; The process of updating the conductance value of each synaptic transistor comprises the following steps: S201, calculating the theoretical conductance change value of the current first-level synaptic transistor according to the loss function; S202, increasing or decreasing the conductance value of the synaptic transistor that meets the condition in the current first-level synaptic transistor according to the current theoretical conductance change value and the current conductance value of the synaptic transistor; S203, increasing or decreasing the conductance value of the synaptic transistor that meets the condition in the current first-level synaptic transistor according to the residual change value and the current conductance value of the synaptic transistor, wherein the residual change value is the difference between the current theoretical conductance change value and the actual change value; S204, when the residual change value reaches the accuracy, passing the residual change value to the next level, and returning to S201, and when reaching the last level synaptic transistor, ending the update; The conductance value of the synaptic transistor is increased by an external light pulse stimulus, and the conductance value of the synaptic transistor is decreased by an external electric pulse stimulus; The preparation method of the synaptic transistor comprises the following steps: S211, spin coating a layer of negative photoresist on a flexible substrate; S212, preparing a gate electrode using evaporation technology, preparing an insulating layer on the gate electrode, and photoetching a source and drain electrode pattern on the insulating layer; S213, depositing metal in the source and drain regions using evaporation technology, and then cleaning the negative photoresist; S214, spin coating a layer of positive photoresist, and photoetching an active region pattern template; S215, spin coating an active layer, and using an organic polymer orthogonal solvent to strip the photoresist template to obtain a patterned organic polymer; S3, testing using the trained artificial neural network model to obtain the recognition result of the electrocardiogram signal.

2. The electrocardiogram signal recognition method based on artificial neural network according to claim 1, characterized in that, The flexible substrate is one of polyimide, polyethylene naphthalate, and polyethylene terephthalate. 3.The artificial neural network-based electrocardiogram signal recognition method of claim 1, wherein, The source, drain, and gate electrodes are one of a conductive metal, a conductive alloy, and a conductive oxide.

4. The artificial neural network-based electrocardiogram signal recognition method of claim 1, wherein, The material of the insulating layer is one of silicon dioxide, aluminum oxide, polymethyl methacrylate, polystyrene, polyvinyl alcohol, polyvinyl pyrrolidone, and photoresist SU8 series.

5. The artificial neural network based electrocardiogram signal recognition method of claim 1, wherein, The active layer comprises an organic polymer semiconductor material and a quantum dot material, and the two form a bulk heterojunction structure.

6. An electrocardiogram signal recognition apparatus based on an artificial neural network, comprising a memory, a processor, and a program stored in the memory, characterized by, The processor implements the method of any one of claims 1-5 when executing the program, and the device comprises an electrocardiogram signal acquisition and preprocessing module, an artificial neural network model training module, and an electrocardiogram signal recognition module.

Images