Design of Assistive Human-Machine-Interface control signal classifiers using the Discrete Cosine Transform

• Main Author: Murali, Swetha
• Format: Others
• Published: OpenSIUC 2008
• Online Access: https://opensiuc.lib.siu.edu/theses/445; https://opensiuc.lib.siu.edu/cgi/viewcontent.cgi?article=1452&context=theses

The ultimate success of a human-machine-interface (HMI) designed to assist severely disabled individuals depends on how accurately the control signal generated by the individual is classified. This theses, therefore, is aimed at developing a generalized strategy to accurately classify single-channel and multi-channel HMI control signals. The primary focus is on the design of Gaussian multivariate signal classifiers with particular emphasis on overcoming the dimensionality problem frequently encountered in the design of assistive HMI multivariate signal classifiers. The dimensionality problem is overcome by selecting a small set of linear combinations of the input signal space generated by the discrete cosine transform (DCT). Issues dealing with the selection of the basis vectors of the DCT for multi-class signal classification problems are addressed. Four different class-dependent ranking criteria are introduced to select basis vectors from the transformed training vectors in the 1-dimensional and 2-dimensional DCT domains. The application of the resulting DCT based Gaussian multivariate classification strategies are demonstrated by classifying two distinct HMI physiological control signals that have been proposed for novel assistive applications: 1) single-channel tongue-movement ear-pressure (TMEP) bioacoustics' signals and 2) multi-channel bioelectric event-related potentials (ERPs). Classification results show that the DCT based classifiers outperform classifiers described in previous studies. Most noteworthy is the fact that the Gaussian multivariate control signal classifiers developed in this thesis can be designed without having to collect a prohibitively large number of training signals in order to satisfy the dimensionality conditions. Furthermore, the formulation of the classification strategy is general and it can be customized for each user simply by training the classifier with the physiological control signals of the user. Consequently, the classification strategies will be especially beneficial for designing personalized assistive HMIs for severely disabled individuals from whom only a limited number of training signals can be reliably collected.