NATIONAL INSTITUTE OF TECHNOLOGY,SILCHAR
An improved feature extraction algorithms of EEG
signals
DEPARTMENT OF ELECTORNICS & COMMUNICATION
BY
Anirban Nath (2424223)
M.Tech (2nd Sem)
A Presentation on
Guided By – Dr. Koushik Guha
Contents
• Introduction to EEG Signal
• Introduction to BCI
• Problem Statement
• Methodology
• Result and Discussion
• Future Work
• Conclusion
• References
• EEG = Electroencephalogram
• Records electrical activity of the brain using scalp
electrodes.
• Reflects real-time brain function and neural
communication.
Introduction to EEG
Signals
• Low amplitude: 10–100 μV
• Frequency range: 0.5 – 100 Hz
• Non-stationary and time-varying
• High temporal resolution (captures millisecond-level activity)
Key Characteristics:
Introduction to BCI
• BCI = Brain-Computer Interface
• Connects the human brain directly to external
devices
• Converts brain signals into commands
• Helps users control devices using thoughts only
• Useful for patients with motor disabilities
Introduction to EEG-based
BCI
Key Characteristics of
EEG Signals
• EEG signals: weak, non-linear,
non-stationary
• BCI links brain to external
devices
• Noise sources: EOG, ECG
interfere
• Feature extraction crucial for
accuracy
Importance of Feature
Extraction
Noise heavily impacts the accuracy
of EEG signals affecting analysis.
Successful BCI depends on robust
extraction of different features.
Problem Statement
Conventional
denoising limitations
Fail to separate low-
amplitude useful signals
Signal and noise
overlap
Complicates feature
extraction
Methodology
Overview 1
ICA: Separate
signal components
2 WT: Reduce noise
via thresholding
3
CSP: Enhance
discriminative
spatial features
4 ML algorithms for
final classification
Independent Component Analysis
(ICA)
Purpose
Separate mixed brain signals
Artifact removal
Eliminate EOG & ECG interference
ICA Mathematical
Model
• Observed signal: X(t) = A · S(t)
• Output: Y(t) = W · X(t)
Where,
X(t): Observed Signal
A:Mixing Coefficient Matrix
S(t):Source Signal
• S(t)= A-1 . X= WX
Fig: The Mathemetical model of ICA .[1]
Wavelet Transform (WT) in
EEG
Purpose
Denoising EEG signals
Technique
Decompose signals in
frequency domain
Thresholding
Remove noise components effectively
Procedure
1.Wavelet
Transform
• EEG signals converted to Wavelet
Coefficient .
• High Frq. Coefficient ~ Noise
• Low Frequency Coeff. ~Real Signal
2.Thresholding the Coefficient
• Remove or shrink the High wavelet
coefficient (assumed to be noise)
3.Inverse Wavelet Transform
• Modified wavelet coefficient are
transform back to get clean signal
Common Spatial Pattern
(CSP)
• Feature Extraction algo
• Used to distinguish between two mental task.
• Example -Imagining Left Hand Vs Right Hand Movement
Proposed Workflow
Step 1: ICA
decomposition
Step 2: WT on ICA
components
Step 3: Wavelet threshold denoising
Step 4: CSP feature
extraction
Step 5:
Classification
Fig 2: The process of the propsed method. [1]
Dataset Description
• EEG data collected from 109 participants
• Recordings captured from 64 channels using the BCI2000 system
• Included tasks: motor imagery of hands and feet
• Approximately 1,500 total recordings
Fig 3: Electrodes arrangement position.[1]
ICA Results
Key Benefit
ICA separates signals effectively
Artifact Control
Reduces EOG and ECG artifacts
Improves SNR
WT Results
Retains low-amplitude
neural signals
Removes high-
frequency noise
CSP Feature
Extraction
1
Input
ICA-WT processed signals
2
Output
Features has been extracted
Classification
Algorithms
Bagging Tree (BT)
Fig 4: Bagging Tree
Results Summary
• ICA-WT-CSP approach:
⚬Highest classification accuracy
⚬Robust to noise
⚬Preserves useful motor features
Future Work
1
Improve recognition
accuracy
2 Explore deep learning
classifiers
3
Extend to real-time BCI
applications
Conclusion
Proposed
method is
effective
Combines
strengths of
ICA, WT, and
CSP
Enhances
classification of
motor imagery
EEG
References
• [1]Xiaozhong Geng, Dezhi Li, Hanlin Chen, Ping Yu, Hui Yan, Mengzhe Yue, An improved feature extraction
algorithms of EEG signals based on motor imagery brain-computer interface,Alexandria Engineering
Journal,Volume 61, Issue 6,2022,Pages 4807-4820,ISSN 1110-0168, https://doi.org/10.1016/j.aej.2021.10.034.
• [2]International Journal for Modern Trends in Science and Technology Volume 9, Issue 08, pages 45-50 ISSN:
2455-3778 online, DOI: https://doi.org/10.46501/IJMTST0908008
• [3] A. Mohammad, F. Siddiqui and M. Afshar Alam, "Feature Extraction from EEG Signals: A deep learning
perspective," 2021 11th International Conference on Cloud Computing, Data Science & Engineering
(Confluence), Noida, India, 2021, pp. 757-760, doi: 10.1109/Confluence51648.2021.9377108.
• [4] ] Z. Ling, C. Shuyue, S. Yuqiang, M.a. Zhe, Extraction of Evoked Related Potentials by using the Combination
of Independent Component Analysis and Wavelet Analysis, J. Biomed. Eng. 27 (4) (2010) 741–745.
• [5] International Conference on Computational Intelligence and Networks. IEEE, (2016) 84-89
Thank You
Any Questions?

