A novel approach to processing electroencephalography (EEG) signals has emerged, leveraging the utilization of signal images. The application of deep learning techniques in bypassing complex signal and image processing tasks has generated significant interest in this field. However, challenges remain in signal image processing, particularly in handling significant features and image sizes. This study presents a comprehensive investigation of EEG motor imagery signal processing, focusing on the classification of three tasks: eating, drinking, and seeking assistance. Fast Fourier Transform (FFT) is employed to extract signal image features, which are subsequently utilized in a deep learning framework. EEG data were collected from five subjects, and four transfer functions of deep learning models, namely VGG16, VGG19, ResNet50, and ResNet101, were employed for training and classification purposes. The performance of the four models was meticulously evaluated and compared. Notably, VGG16 exhibited superior performance in accurately classifying the EEG motor imagery tasks, achieving an impressive accuracy of 90%, sensitivity of 84%, and specificity of 92%. In conclusion, this study underscores the efficacy of EEG signal image processing through deep learning-based classification techniques. The findings highlight the potential of utilizing signal images in EEG analysis for motor imagery tasks, thereby contributing to the advancement of brain-computer interface technology and enhancing our understanding of neural dynamics.
