基于深度迁移学习的心电信号分类方法  

ECG SIGNAL CLASSIFICATION METHOD BASED ON DEEP TRANSFER LEARNING

作　　者：孙梦莉 王演[1] 刘晓凯 林卉 张世强 Sun Mengli;Wang Yan;Liu Xiaokai;Lin Hui;Zhang Shiqiang(College of Information Science and Technology,Dalian Maritime University,Dalian 116026,Liaoning,China;Hualu Zhida Technology Co.,Ltd.,Dalian 116023,Liaoning,China)

机构地区：[1]大连海事大学信息科学技术学院,辽宁大连116026 [2]华录智达科技有限公司,辽宁大连116023

出　　处：《计算机应用与软件》2020年第5期133-139,共7页Computer Applications and Software

基　　金：国家自然科学基金项目(61802044)。

摘　　要：针对传统的心电信号分类方法中存在模型复杂度较高且训练时间长等问题,提出深度迁移分类方法。预先使用ImageNet数据集在VGG-16网络中训练过的全网络连接层,对心电数据库(MIT-BIH)采用微调VGG-16预训练,实现基于同构数据下的特征迁移,利用分类模型训练,得到心电信号分类结果。分类训练模型采用算法复杂度评估,深度迁移学习模型相较于深度学习模型时间复杂度减少了O(102)。实验结果表明,在提升深度模型分类性能和加速心电信号训练过程上,该方法优于深度学习方法。Aiming at the problems of traditional ECG signal classification methods,such as high model complexity and long training time,this paper proposes a deep transfer classification method.It used ImageNet data set in advance to train the full network connection layer in VGG-16 network.The ECG database(MIT-BIH)was pre-trained with fine tuning VGG-16 to realize feature migration based on isomorphic data.Then,the classification model was used to train,and the result of ECG classification was obtained.The classification training model adopted algorithm complexity evaluation,and the time complexity of the deep transfer learning model was reduced by O(102),compared with the deep learning model.The experimental results show that this method is superior to the deep learning method in improving the classification performance of the depth model and accelerating the ECG signal training process.

关 键 词：心电信号分类 迁移学习 深度学习 算法复杂度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]