Breathing Pattern Recognition By the Fusion of EMG and Acceleration Signals

Breathing plays an important part for human beings in our daily life. Besides physical parameters like tidal volume or respiratory rate, biomedical signals like electromyography (EMG) signals can be a potential candidate for breathing activity monitoring. In this work, we propose a novel scheme for breathing activity pattern recognition by fusing features extracted from both EMG and acceleration signals. The EMG signals and acceleration signals during four breathing activities usually used in our daily life, including normal breathing, fast breathing, coughing, and deep breathing, are captured. The raw data is preprocessed, feature extracted by several hand-crafted features, and pattern classified. The performance of five EMG feature sets, five acceleration feature sets, and two machine learning algorithms are evaluated. The best result achieves an accuracy of 82.20% using an EMG feature and an acceleration feature with a support vector machine (SVM) classifier. It shows that fusing EMG and acceleration data is better than EMG signals alone or acceleration signals alone, and it also raises the problem of finding the best features to reach higher performance. To the best of our knowledge, this is the first time that EMG signals are combined with acceleration signals for human breathing activity classification. The proposed approach is effective and explores a new way of human breathing monitoring.