Enhancing EMG signal classification using convolution neural network optimized with fractional order bat algorithm

The galvanizing activity of nerves and muscles, known as electromyograms (EMG), is a valuable diagnostic tool for identifying muscle and nerve disorders. Effective identification and classification of normal and abnormal EMG signals such as myopathies and Amyotrophic Lateral Sclerosis (ALS) plays a pivotal role in supporting computerized diagnostic tools, which is particularly important because of the non-stationary characteristic of EMG signals. Therefore, developing an efficient classification system for EMG signals is indispensable to facilitate the computer-assisted prediction of abnormalities. Initially, the smoothed pseudo-Wigner–Ville transform is employed to reconstruct the time series of EMG into time–frequency images. Convolution neural network (CNN) is reconstructed with fractional order bat optimization algorithm for classifying the EMG signals. The efficiency of the developed fractional order bat-CNN was compared with the bat-CNN. Results express that the CNN combined with fractional order bat optimization proves to be highly effective in accurately classifying normal and abnormal EMG images, achieving an accuracy rate of 99.11%. It is also proved that the unification of fractional order bat optimization in CNN has exhibited superior performance compared to CNN with bat optimization algorithm.