Design of a Computationally Efficient 1-D Convolutional Neural Network Hardware for ECG Classification in Portable Devices

Even though several machine learning techniques already exists for classification of Electrocardiogram (ECG) signals, most of them are less resource efficient for a viable incorporation inside a portable device especially wearable medical device. This paper proposes a computationally efficient 1-D convolutional neural network (CNN) with global maxpooling layer, which is able to classify the ECG signals with an accuracy of 98.26 % while the number of parameters is reduced by 62%, when compared to the latest existing work. A fully pipelined hardware architecture is also proposed for the implementation of this 1-D CNN structure. The synthesis results on field programmable gate array (FPGA) platform shows the resource efficiency of the proposed hardware over the other CNN architectures for ECG classification.