Riemannian feature migration-based magnetoencephalogram signal classification method and device and medium

Abstract

The invention discloses a Riemannian feature migration-based magnetoencephalogram signal classification method and device and a medium, and the method comprises the following steps: obtaining magnetoencephalogram signals, and carrying out the filtering of the magnetoencephalogram signals through a band-pass filter and a spatial filter, and obtaining filtered signals; constructing Riemannian features according to the sample covariance matrix of the filtering signal; introducing a discrimination subspace alignment method according to the tag information of the existing subjects, and aligning Riemannian features of the existing subjects and the target subject; and training a classifier according to the existing subject features after feature alignment and the labels, and predicting the category of the magnetoencephalogram signal of the target subject by adopting the trained classifier. According to the method, the feature distribution difference between the source domain and the target domain is reduced through the discriminant subspace alignment algorithm, the quality of model training features is improved, compared with a traditional cross-subject transfer learning method, the prediction accuracy can be better improved, and the method and device can be widely applied to the field of brain-computer interfaces.

Description

technical field [0001] The present invention relates to the field of brain-computer interface, in particular to a magnetoencephalogram signal classification method, device and medium based on Riemann feature transfer. Background technique [0002] Magnetoencephalography is a functional neuroimaging technique for recording brain activity information in the field of brain-computer interface, which has the advantages of high temporal and spatial resolution. The process of collecting magnetoencephalographic signals is usually time-consuming and expensive. Using the labeled data of previous subjects can reduce the dependence on new subject data and the training time of classification models. Due to the obvious individual differences and non-stationarity of magnetoencephalogram signals, there is a large difference between the data distribution of previous subjects and the data distribution of target subjects, resulting in a lengthy calibration process for training models and a dec...

AI Technical Summary

Problems solved by technology

Migration learning can help the system complete feature migration and model calibration across subjects, but the existing migration learning algorithms lack the use of label information and often fail to achieve optimal results

Images