Multidimensional Feature Extraction of EMG Signals: RMS, Median Frequency, and Mean Power Frequency

Multidimensional feature extraction of electromyographic (EMG) signals represents a crucial component in biosignal processing. By extracting different characteristic parameters, we can comprehensively reflect the state and variations of muscle activity. Commonly used time-domain and frequency-domain features include Root Mean Square (RMS), Median Frequency (MDF), and Mean Power Frequency (MPF), each describing EMG signal characteristics from distinct perspectives.

Root Mean Square (RMS) serves as a typical time-domain feature that reflects amplitude variations in the signal. The computational algorithm involves squaring the signal values, calculating their mean, and then taking the square root of the result. This effectively characterizes the intensity level of EMG signals and is commonly applied in assessing muscle activation levels and fatigue conditions. Implementation typically involves: RMS = sqrt(mean(signal.^2)) where the signal array undergoes element-wise squaring before averaging.

Median Frequency (MDF) belongs to frequency-domain features, representing the midpoint frequency in the power spectrum where the spectral power is divided equally into two halves. In muscle fatigue analysis, median frequency typically decreases as fatigue increases, serving as a crucial indicator for monitoring muscle functional states. Algorithm implementation requires computing the power spectral density (using FFT or Welch's method) and finding the frequency that bisects the total power.

Mean Power Frequency (MPF) similarly employs frequency-domain analysis, calculating the weighted average frequency of the power spectrum. Like median frequency, it's frequently used in muscle fatigue monitoring, but its calculation method demonstrates higher sensitivity to high-frequency component changes, thereby offering better discrimination capability in certain applications. Computational approach involves: MPF = sum(frequency × power_spectrum) / sum(power_spectrum) where frequency bins are weighted by their corresponding power values.

By integrating these multidimensional features, researchers can achieve more comprehensive analysis and evaluation of EMG signals, providing richer data support for studies in sports medicine, rehabilitation engineering, and human-computer interaction domains. Feature extraction pipelines often combine these algorithms with signal preprocessing steps like filtering and segmentation for robust biomedical applications.