Feature extraction for EEG motor imagery signals using a deep neural network

Date

2023-8

Author

Soysal, Rıdvan

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

318
views

160
downloads

Cite This

Compared to traditional machine learning methods, deep learning methods generally have better performance, and they are more able to handle complex data. Moreover, these models learn the features directly from the raw data, eliminating the need for additional feature extraction step. However, in order to benefit from these advantages, deep learning methods need high amount of data. In this study we examined the use of deep learning methods in the field of EEG motor imagery signal (MI) classification. Although in recent years, many researchers have been applying deep learning methods in this area, we notice that EEG MI signal datasets that are highly used in these researches have insufficient amount of data for deep learning. In this thesis, in order to benefit more from advantages of deep learning methods on EEG MI signal classification we looked for a solution to combine the datasets collected in different studies which cannot be combined directly due to variations in the protocols used in collecting data. After combining available datasets each having little amount of data, we created a larger dataset and used this mega-dataset to train a convolutional autoencoder (CAE) based network that can be used as a deep feature extractor (DFE) for MI signals. Afterwards, trained DFE network is tested as a feature extractor for small datasets. Our experimental results show that using such DFE network improves the performance of EEG MI signal classification on these datasets.

Subject Keywords

EEG MI signal classification, Convolutional autoencoder, Deep clustering, Feature extractor, Brain computer interface

URI

https://hdl.handle.net/11511/105182

Collections

Graduate School of Natural and Applied Sciences, Thesis