Enhanced Electromyogram Signal Denoising Using Canonical Correlation Analysis Informed Variational Mode Decomposition

Abstract

Electromyogram (EMG) signal is often contaminated with noise and artifacts of different sources. These negatively affect the classification and recognition of motion and gestures. Hence, denoising the signal prior to further processing is crucial for better EMG applicability. Different denoising techniques have been proposed in the past which are based on either wavelet transform, blind source separation, or mode decomposition techniques; however, improvement in these techniques is warranted nonetheless. In this paper, a novel approach for noise and muscle artifact removal is proposed, employing a hybrid framework combining variational mode decomposition (VMD) with canonical correlation analysis (CCA). The proposed framework uses VMD to decompose the signal into intrinsic mode functions (IMFs) and subsequently leverages CCA to isolate and refine the noisy IMFs prior to signal reconstruction. The framework outperforms state-of-the-art EMG denoising techniques like EMD and VMD. The framework was tested on multisubject data acquired for multiple motions. Experimental results show significant improvements in signal quality, evaluated using signal-to-noise ratio (SNR), percentageroot-mean-square difference (PRD), and root mean square error (RMSE) metrics. This approach is an effective signal processing tool especially for post-acquisition analysis in medical diagnostics and research.