Eye blink artefact removal framework for EEG signals using DWT and autoencoder

Electroencephalography (EEG) is extensively used in brain-machine interface (BMI) systems because of its compact size and high temporal resolution. However, artefacts such as eye blinks distort the underlying brain activity, requiring effective artifact removal for accurate analysis. This paper presents an automatic method for removing eye-blink artifacts using a combination of discrete wavelet transform (DWT) and an autoencoder. The contaminated EEG signal is first preprocessed with DWT to generate clean signal, which serves as a reference during training. The autoencoder is trained by comparing its reconstructed output with the clean signal. A novel loss function combining correlation coefficient (CC) and mean squared error (MSE) minimizes reconstruction error while preserving cerebral information. This method achieves an average signal to artifact ratio (SAR) of 3.21 and CC of 0.88, demonstrating superior artifact removal and signal preservation, while maintaining computational efficiency with an average processing time of 9 ms across 10 runs.