Neural network based EEG denoising.

Area of Science:

• Signal Processing
• Artificial Intelligence
• Biomedical Engineering

Background:

• Additive and multiplicative white noise significantly corrupts signal data.
• Existing noise reduction filters often compromise signal integrity.
• Advanced filtering methods are needed for accurate signal analysis, particularly in noisy environments.

Purpose of the Study:

• To introduce a novel noise reduction filter based on a back propagation neural network (BPNN) ensemble.
• To evaluate the filter's efficacy in reducing both additive and multiplicative white noise.
• To assess the filter's ability to preserve essential signal characteristics during noise reduction.

Main Methods:

• Developed a neural network (NN) ensemble filter using back propagation.
• Trained the NN ensemble with identical noisy and reference signals.
• Compared performance against improved epsilon nonlinear filters and single NN filters.

Main Results:

• The NN ensemble filter demonstrated superior noise reduction capabilities compared to other methods.
• Performance advantages were particularly evident at lower signal-to-noise ratios.
• The filter successfully preserved critical signal characteristics post-processing.

Conclusions:

• The proposed NN ensemble filter offers a robust solution for noise reduction in various signal types.
• It provides significant improvements over existing filtering techniques, especially under challenging noise conditions.
• Validated effectiveness through simulations and real-world electroencephalogram (EEG) signal processing.