Surface myoelectric signal analysis and enhancement for improved prosthesis control

In this thesis, novel signal processing and machine learning techniques are presented in the field of myoelectric control. Specifically, algorithms for activity detection, noise identification and noise reduction are introduced, evaluated and discussed. The ultimate aim has been to develop algorithms to improve the performance of prosthetic control systems that use myoelectric signals. Such systems must be an ability to distinguish between electromyographic signals and background noise. For this, the behaviour of One-Dimensional Local Binary Pattern histograms were used to identify the presence of myoelectric activity in recorded signals that originated from electrode sensors on the surface of the skin. This technique was compared against two other activity detection methods and it was found to give better performance in some circumstances. In particular, a lower False Positive Rate was achieved. Noise is always present in myoelectric signals, and if it can be identified then steps can be taken to quantify and/or mitigate it. Pattern recognition was used to identify a single noise type in pre-recorded myoelectric signals. A set of Radial Basis Function Support Vector Machines were trained and tested on clean myoelectric signals that have been artificially contaminated with five typical noise types. The behaviour of the features and the nature of the confusion are discussed. Identification was shown to be possible, but confusion between noise types grew as the SNR increased. Spectral Enhancement, which is normally used on speech signals, is applied to myoelectric signals in an attempt to mitigate noise. Spectral Enhancement based on Improved Minima Controlled Recursive Averaging (IMCRA) was found to improve the classification accuracy, and by corollary the signal quality, with signals that had white noise artificially added (which can be present in recorded myoelectric signals) and with intrinsically noisy signals. The improvement was higher when fewer channels were used.

Name dc:type.qualificationname

phd

Level dc:type.qualificationlevel

doctoral-pg

Grantor dc:publisher.institution

University of Strathclyde

Year dc:date.issued

2014

Author dc:creator

• McCool, Paul

DOI

10.48730/7xqn-je53

Identifier

T13733

OAI identifier oai:identifier

oai:strathclyde:pr76f344p