| Titre : | EEG Classification for Mind Controlling Applications using Multi-Method Approach |
| Auteurs : | Imad eddine Tibermacine, Auteur ; Christian Napoli, Auteur ; Ahmed Tibermacine, Directeur de thèse |
| Type de document : | Monographie imprimée |
| Editeur : | Biskra [Algérie] : Faculté des Sciences Exactes et des Sciences de la Nature et de la Vie, Université Mohamed Khider, 2022 |
| Format : | 1 vol. (161 p.) / couv. ill. en coul |
| Langues: | Anglais |
| Mots-clés: | Brain-Computer Interface, Deep Learning, Machine Learning, Motor Imagery, Robot controllati dalla mente, Classification. |
| Résumé : |
Since the first discovery of electroencephalography (EEG) principles in the 20’sby Berger, scientists have used EEG signals in diagnosing brain conditions and other wide applications. In the last 2 decades, brain-computer interfaces (BCIs) and their technological advances have allowed people to use EEG for mind-controlling tasks, especially controlling robots by decoding and classifying EEG signals using Deep Learning (DL). However, anomalies in EEG signals induced by human abuses like alcohol and drugs or some diseases like Parkinson’s, have made the classification task very hard. In this thesis, we have implemented a multi-method approach that uses two simultaneous models in order to generalize the motor imagery classification for mindcontrolled robots from healthy patients to drug-addicted and alcoholic patients, and we discuss their applications to quad-rotors and wheeled mobile robots. We also accomplished the mind-controlling task in real time by accelerating our model’s predictions. Finally, we realized the multi-robot controlling task, which would enable patients to control multiple robots (UAVs, wheeled mobile robots,... etc). The test results showed that the patients were able to use the proposed multi-method approach to control the mobile robot. The effectiveness of our study shows the high precision of attentionbased Bi-LSTM compared to the SVM model, and by GCN compared to others, in classifying motor imagery EEG. The results were accurate and achieved the goals of the study. This will be a motivation to apply it to more complicated problems, like Parkinson’s EEG studies. |
| Sommaire : |
Abstract i Resum ´ e ii ´ Astratto iii Acknowledgements iv List of Figures xii List of Tables 1 Introduction 2 1 Theoretical Background 4 1.1 Overview 4 1.2 Definitions of BCI 6 1.2.1 Synchronous or Asynchronous BCI 6 1.2.2 Invasive and Non-invasive BCI 6 1.3 Neurophysiological Foundations of BCI 7 1.3.1 Brain Structure . 7 1.3.2 Electroencephalography Rhythms 9 1.3.3 Effects of Drugs on EEG 10 1.3.4 Brain Imagine Techniques 11 1.3.5 Event-Related Desynchronisations/Synchronisations12 1.3.6 Event-related potentials 13 1.3.7 Steady State Visually Evoked Potential . 15 1.3.8 Concentration and relaxation mental states 16 1.4 Hybrid Brain-Computer Interfaces 17 1.5 Principal ideas behind hybrid brain-computer interfaces 18 1.5.1 Brain-computer interfaces that are a mix or a pure hybrid19 1.5.2 Sequential or simultaneous processing 19 1.5.3 Most prevalent input devices utilized in hybrid BCI 20 1.6 Some instances of hybrid brain-computer connections 21 1.6.1 Pure hybrid BCIs 21 vii1.6.2 Mixed hybrid brain-computer interfaces 23 1.7 Related Works 27 1.7.1 Control of a wheelchair by motor imagery in real time 27 1.7.2 Quad-copter control in three-dimensional space . 27 1.7.3 EEG based BCI for controlling a robot arm movement 3.2.5 Naive Bayes 65 3.2.6 Decision Tree and Random Forest 66 3.2.7 Ensemble Learning 67 3.2.8 Fuzzy Logic 68 3.2.9 Linear Discriminant Analysis 69 3.2.10 K-Means 70 3.3 Machine Learning for EEG Classification 71 3.4 Deep Learning 74 3.4.1 Architecture design choices 74 3.4.2 Activation functions 76 3.4.3 Task specific deep learning trends 76 3.4.4 Input formulation by deep learning architecture 77 3.5 Deep Learning Models 79 3.5.1 Discriminative Deep Learning Models 81 3.5.2 Representative Deep Learning Models 87 3.5.3 Generative Deep Learning Models 92 3.5.4 Hybrid Model. 96 3.6 Conclusion 96 4 Experimental implementation and results 97 4.1 Introduction 97 4.2 Development software and hardware 97 Page ix4.2.1 EEG Headset 97 4.2.2 EEG Electrodes Gel 98 4.2.3 EEG Mice Software 98 4.2.4 Training Hardware 99 4.2.5 Languages and FrameWorks 99 4.3 Proposed Model. 101 4.4 Data Set . 102 4.5 System Overview. 103 4.5.1 Support Vector Machine 103 4.5.2 Attention-based Bi-LSTM109 4.6 Graph Convolutional Neural Network 116 4.6.1 Mathematical Background 116 4.7 Diverse Features Graph Convolutional Neural Network 118 4.7.1 Implementation 118 4.7.2 Results 121 4.8 Time Domain Graph Convolutional Neural Netowk 122 4.8.1 Preprocessing 122 4.8.2 Feature Extraction123 4.8.3 Architecture123 4.8.4 Training 124 4.8.5 Results 124 4.9 Discussion 126 4.9.1 Evaluation study: 126 4.9.2 Comparative study: 128 4.10 Contributions 128 5 Conclusion and future works 130 References |
Disponibilité (1)
| Cote | Support | Localisation | Statut |
|---|---|---|---|
| MINF/779 | Mémoire master | bibliothèque sciences exactes | Consultable |
