EMG Signal Analysis and Machine Learning for Abnormal Movement Identification

Electromyography (EMG) enables the measurement of electrical activity generated by human muscles. This technique has widely been applied to rehabilitation, sports science, prosthetics development, and diagnosis of neuromuscular disorders. In this paper, we use three different EMG-based analysis methods, that is, Fast Kurtogram (FK), Hilbert-Huang Transform (HHT), and Discrete Wavelet Transform (DWT), to detect human false gait cycles. This work uses EMG signals from healthy subjects during walking cycles as reference data. The efficiencies of the analysis methods are shown and then compared. Moreover, the features extracted from the HHT and DWT methods are labeled and input into a machine learning model, which is trained via a medium neural network, to identify abnormal movements of humans.