.png){kind=icon}
.png){kind=icon}
Return
Structure-constrained low-rank and sparse representation with label embedding graph regularization for classification of cardiac arrhythmias
.png){kind=icon}
PRE
AI
B
P
Y
X
F
J
DOI：10.1016/j.bspc.2026.111249.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The electrocardiogram (ECG) is an effective tool for cardiovascular disease diagnosis and arrhythmia detection. Representation learning of ECG has been an active research field for the automated detection of cardiac disease. The locality and label information of training samples play an important role in representation learning classification. However, previous dictionary learning algorithms did not simultaneously consider the locality and label information of training samples during the learning process, or did not consider the distribution differences of similar data, so their performance was limited. In this paper, we propose a novel structure-constrained low-rank and sparse representation with label embedding graph regularization (SCLSR-LEGR) model for automatic electrocardiogram classification. First, the global and local structures of the ECG signal are maintained through low-rank and sparse constraints, At the same time, a local constraint term is introduced to inherit the manifold structure of the training samples, so that similar samples have similar coding coefficients. Second, to reduce the distribution difference of encoded data from the same class, construct a Laplacian matrix using label information, making the learned sparse representation minimize intra-class compactness and maximize inter-class separability, with stronger discriminative ability. Finally, a dictionary with stronger discrimination capabilities is obtained through dictionary learning algorithms.
Journal
IF:
4.9
Papers:
9.7K
Citations:
2.4W
Organization
Cited Papers