Power Spectra Images for Mental Imagery Tasks EEG Classification

Access to daily activity has a potential impact on differently-abled individuals.BCI-based EEG devices have emerged as a potential aid to improve daily assistance, using only brain signals as a communication path.The EEG signals of mental imagination of any action, specifically visual imagery, are challenged in classification due to the diversity and variety of neural activity patterns.This study specifically concentrates on the EEG signal classification of imagery mental tasks, employing the imagination of turning light on and off.Electroencephalogram signals were recorded using a NeuroSky headset.Our methodology involved comparing raw data, extracted features, and power spectra images.These data were then fed into recurrent neural networks (RNNs) and deep neural networks (DNNs) for task recognition.Results indicate that image power spectra images, which identify EEG signal frequencies, are the most significant, and the classification outperformed raw data and extracted features.Notably, greyscale power spectra images achieved the highest accuracy, reaching 97.4% through a deep-learning network.The superior performance of image classification suggests its efficacy in discerning imagery tasks.In conclusion, greyscale power spectra images emerge as the most suitable data type for classifying imagery tasks, showing a clear pattern of imagery tasks.