ANALYSIS OF STFT WINDOW PARAMETERS ON EMG SIGNAL BASED FINGER MOVEMENT CLASSIFICATION

Abstract

A spectrogram is essential in analyzing EMG signals for finger motion recognition. It relies on STFT parameters like window size, overlap, and window type for accuracy. Optimal parameter selection is challenging due to EMG sensitivity to minor changes affecting recognition accuracy. The study employs AlexNet to recognize spectrograms from EMG signals, using various STFT parameter combinations for five finger movements.

Results show that a window size of 100, 50% overlap, and Hamming window outperform other combinations. A window size of 100 consistently outperforms 200 and 300, while a 50% overlap is better than 25% and 75%. Hanning window types consistently outperform Hamming, Blackman, and Tukey. This research streamlines EMG spectrogram analysis for efficient finger motion recognition.