Artifact Removal in EEG Signals

Automated artifact removal for multi-channel EEG signals.

year: Feb 2024 — Sep 2025
role: Bachelor Thesis
location: Amrita Vishwa Vidyapeetham
category: Academic

[ hardware & pipeline ]

8-channel Neuphony dry-sensor EEG cap — fig.01 // 8-channel neuphony eeg cap

Comparison of three artifact removal methodologies — fig.02 // three-method comparison — manual ica // wavelet denoising // ml-assisted detection

Context

Engineered a fully automated, hybrid machine learning framework to isolate and eliminate disruptive ocular eye-blink artifacts from multi-channel EEG streams without manual intervention.

Pipeline

Managed the entire data acquisition pipeline, streaming raw neuro signals at 250 Hz from custom participant trials using an 8-channel Neuphony dry-sensor EEG cap.

Design

Implemented a 4th-order Butterworth bandpass filter (0.5–40 Hz) paired with Independent Component Analysis to decompose complex scalp mixtures into localized independent source components.

Automation

Trained an RBF-kernel SVM classifier on a 10-dimensional statistical feature matrix to automatically identify artifact components, selectively applying Daubechies-4 (db4) wavelet soft-thresholding to preserve clean neural data.

Performance

Validated signal integrity over alternative methods using Higuchi Fractal Dimension (HFD), Shannon Entropy, and Power Spectral Density (PSD) metrics to confirm zero over-filtering or signal distortion.

Signal ProcessingIndependent Component Analysis (ICA)Support Vector Machines (SVM)Real-Time PipelinesPython

[MILESTONE]

Peer-reviewed, accepted, and personally presented this standalone architecture at the IEEE CONNECT 2025 international conference in Bangalore, India.

View published paper

UAV Autopilot Architecture — Avionics System Design

open →