EEG Image Classification using Deep Metric Learning with Nearest Neighbor Technique

Many Deep Learning (DL) applications have been able to attain the best possible outcomes. The DL model is used for categorizing electrical brain signals for the interface between the brain and computer. Fast Fourier Transform converts time-domain Electroencephalogram (EEG) data to frequency-domain data. The Deep Metric Learning (DML) model using the nearest neighbor is used to train a spectrogram composed of 72 concatenated EEG channels. There were three sets of triplet pairings of epochs labeled positive, negative, and anchor in the training batch. The Recurrent Neural Network (RNN) is combined with a triplet loss function for minimizing the Euclidean distance in embeddings of the identical class. RNN can also increase the distance among the embeddings of various labeled images to train an embedding feature. An EEG dataset of 22 untrained people was used to test the suggested approach. It was demonstrated that the DML method using the nearest neighbor can be used for training samples in a way that 100 EEG trials for a single subject. This model is used in the modification of the well-known problem of the substantial inter-individual inconsistency of human Motor-Imagery Brain-Computer Interfaces (MI-BCI) EEG with an accuracy of 0.75 percent.