Towards Improving the Saccade Angle Recognition Using the Sensitivity Weights of Channels

The classification of eye movements provides an effective means for human–computer interaction (HCI). Although numerous interfaces based on eye movement have been developed, systems employing the electroencephalogram (EEG) face challenges due to weak features, often resulting in unstable classification performance. Consequently, exploring methods to accurately identify saccade angles through EEG signals holds great significance.

This study presents a feature weighting method optimized for recognizing saccade angles, accounting for the sensitivity of different channels to saccade movement. The proposed method involves calculating the sensitivity weights for each channel in response to saccade signals and then integrating these sensitivity weights with original feature data to generate enhanced features.

In our study, ten subjects were recruited to participate in the collection of saccade signals. To validate the proposed method, we used four distinct feature extraction methods and conducted four comparative experiments. The results show that the average classification accuracy using the sensitivity weights reached 91%, significantly outperforming other methods by margins of 43.71%, 4.98%, 23.19%, and 20.99% respectively.

The sensitivity weights enhance classification effectiveness across different feature extraction methods. Experimental results confirm the method's efficacy in reflecting the sensitivity of different channels to saccade movement. Future research will explore the application of the sensitivity weights in regression analysis of saccade angles to further its application in continuous control.