Efficient deep neural network model for classification of grasp types using sEMG signals

Grasping is a challenging problem in robotics and prosthetic applications due to its control requirements. The visual perception and analyzing electromyography (EMG) signals are the two ways to give the inputs to robots and prosthetic amputees for grasping abilities. The EMG is a diagnostic manner that evaluates the fitness condition of skeletal muscles. Examination or evaluation of the EMG signals is time-consuming and arduous for experts. Hence, the state-of-the-art methods in artificial intelligence (AI) is employed to improve the accuracy rate for the detection and classification of EMG signals for grasping. Recently, deep learning architectures have been used in many engineering applications such as diagnosis of health conditions, computer vision, and human machine interaction (HMI). In this study, a new deep one-dimensional convolutional neural network model (1D-CNN) is proposed to classify six types of hand movements. Our proposed 1D-CNN model implemented using surface EMG (sEMG) has obtained the highest accuracy of 94.94% in classifying six hand movements. The strength of our model is that, it can perform the automated classification of various hard grasps using only one channel data. Our developed prototype model is ready to be tested with more data and can be used to assist in musculoskeletal disorders.

Automated summary

The study introduces a new deep one-dimensional convolutional neural network model (1D-CNN) for classifying six types of hand movements. Using surface EMG (sEMG), the proposed 1D-CNN model achieved the highest accuracy of 94.94% in classifying six hand movements. The developed prototype model is prepared for further testing with more data and can be utilized to assist with musculoskeletal disorders.