Feature analysis for Intention Decoding in EMG Signals

Abstract

Prosthesis implants had been adored around the globe for amputee. The further advanced hand prostheses using electromyography signals getting common. The immense importance assisting biomedical domain the robotic applications are immersing. Command to our muscles generated by our mind are equivalent to EMG signals in electrical terms, therefore configuration of these signals reckon in the muscle activities. The aim of this study is synthesis of EMG signals produced during hand movement and analysis of the data by adopting different methods. To get the best results these signals are needed to be preprocessed to reduce the noise and unwanted signals inferred while recording data, as data can be collected by installing electrodes on patient’s hand. NinaPro shared the data of previous the researchers of various streams on numerous hand movements, which simplifies the prosthesis implants to accomplish better movements. Intention of this thesis is to dive deep into the analysis of this available data set for its different classes, strategy to be followed is the processing of the EMG signals and then formation of window segmentation facilitating for smooth management of data for further steps. Moreover, to extract features which is the task to be achieved in a way to get a control on the hand prosthesis in frequency domain. Furthermore, the classifiers Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), K-Nearest Neighbor (KNN) and Artificial Neural Network (ANN) applied on the extracted features and highest of 85.31% accuracy achieved with conventional classifier KNN. However ANN from deep learning gave far better results than the machine learning techniques.