Classification of Body Movements in Wearable ECG (W-ECG) Signals Using Artificial Neural Networks

The wearable electrocardiogram (W-ECG) signal inherently contains motion artifacts due to various body movements of the wearer. The W-ECG signals with four body movement activit ies (BMAs) ‒ left arm up-down, right arm up-down, waist-twist and walking of five healthy subjects have been acquired using the wearable ECG recorder. The classification of these four BMAs has been performed using artificial neural networks (ANN). In the process, the motion artifacts contained in the captured W-ECG signals have been extracted using Wavelet transform and the features of the motion artifacts have been extracted using Gabor transform. These feature vectors are fed to a multi-layered perceptron neural network (MLPNN) consisting of ten neuron hidden layer. The overall classification accuracy achieved using ANN is close to 92%.