Research on Electrooculography Classification based on Multiple Features

The increase in disabled motor functions has made the bio-based human computer interface (HCI) a research hotspot for assistive robots. This paper studies the electrooculography (EOG) classification algorithm, which plays a crucial role in EOG-based HCI. Only when EOG signals are correctly classified can HCI obtain correct control commands. First, the entropy concept is introduced to describe the four kinds of EOG signals characterized through waveform analysis. Entropy, spatial entropy, and peak-to-valley ratio are used to compose feature vectors that have less feature dimensions. Second, the back-propagation (BP) neural network, classification support vector machine (C-SVM), and particle swarm optimization-SVM (PSO-SVM) have been used to classify four kinds of eye movement signals. Finally, the BP and SVM models are built, and the penalty parameter C and kernel parameter y are optimized. The BP and C-SVM classifiers yield the same classification accuracy rate of approximately 96%, but the hybrid PSO-SVM obtains the highest classification accuracy rate of 98.71% and exhibits more stable classification performance than the BP network. The correct classification rates of the PSO-SVM application techniques are very high. Thus, such techniques can be used to classify EOG signals for HCI.