Building EEG-based CAD object selection intention discrimination model using convolutional neural network (CNN)

Currently, building natural interaction systems based on physiological signals has become an crucial requirement for the development of Computer Aided Design (CAD). As the first step of model operation in CAD, object selection is essential and the efficiency of selecting has a great impact on the experience of users. In the research community, gaze-based interaction for object selection has been well-established. However, this interactive mode is still imperfect due to Midas touch problem. In this work, a selection intention discrimination (SID) model is implemented to decode electroencephalogram (EEG) signals generated during object selection process. Common Spatial Pattern (CSP) is applied to extract spatial features from EEG in four frequency bands. Then these features are learned by a Convolutional Neural Network (CNN) equipped with an adaptive weights training module to realize the SID. To verify the decoding feasibility of this model, a cognitive experiment related to object selection is conducted. The empirical result shows that the performance of this model is good. It turns out that EEG-based object selection is feasible, which can be a intuitive and natural interaction mode for CAD.