基于生成对抗网络的心电数据增强算法研究
首发时间:2020-06-28
摘要:目前已经公开的标准心电图数据库中,大多都存在严重的类别失衡现象,这就给心电图分类识别准确率的提升带来了巨大阻碍。为了解决这一问题,本文通过构建基于BILSTM和CNN的生成对抗网络模型合成高质量复杂类型的心电数据,对数据集中少数类样本进行扩充。仿真结果表明,该生成模型产生的心电信号在波形形态和频域特征上都接近于真实信号。为了进一步验证生成数据的有效性,本文设计了CNN分类器,用于检测心电数据集扩充前后分类指标值accuracy和macro-F1的变化。实验结果表明,本文提出的数据增强算法可以有效提升心电分类准确率。
关键词: 人工智能 生成对抗网络 BILSTM 卷积神经网络 差分阈值法
For information in English, please click here
Research on ECG synthesis Based on Generative Adversarial Network
Abstract:In Most of the standard ECG databases that have been published so far have serious category imbalances, which has brought great obstacles to the improvement of the accuracy of ECG classification and recognition. In order to solve this problem, this paper expands the minority samples of the data set by building a generative adversarial network model based on BILSTM and CNN, synthesizing high-quality and complex types of ECG data. The simulation results show that the ECG signal generated by the generated model is close to the real signal in waveform shape and frequency domain characteristics. In order to further verify the validity of the generated data, the CNN classifier is designed to detect the changes of the classification index value accuracy and macro-F1 before and after the expansion of the ECG data set. The experimental results show that the data enhancement algorithm proposed in this paper can effectively improve the ECG Classification accuracy.
Keywords: Artificial intelligence Generative adversarial network BILSTM Convolutional Neural Network Difference threshold method
基金:
引用
No.****
同行评议
勘误表
基于生成对抗网络的心电数据增强算法研究
评论
全部评论0/1000