Multichannel Eigenvalue Decomposition of Hankel Matrix-Based Classification of Eye Movements From Electrooculogram

In this letter, we propose a new method named multichannel eigenvalue decomposition of Hankel matrix (MCh-EVDHM) for decomposition of multichannel signal into a set of significant elementary components. The decomposition of a multichannel synthetic signal using MCh-EVDHM is presented, and the decomposed significant elementary components of each channel are found to be perfectly aligned. The proposed MCh-EVDHM technique is successfully used to design a framework for classification of six eye movements from the two-channel electrooculogram (EOG) signal. The validation performance metrics of the proposed framework are compared with the existing methods and found to be achieving highest specificity of 99.27% and comparable accuracy, sensitivity, and precision. Further, the proposed framework with ensemble bagged tree classifier has achieved the testing accuracy of 100%. The model memory size of the classifier in the developed framework is also presented, and the linear support vector machine classifier has the smallest model size of 147 kB with testing accuracy of 98.33%. The experimental results show that the proposed framework can be used for the development of an EOG-based lightweight human–computer interface system.