Motor imagery electroencephalogram recognition method based on relative wavelet packet entropy brain network and improved lasso

Abstract

The invention discloses a motor imagery electroencephalogram recognition method based on a relative wavelet packet entropy brain network and improved lasso, which comprises the following steps: calculating an R2map according to power spectral density to obtain a maximum frequency band and performing band-pass filtering; extracting and calculating detail coefficients and approximation systems of the electroencephalogram signals through a wavelet packet method to obtain wavelet packet energy entropy features, constructing a brain function network through wavelet packet energy entropy values, and extracting topological features of the brain network. Variance features are obtained according to an SCSP algorithm in data preprocessing; The three features are fused to obtain a feature matrix with a relatively high dimension. The method comprises the following steps: carrying out feature selection through a mutual information and correlation Lasso method in combination with a Relief-f algorithm, and screening out a feature matrix with a smaller dimension. According to the method, time-space domain features are extracted, topological features of the brain network are extracted together, and more electroencephalogram feature information is reserved; and feature screening is carried out by combining a mutual information and correlation Lasso method and a Relief-f algorithm, so that the selected features are more excellent.

Description

technical field [0001] The invention relates to the field of EEG signals, in particular to a motor imagery EEG recognition method based on a relative wavelet packet entropy brain network and an improved version of lasso. Background technique [0002] As a new interactive technology, Brain Computer Interface (BCI) aims to establish a connection between the human brain and computers, and to combine the biomedical and computer fields. As a hot area of ​​BCI, MI has been widely used in medical rehabilitation. Although the field of motor imagery has great value at this stage, there are still some challenges in this research, such as insufficient data, poor classification effect, etc. [0003] MI is mainly used to collect EEG signals of subjects imagining limb movements. Machine learning is used to classify the collected signals and feed back the classification results to external devices to help subjects perform limb movements, so as to achieve the purpose of helping physically...

AI Technical Summary

Problems solved by technology

Javier proposed an improved shrinkage covariance matrix in 2019 to better handle small sample data, and finally processed the above data set through the Riemann Minimum Mean Distance (RMDM) classifier with an average classification accuracy of 79.6% [ 7]. However, the survey found that there are still some problems in the recognition of motor imagery EEG, including the low accuracy of feature classification, usually only about 80%

Images