Improved Classification Scheme Using Fused Wavelet Packet Transform Based Features for Intelligent Myoelectric Prostheses

Electromyography (EMG) signal is gaining popularity to develop intelligent bionics and prosthetic devices using machine learning techniques. Feature extraction is essential step for the EMG pattern recognition based application. In this article, a fused wavelet packet transform based feature extraction approach is proposed for EMG pattern classification. Total nine subjects (six intact and three amputees) are recruited for the data acquisition. Data acquisition is performed by an ADS1298-based system with eight bipolar electrodes. Further 11 activities are performed by each subject at the time of EMG signal recording including lateral grasp, cylindrical grasp, spherical grasp, and grasp with force. The visual feedback system is utilized for EMG signal acquisition of amputees. The comparison of commonly used wavelet transform based features and proposed fused wavelet transform based features is also presented with respect to classification accuracy and time complexity. The proposed method exhibits highest classification accuracy up to 98.32% for the amputees using discriminant analysis classification with marginal variation in time complexity. Similar trends in results are observed when standard dataset (NinaPro) has been utilized. The results validate the enhanced performance of the proposed technique over conventional counterparts.