Electromyography (EMG) signal based pattern recognition have been applied for various applications especially in Human-Machine Interface (HMI). Most of previous research works focused on human muscles which related to movements of arms and fingers for controlling body parts through intention. Statistical analysis has been used in most research works on muscle assessment and performance measurement. There are only few literatures on EMG pattern recognition study that used other than arms or hands. Ths thesis will focus on the use of EMG pattern recognition to classify the human breathing task. Four respiratory muscles have been chosen for EMG data acquisition i.e. sternocleidomastoid, scalene, external intercostal and diaphragm. The selected human subjects are used to perform four breathing tasks which are quiet, deep, deep & hold and fast breathing. The feature extraction of EMG consists of basic features and its convolution. The basic features are Root Mean Square (RMS), Zero Crossing (ZC), Mean Frequency (MF) and Mean Power (MP). The convolution is performed between pairs of the bsic features. The EMG pattern recognition is performed by using K-Nearest Neighbor (K-NN), Multilayer Perceptron Artificial Neural Network (MLP-ANN) and Support Vector Machine (SVM). Segmentation window size is configured at 1000 ms. Results showed highest accuracy of 96.86% on convolution of RMS and MF features using SVM classifier. The convolution feature extraction will enhance the EMG classification accuracy for human breathing compare to basic features of time and frequency domain. Study on classification of human breathing activity based on EMG of respiratory muscles can be implemented in biofeedback rehabilitation. The EMG from respiratory muscles can be used in physical therapy for disabled person by controlling an assistive device such as robotic limb and electric wheelchair.