An improved feature extraction algorithms of EEG signals

  • 1.
    NATIONAL INSTITUTE OFTECHNOLOGY,SILCHAR An improved feature extraction algorithms of EEG signals DEPARTMENT OF ELECTORNICS & COMMUNICATION BY Anirban Nath (2424223) M.Tech (2nd Sem) A Presentation on Guided By – Dr. Koushik Guha
  • 2.
    Contents • Introduction toEEG Signal • Introduction to BCI • Problem Statement • Methodology • Result and Discussion • Future Work • Conclusion • References
  • 3.
    • EEG =Electroencephalogram • Records electrical activity of the brain using scalp electrodes. • Reflects real-time brain function and neural communication. Introduction to EEG Signals
  • 4.
    • Low amplitude:10–100 μV • Frequency range: 0.5 – 100 Hz • Non-stationary and time-varying • High temporal resolution (captures millisecond-level activity) Key Characteristics:
  • 6.
    Introduction to BCI •BCI = Brain-Computer Interface • Connects the human brain directly to external devices • Converts brain signals into commands • Helps users control devices using thoughts only • Useful for patients with motor disabilities
  • 7.
    Introduction to EEG-based BCI KeyCharacteristics of EEG Signals • EEG signals: weak, non-linear, non-stationary • BCI links brain to external devices • Noise sources: EOG, ECG interfere • Feature extraction crucial for accuracy Importance of Feature Extraction Noise heavily impacts the accuracy of EEG signals affecting analysis. Successful BCI depends on robust extraction of different features.
  • 8.
    Problem Statement Conventional denoising limitations Failto separate low- amplitude useful signals Signal and noise overlap Complicates feature extraction
  • 9.
    Methodology Overview 1 ICA: Separate signalcomponents 2 WT: Reduce noise via thresholding 3 CSP: Enhance discriminative spatial features 4 ML algorithms for final classification
  • 10.
    Independent Component Analysis (ICA) Purpose Separatemixed brain signals Artifact removal Eliminate EOG & ECG interference
  • 11.
    ICA Mathematical Model • Observedsignal: X(t) = A · S(t) • Output: Y(t) = W · X(t) Where, X(t): Observed Signal A:Mixing Coefficient Matrix S(t):Source Signal • S(t)= A-1 . X= WX Fig: The Mathemetical model of ICA .[1]
  • 12.
    Wavelet Transform (WT)in EEG Purpose Denoising EEG signals Technique Decompose signals in frequency domain Thresholding Remove noise components effectively
  • 13.
    Procedure 1.Wavelet Transform • EEG signalsconverted to Wavelet Coefficient . • High Frq. Coefficient ~ Noise • Low Frequency Coeff. ~Real Signal 2.Thresholding the Coefficient • Remove or shrink the High wavelet coefficient (assumed to be noise) 3.Inverse Wavelet Transform • Modified wavelet coefficient are transform back to get clean signal
  • 14.
    Common Spatial Pattern (CSP) •Feature Extraction algo • Used to distinguish between two mental task. • Example -Imagining Left Hand Vs Right Hand Movement
  • 15.
    Proposed Workflow Step 1:ICA decomposition Step 2: WT on ICA components Step 3: Wavelet threshold denoising Step 4: CSP feature extraction Step 5: Classification Fig 2: The process of the propsed method. [1]
  • 16.
    Dataset Description • EEGdata collected from 109 participants • Recordings captured from 64 channels using the BCI2000 system • Included tasks: motor imagery of hands and feet • Approximately 1,500 total recordings Fig 3: Electrodes arrangement position.[1]
  • 17.
    ICA Results Key Benefit ICAseparates signals effectively Artifact Control Reduces EOG and ECG artifacts Improves SNR
  • 18.
    WT Results Retains low-amplitude neuralsignals Removes high- frequency noise
  • 19.
    CSP Feature Extraction 1 Input ICA-WT processedsignals 2 Output Features has been extracted
  • 20.
  • 21.
  • 22.
    Results Summary • ICA-WT-CSPapproach: ⚬Highest classification accuracy ⚬Robust to noise ⚬Preserves useful motor features
  • 23.
    Future Work 1 Improve recognition accuracy 2Explore deep learning classifiers 3 Extend to real-time BCI applications
  • 24.
    Conclusion Proposed method is effective Combines strengths of ICA,WT, and CSP Enhances classification of motor imagery EEG
  • 25.
    References • [1]Xiaozhong Geng,Dezhi Li, Hanlin Chen, Ping Yu, Hui Yan, Mengzhe Yue, An improved feature extraction algorithms of EEG signals based on motor imagery brain-computer interface,Alexandria Engineering Journal,Volume 61, Issue 6,2022,Pages 4807-4820,ISSN 1110-0168, https://doi.org/10.1016/j.aej.2021.10.034. • [2]International Journal for Modern Trends in Science and Technology Volume 9, Issue 08, pages 45-50 ISSN: 2455-3778 online, DOI: https://doi.org/10.46501/IJMTST0908008 • [3] A. Mohammad, F. Siddiqui and M. Afshar Alam, "Feature Extraction from EEG Signals: A deep learning perspective," 2021 11th International Conference on Cloud Computing, Data Science & Engineering (Confluence), Noida, India, 2021, pp. 757-760, doi: 10.1109/Confluence51648.2021.9377108. • [4] ] Z. Ling, C. Shuyue, S. Yuqiang, M.a. Zhe, Extraction of Evoked Related Potentials by using the Combination of Independent Component Analysis and Wavelet Analysis, J. Biomed. Eng. 27 (4) (2010) 741–745. • [5] International Conference on Computational Intelligence and Networks. IEEE, (2016) 84-89
  • 26.