A Framework for Motion Recognition Using Electroencephalogram-Based Brain–Computer Interface and Pretrained Convolutional Neural Networks Enhanced by High-Resolution Superlet Transform

The precise classification of motor imagery (MI) electroencephalography (EEG) signals is foundational for advancing brain–computer interface (BCI) systems applicable in practical scenarios. This study proposes a novel methodology aimed at the accurate recognition of MI activities derived from cerebral signals. Such advancements hold the potential to significantly enhance the interaction between individuals with mobility impairments and their environment, facilitating a higher degree of autonomy. A critical obstacle in this domain is the creation of algorithms capable of reliably identifying MI tasks, thereby enabling those with mobility challenges to securely and efficiently manipulate external devices and appliances through cognitive commands alone. There are several ways to learn the characteristics of EEG signals, but deep learning has received less attention to develop novel models of EEG features and enhance the categorization of motor imagery performance. This paper utilizes superlet transform (SLT) to transform an EEG signal into its two-dimensional (2-D) representation. This 2D representation of segmented EEG signals is fed to an adapted pretrained network and then the classifier classifies the EEG signal into hand and foot movements. The classification accuracy attained by the proposed framework is 99.9%. The effectiveness of the suggested technique is evaluated in comparison to other current algorithms in terms of performance metrics. The findings show that the proposed approach accurately recognizes motor image classes using EEG inputs.