Wavelet-Based Detrending for EMG Noise Removal

Myoelectric Signals (MES) have a long tradition with regard to prostheses control. Due to the signals' nature, MES are prone to interference and noise. Various methods exist for preprocessing these signals before classification algorithms to derive control information are applied. While these methods help to improve the source signals, parameters must be carefully selected and implemented on a case-to-case basis. After presenting several noise removal methods and drawbacks, we introduce a novel approach by applying wavelet detrending to the signal. The approach brought forward yields an excellent signal-to-noise ratio and provides in some cases a complete removal of noise interference. Weak signals and muscle fatigue do not impact the results. Besides serving as input for various classification methods, the detrended signal can also be directly used for implementing robust control methods like Cookie Crusher or threshold algorithms. A basic Cookie Crusher control model was chosen to verify the approach in comparison to traditional amplitude level schemes. Results show that detrended signal data can be utilized for reliable prosthesis control even for users exhibiting low amplitude MES.