基于1-D CNN的QRS波群检测算法  

QRS complex detection algorithm based on 1-D CNN

作　　者：郭新杰 曾成 孙尚云 赵地 花中秋 GUO Xin-jie;ZENG Cheng;SUN Shang-yun;ZHAO Di;HUA Zhong-qiu(School of Electronic and Information Engineering,Hebei University of Technology,Tianjin 300401,China;Tianjin Branch of Computing Technology Institute,Chinese Academy of Sciences,Tianjin 300380,China;Institute of Computing Technology,Chinese Academy of Sciences,Beijing 100190,China)

机构地区：[1]河北工业大学电子信息工程学院,天津300401 [2]中国科学院计算技术研究所天津分所,天津300380 [3]中国科学院计算技术研究所,北京100190

出　　处：《计算机工程与设计》2020年第9期2469-2475,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(61501167);2018年度首都卫生发展科研专项基金项目(首发2018-2-4064);北京中医药科技发展基金项目(QN2018-07)。

摘　　要：为提高QRS波群检测算法的准确性、鲁棒性以及抗噪声能力,提出基于一维卷积神经网络(1-D CNN)的QRS波群检测算法。以心电信号的R波/非R峰波位置为中心截取固定长度采样点制作训练集和验证集,将训练集数据喂入1-D CNN模型,实现QRS/非QRS波群2分类,通过非极大值抑制方法实时输出QRS波群检测结果。1-D CNN模型在MIT-BIH心律失常数据库验证集的准确率为99.54%,在可穿戴心电检测设备上QRS波群检测准确率为98.51%。实验结果表明,该算法具有准确度高、鲁棒性好和抗干扰性能强的特点。To improve the accuracy,anti-noise ability and robustness for QRS complex detection,an algorithm of QRS complex detection based on one-dimensional convolutional neural network(1-D CNN)was proposed.The training and validation sets were formed by fixed-length samples truncated with R wave/non-R wave positions of ECG signal as the center.The 1-D CNN model was trained with training sets to identify QRS/non-QRS complex.Its outputs were processed using a non-maximum suppression method to get the final decisions.Experimental results on validation sets indicate that the proposed model achieves the detection accuracy of 99.54%.In another experiment on the data acquired from wearable ECG detection devices,the proposed method achieves a promising classification accuracy of 98.51%.Experimental results show that the proposed algorithm has the characteristics of high accuracy,good robustness and anti-noise ability.

关 键 词：心电图 QRS波群检测 一维卷积神经网络 非极大值抑制 噪声鲁棒性 

分 类 号：TP391.5[自动化与计算机技术—计算机应用技术]