An ECG Sparse Noise Reduction Method based on Deep Unfolding Network

ECG is a kind of weak body surface signal that is easily disturbed by noise during the collection process. The traditional ECG signal denoising technology depends on effective filters, which is artificially created by experience. Once the form of the signal is updated, the inherent space may no longer be suitable for this problem. As the deep learning method can learn sparse features from the data without manual intervention. We designed a deep learning process to apply the powerful functions of neural networks to the inference of the ECG sparse noise reduction model, which can also solve the optimization problem in sparse signal processing. By using this method of deep expansion, an optimization strategy is proposed, which turns the iterative optimization problem into constructing a new network framework. In this way, the model parameters can be easily solved through cross-layer. Through experimental verification, our method improves the SNR by 83.29% compared with the current advanced method.