Binary neural network-based lightweight ECG recognition for edge devices

Automatic electrocardiogram (ECG) recognition is an important topic in medical diagnosis. However, traditional deep learning models usually require large computational resources when requiring high-precision recognition, which is not suitable for resource-constrained edge devices. To tackle this challenge, a binarized neural network (BNN) based lightweight ECG recognition method is proposed in this paper, which effectively reduces the computational costs by binarizing the network weights and introducing a scaling factor to simulate a full-precision model. But when the model parameters are binarized, slight numerical changes near the threshold may cause huge output fluctuations, so an improved batch normalization is proposed, which increases the stability of the model by adding bias to the sample features. During model training, we selected Adam as the optimizer, which has faster convergence speed and model accuracy compared to the original SGD. We conducted comprehensive experimental analysis including ECG image and ECG sequence recognition on the MIT-BIH arrhythmia dataset. The experimental results show that the proposed BNN model maintains high recognition accuracy while increasing model efficiency compared to traditional floating-point neural networks. The method provides an effective solution for low-power, real-time ECG recognition for edge devices.