期刊
MULTIMEDIA TOOLS AND APPLICATIONS
卷 81, 期 11, 页码 16047-16065出版社
SPRINGER
DOI: 10.1007/s11042-022-12614-8
关键词
Electrocardiogram signal; Nonnegative representation; Dictionary learning; Sparse representation
类别
资金
- National Natural Science Foundation of China [61863027]
- Key Research and Development Plan of Jiangxi Province [20202BBGL73057]
- Natural Science Foundation of Jiangxi Province [20171BAB201013]
- Project of Nanchang Key Laboratory of Medical and Technology Research [2018-NCZDSY-002]
This research proposes a novel ECG biometric method called Label Consistent Non-negative Representation (LCNR) for ECG classification. The method utilizes non-negative constrained least squares model and blockwise coordinate descent algorithm, and achieves superior performance in experiments.
Electrocardiogram (ECG) is a common and powerful tool for studying heart function and diagnosing several abnormal arrhythmias. This paper aims to propose a novel robust ECG biometric method, named the Label Consistent Non-negative Representation (LCNR), for ECG classification. We propose an objective function consists of the reconstruction error, classification error and discriminative sparse-code error with the non-negative regularization term on the coding coefficients. The coding vector was restricted to be non-negative using a non-negative constrained least squares model, and a blockwise coordinate descent algorithm was used to simultaneously learn a compact discriminative dictionary and a multiclass linear classifier. The experiments are carried out for the proposed methods using benchmark MIT-BIH data and evaluated under standard scheme and category-based scheme. The evaluation and experimental results show that our proposed LCNR algorithm achieves state-of-the-art performance, specifically surpassing the label consistent KSVD algorithm in terms of classification accuracy. By means of the dictionary learning algorithm, we can improve the efficiency for a large-size training database with a significantly faster execution time (more than 5 times) than NRC.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据