In this project, I tried to develop a new DL architecture for extracting focus index from EEG signals. The designed Neural Network has been compared with the famous structure like VGG16, Resnet,... and had better results.
Data Augmentation:
to make data able to use the two-dimensional convolutional network, the shape of the data has been reformed. In this way, first, all of the seconds related to the same subject and same Video ID, resulting in the same label have been founded. Then, chosen every 50 time-point with their features has been chosen as input. The step size to have the next data is one. By doing so, the form of the dataset changed to (7811,50,14).
Training:
The designed network has been shown in the following network. This network has been inspired by the EEGNet which is well-known for analyzing EEG signals. The obtained results from this network have been compared with the famous networks for image and signal processing like VGG16, Inception, and Resnet. Results show that the designed architecture is able to extract the focus index from EEG signals more accurately.
The full report could be found here: