A user-independent motor imagery classification model training method based on transfer learning

Abstract

A user-independent motor imagery classification model training method based on transfer learning. First, the EEG signal is transformed from a time series signal into a time-frequency domain image signal through a short-time Fourier transformation, and then a perceptual hash algorithm is used. The data of different trials between trials is measured, and the migration weight coefficient is calculated. Then, when training the current subject, the calculated weight is used to weight the data of other subjects to complete the sample migration; Verification; the present invention realizes data multiplexing across subjects under the condition of small samples, thereby improving the generalization ability and accuracy of the current subject classification model.

Description

technical field [0001] The invention belongs to the technical field of biological information, and in particular relates to a method for training a user-independent motor imagery classification model based on migration learning. Background technique [0002] The motor imagination thinking activity is a kind of thinking mode of thinking simulation of motor intention without real motor output, that is, the brain imagines the whole movement process without actual muscle contraction. similarity. Brain-computer interface (BCI) is essentially a human-computer interface system that realizes communication between people and external devices. It does not rely on normal peripheral neuromuscular channels, but uses EEG signals as the carrier of brain intent to interact with the outside world. . [0003] In practical applications, the brain-computer interface method based on motor imagery can be used for robot-assisted operations or intelligent human-computer interaction, which is a ne...

AI Technical Summary

Problems solved by technology

However, since brain-computer interaction tasks have strict requirements on the experimental environment and participants, it is difficult to collect a large amount of effective EEG data in actual operation.

If you want to ensure the accuracy of the classification model, you need a longer experimental period to ensure sufficient data volume, but this will greatly affect the experimental process, which is not conducive to practical applications; and using small samples to train the model independently will easily lead to insufficient data in the model. Problems such as low fitting and robustness greatly affect the classification performance of the model

Images