Electroencephalogram denoising method

Abstract

The electroencephalogram denoising method comprises the following steps: S1, decomposing an original electroencephalogram signal into different sub-components through singular spectrum analysis, setting an autocorrelation coefficient threshold to extract sub-components related to main components of the electroencephalogram signal, and reconstructing the sub-components to obtain a corresponding time sequence to form the main components of the electroencephalogram signal; s2, subtracting the main components of the electroencephalogram signals from the original electroencephalogram signals to obtain residual electroencephalogram components; and S3, inputting the residual electroencephalogram components into a deep convolutional neural network to extract detail features, and combining the output of the deep convolutional neural network with the electroencephalogram signal main components to reconstruct the electroencephalogram signals after noise removal. The electroencephalogram denoising method can well remove myoelectricity noise in electroencephalogram signals, has better denoising performance when the signal-to-noise ratio is larger than-1dB compared with a traditional denoising method, and has smaller distortion compared with a single path.

Description

technical field [0001] The invention relates to the field of EEG signal processing, in particular to an EEG denoising method. Background technique [0002] Electroencephalography (Electroencephalography, EEG) is a sophisticated electronic instrument that records spontaneous biopotentials of the brain. It is widely used in brain science research, clinical diagnosis of neurological diseases, and brain-computer interface and other fields. However, EEG is often inevitably mixed with various other types of physiological electrical signals during the acquisition process, such as electro-oculogram, electrocardiogram, and myoelectricity. In most application scenarios such as EEG, these physiological electrical signals will interfere with the analysis or processing of EEG signals, so they are considered as noise [1]. Compared with the influence of eye electricity, electrocardiogram or emotional disturbance on EEG, electromyography (EMG) disturbance is difficult to detect and separat...

AI Technical Summary

Problems solved by technology

Although the deep network can partially obtain the EEG information under extremely high pollution, the results show that the true value of the EEG morphology will be more distorted than the singular spectrum analysis in the case of a higher signal-to-noise ratio (SNR).

Images