基于CNN与Transformer相融合的心跳分类算法  

Heartbeat Classification Algorithm Based on Combination of CNN and Transformer

作　　者：刘子杰 杨晨[1,2,3] LIU Zijie;YANG Chen(College of Big Data and Information Engineering,Guizhou University,Guiyang Guizhou 550025,China;Ministry of Education Engineering Centre for Reliability of Functional Semiconductor Devices,Guiyang Guizhou 550025,China;State Key Laboratory of Public Big Data,Guizhou University,Guiyang Guizhou 550025,China)

机构地区：[1]贵州大学大数据与信息工程学院,贵州贵阳550025 [2]功能半导体器件可靠性教育部工程中心,贵州贵阳550025 [3]贵州大学公共大数据国家重点实验室,贵州贵阳550025

出　　处：《通信技术》2024年第6期556-562,共7页Communications Technology

基　　金：国家自然科学基金(62065003);贵州省科技项目(ZK[2022]Key-020)。

摘　　要：心电图(Electrocardiogram,ECG)对于心血管疾病的诊断有着重要的作用,精确的心跳分类有助于后续的疾病治疗。但原始的心电图信号存在的大量噪声会影响计算机的判断,而去噪可能会导致信号特征丢失。为此,提出了利用格拉姆角场(Gramian Angular Field,GAF)将一维信号转化为图像,以保证信息完整性,然后利用计算机视觉处理技术分析心跳信号。此外,针对视觉Transformer(Vision Transformer,ViT)在局部特征捕获能力受限的问题,提出了在ViT模型内部和外部引入带残差连接的卷积神经网络(Convolutional Neural Network,CNN),使整体网络能够同时关注图像的局部特征和全局特征。同时,通过将ViT中单一的自注意块改进为双支路通道空间注意力结构,提升了模型的整体分类性能。实验结果表明,所提出的方法有效提高了网络的性能。ECG(Electrocardiogram)plays an important role in the diagnosis of cardiovascular diseases,and accurate classification of heartbeats helps in the subsequent treatment of the disease.However,the large amount of noise present in the raw ECG signals affects the computer’s judgment,and denoising may lead to loss of signal characteristics.For this reason,this paper proposes to convert 1D signals into images using GAF(Gramian Angular Field)to ensure information integrity,and then analyze the heartbeat signals using computer vision processing techniques.In addition,to address the problem that ViT(Vision Transformer)is limited in its ability to capture local features,this paper proposes to introduce CNN(Convolutional Neural Network)with residual connections inside and outside of the ViT model,so that the overall network can simultaneously focus on the local and global features of the image.Meanwhile,the overall classification performance of the model is enhanced by improving the single self-attention block in ViT to a dual-branch channel spatial attention structure.Experimental results indicate that the proposed method effectively improves the performance of the network.

关 键 词：心跳分类 TRANSFORMER 卷积神经网络 格拉姆角场 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]