Generating ten BCI commands using four simple motor imageries and classification by divergence-based DNN

The brain computer interface (BCI) systems are utilized for transferring information among humans and computers by analyzing electroencephalogram (EEG) recordings. However, in this emerging research field, the number of commands in the BCI is limited in relation to the number of motor imagery (MI) tasks; in the current literature, mostly two or four commands (classes) are studied. As a solution to this problem, it is recommended to use mental tasks as well as MI tasks. Unfortunately, the use of this approach reduces the classification performance of MI EEG signals. The fMRI analyses show that the resources in the brain associated with the motor imagery can be activated independently. It is assumed that the activity of the brain produced by the MI of the combination of body parts corresponds to the superposition of the activities generated during each body part’s simple MI. In this study, in order to create more than four BCI commands, we suggest to generate combined MI EEG signals artificially by using tongue, feet, left and right hands’ motor imageries in pairs. For the first time in the literature, combined MI EEG signal is artificially generated by using the superposition of simple MI EEG signals from two different sources. We observe in the literature that the classification performances are adversely affected as the number of classes is increased, and the classification performances for the MI with more than four classes are poor. The aim of this study is to increase the BCI commands by generating artificially combined MI EEG signals and to achieve high success rates for ten BCI commands by using a small-sized deep neural network (DNN). In this context, by analyzing the ERD signal of combined MI tasks, we investigate how to generate combined MI signals artificially using the superposition of simple MI signals. The proposed method is validated on real data which consist of simple and combined MI EEG signals, and average classification performance of 81.8% is achieved for ten BCI commands generated from the BCI Competition 3 and 4 datasets.