Enhancing Motor Imagery Classification Accuracy in Controlled and Uncontrolled Environments Using Convolutional Neural Networks

The classification of motor imagery (MI) tasks is one of the main issues in the field of brain–computer interfaces (BCIs). However, the majority of these BCI applications are confined to controlled environments to prevent out-of-lab artifacts from contaminating brain signals and consequently influencing the decoding performance. Moreover, recently, deep learning algorithms have revolutionized many applications, such as natural language processing and image recognition. Hence, the BCI community has grown interest in DL approaches. In this paper, we utilize one of the prominent DL models, i.e., a convolutional neural network, to evaluate a motor imagery under distraction dataset, which acquires data in simulated conditions of real-life environments. The outcomes of this experiment indicate that the proposed model can enhance the classification accuracy of MI signals in controlled and uncontrolled environments.