Finger movements classification based on fractional Fourier transform coefficients extracted from surface EMG signals

Abstract

EMG signals have played a pivotal role as a fundamental component of myriad modern prostheses to control prostheses’ movements as well as identifying individual and combined hand or finger gestures. Despite a great deal of interest in these signals, the non-stationary nature of biological EMG signals has led to complications in EMG applications. Stationary signals have been analyzed plainly by time domain approaches like Fourier Transform, while non-stationary signals analysis is not satisfactory to be carried out with such method as it is not capable to illustrate the incidence time of various frequency components, besides, extracting both time and frequency information is essential. The Fractional Fourier Transform (FrFT), which is a generalization of classical Fourier Transform, is able to demonstrate the variable frequency of non-stationary signals. In this paper, FrFT technique with different fractional orders is employed as a novel and sophisticated feature extraction method for EMG signals of 8 subjects including 6 men and 2 women recorded in 10 different finger movements which are 5 individual and 5 combined postures. Windowing method and the t-test approach are utilized to select the best FrFT extracted coefficients as features. Employing KNN method to classify 10 different classes and the average classification accuracy of 98.12% of proposed method is a significant indicator of its sufficient performance.