基于三域特征提取和GS-SVM的ECG信号智能分类技术研究

Research on intelligent classification of ECG signals based onthreedomain features extraction and GS-SVM

摘要:

近年来，基于单域的特征提取方法已经得到广泛的研究，并被用于心律失常的分类检测.事实上,多域特征提取在其分类中往往表现得更好.本文利用MIT/BIH心律失常数据库中的48组ECG信号进行预处理，从时域、频域和小波域提取了信号的三域特征，这些特征从各个方面充分表征了ECG信号的性质.再利用基于网格搜索的SVM结合归一化特征可将ECG信号划分为常见的4类.该方法的总体精度达到98.01%，f1分值为0.9800，对ECG信号的检测性能良好，相对目前绝大多数ECG信号分类器具有更强的泛化能力.

Abstract:

Singledomain based feature extraction has been extensively studied and is used to detect and classify Arrhythmia recently. In fact, multidomain feature extraction tends to perform better in classification. In this paper, threedomain features are extracted from timedomain, frequencydomain and waveletdomain using preprocessed ECG signals taken from 48 data sets in MIT/BIH arrhythmia database. These features fully characterize the nature of the ECG signals from various aspects. In the final stage, ECG signals are classified into four classes by the normalized features combined with gridsearch based SVM, the overall accuracy and F1 score of the proposed method is 98.01% and 0.9800 respectively, it performs well in detection and classification of ECG signals and has better generalization against the most results reported so far.