Automatic Reconstruction of Noisy Electroencephalography (EEG) Channels with Transformer-Based Architectures for Sustainable Systems

In the field of Brain-Computer Interface (BCI), electroencephalography, commonly referred to as EEG, is a neuroimaging technology that experiences widespread application. EEG signals, on the other hand, are susceptible to a wide variety of noises and artifacts, which can lead to a decrease in the quality of the data. Using transformer-based architectures, including attention-based models, the purpose of this research is to establish an automatic approach for reconstructing noisy EEG channels. There were three different noise scenarios that were taken into consideration, and they were zero-mashing, noise masking, and zero-noise masking. The reconstructing model combines convulsion neural networks (CNN) with a multi-head self attehtion transformer model. We designed the model to balance both performance and sustainability considerations. The findings indicate that the model was able to accomplish a high level of reconstructing skills, as indicated by the value of R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>, which is equivalent to 0.8563 within a margin of error of 0.0543. In addition, considerations of sustainability were taken into account for the actual execution where the model size was restricted to less than 0.5 MB.