Please use this identifier to cite or link to this item:
Title: Self-regulating interval type-2 neuro-fuzzy inference system for non-stationary EEG signal processing
Authors: Ankit Kumar Das
Keywords: DRNTU::Engineering::Computer science and engineering
Issue Date: 2017
Source: Ankit Kumar Das. (2017). Self-regulating interval type-2 neuro-fuzzy inference system for non-stationary EEG signal processing. Doctoral thesis, Nanyang Technological University, Singapore.
Abstract: Motor-imagery based Brain Computer Interface (BCI) provides a direct communication pathway between the brain and a computer based on the neural activities generated by the brain. Such a technology enables people with physical disabilities to communicate with the external world without using their peripheral nerves and muscles. Moreover, BCI systems can also be used in field of gaming, robotics and human-computer interaction, in general. A typical motor-imagery based BCI system consists of a brain activity acquisition phase to obtain ElectroEncephaloGram (EEG) data using multiple channels, a brain signal processing phase to decode the user intentions and a translation phase for transforming the decoded information to control external devices. However, the usage of multiple channels may lead to several issues: a) Longer preparation time; b) Redundant channels with noisy information. In addition, the thoughts of users vary, which may change the data distribution over time (leads to non-stationary nature of EEG). Some of the trials are also affected by the presence of artifacts in EEG data. Further, the frequency response for a motor-imagery task varies for different individuals. These issues can render the BCI system inaccurate with deteriorated performance. The main goal of this thesis is to develop a unified BCI system that address the above mentioned issues. There are four major contributions in this thesis. The first contribution is development of a Robust Common Spatial Pattern (RoCSP) feature extraction algorithm that eliminates the trials affected by artifacts and discards the redundant channels to improve the classification performance of the BCI system. Experimental results show that RoCSP is able to reduce the number of channels and produce a good classification performance. Although, RoCSP provides robust features and improves performance, it does not handle non-stationarity in the EEG data. Recently, it has been shown that interval type-2 fuzzy systems are capable of handling non-stationarity. However, there is a challenge in developing an evolving sequential learning algorithm for interval type-2 fuzzy systems which can learn and evolve the architecture automatically and select the appropriate samples. The second contribution is the development of a Self-Regulated Interval Type-2 Neuro-Fuzzy Inference System (SRIT2NFIS) to handle the non-stationarity in data. A five-layered modified Takagi-Sugeno-Kang interval type-2 fuzzy inference mechanism forms the structure and the learning algorithm is a self-regulatory learning mechanism. The learning starts with zero rules and as a sample is presented to the network, the learning algorithm monitors the knowledge in the sample and decides whether to delete it, learn it or keep it for future use. When a new rule is added to the network, a regularized projection based learning is employed to compute the optimal weights of the network. The RoCSP feature extraction algorithm is combined with SRIT2NFIS to form an Intelligent-BCI (I-BCI) system. The features generated by the RoCSP algorithm are used as input to the SRIT2NFIS classifier. I-BCI system handles non-stationarity in the RoCSP features by modeling it as uncertainty in the classifier. Experimental results show that I-BCI system is able to handle both intra/intersession non-stationarity. The I-BCI system employs a broad frequency range for all the subjects. However, it has been shown in the literature that there is a variation in frequency responses of different subjects. Hence, we improve the I-BCI system for individual subjects. The third contribution is the development of a mechanism for selecting spectral (Spectral-I-BCI (S-IBCI)) and spatio-spectral filters (Spatio-Spectral-I-BCI (SS-IBCI)). S-IBCI system selects spectral filters by eliminating redundant bands from a filter bank in an iterative manner. Further, the SS-IBCI system selects spatial filters from the selected spectral filters. Based on the subject-wise studies, the experimental results indicate a significant performance improvement by both the spectral and spatio-spectral filter selection scheme. Finally, the I-BCI system is realized for control of a quadcopter using brain signals. The quadcopter is equipped with proportional integral derivative controller for motion in x-y plane. Experiments show that the user is able to control the quadcopter successfully. The proposed system improves the accuracy of EEG based BCI highlighting the significance of robust features, handling non-stationarity and providing subject-specific spatio-spectral filters.
DOI: 10.32657/10356/69545
Fulltext Permission: open
Fulltext Availability: With Fulltext
Appears in Collections:SCSE Theses

Files in This Item:
File Description SizeFormat 
main.pdf6.68 MBAdobe PDFThumbnail

Page view(s)

Updated on Oct 20, 2021

Download(s) 50

Updated on Oct 20, 2021

Google ScholarTM




Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.