倪娟 王剑卓
（广东工业大学，广东 广州 510006）
摘要：针对心电图不同样本间的高变异性，以及深度学习模型泛化能力不足的问题，提出一种基于Mixup的心电图多标签异常心律检测方法。首先，通过Mixup方法将心电图与白噪声混合；然后，利用混合样本训练深度学习模型；最后，在CPSC 2018数据集上进行实验，该方法的F1分数相较于Inception-ResNet-v2、MLC-CNN、STA-CRNN分别提升了0.014、0.031、0.023。
关键词：Mixup方法；心电图；多标签；异常心律检测；深度学习模型
中图分类号：TP391 文献标志码：A 文章编号：1674-2605(2024)03-0008-05
DOI：10.3969/j.issn.1674-2605.2024.03.008
Multi Label Abnormal Heart Rate Detection Method for ECG Based on Mixup
NI Juan WANG Jianzhuo
(Guangdong University of Technology, Guangzhou 510006, China)
Abstract: Multi label abnormal heart rate detection method based on Mixup is proposed to address the high variability of electrocardiograms between different samples and the insufficient generalization ability of deep learning models. Firstly, the electrocardiogram is mixed with white noise using the Mixup method; Then, use mixed samples to train deep learning models; Finally, the experiment was conducted on the CPSC 2018 dataset, compared to Inception-ResNet-v2, MLC-CNN, STA-CRNN, the F1 scores of this method have increased by 0.014, 0.031, and 0.023, respectively.
Keywords: Mixup method; electrocardiogram; multi label; abnormal heart rate detection; deep learning models
