The ECG Signal Denoising Method Based on Temporal Convolutional Network and Attention Mechanism

Denoising electrocardiogram (ECG) signals is a critical task in arrhythmia analysis and diagnosis. However, traditional signal processing methods face limitations when addressing the challenge of noise spectrum overlapping with the signal spectrum. To tackle this issue, this paper proposes a deep learning model, TACA-Net, which is based on Temporal Convolutional Networks (TCN) and channel attention mechanisms, for efficiently denoising noisy ECG signals. TACA-Net leverages temporal dilated convolutions to capture multi-scale temporal features, integrates ECA and SE channel attention modules to enhance the extraction of critical information, and incorporates residual connections to improve training stability and signal reconstruction quality. Experiments conducted on the MIT-BIH Arrhythmia Database and Noise Stress Test Database demonstrate that TACA-Net outperforms traditional deep learning methods in terms of Signal-to-Noise Ratio Improvement (SNRimp), Root Mean Square Error (RMSE), and Percentage Root Mean Square Difference (PRD), particularly under low signal-to-noise ratio (SNR) conditions. This study highlights TACA-Net's robustness and efficiency in complex noisy environments, offering a novel approach to ECG signal denoising and providing significant reference value for signal denoising research.