Long-duration electrocardiogram classification based on Subspace Search VMD and Fourier Pooling Broad Learning System

Abstract

Detecting early stages of cardiovascular disease from short-duration Electrocardiogram (ECG) signals is challenging. However, long-duration ECG data are susceptible to various types of noise during acquisition. To tackle the problem, Subspace Search Variational Mode Decomposition (SSVMD) was proposed, which determines the optimal solution by continuously narrowing the parameter subspace and implements data preprocessing by removing baseline drift noise and high-frequency noise modes. In response to the unclear spatial characteristics and excessive data dimension in long-duration ECG data, a Fourier Pooling Broad Learning System (FPBLS) is proposed. FPBLS integrates a Fourier feature layer and a broad pooling layer to express the input data with more obvious features, reducing the data dimension and maintaining effective features. The theory is verified using the MIT-BIH arrhythmia database and achieves better results compared to the latest literature method.