Intra-class distance and inter-class distance combined transfer learning method based on motor imagery classification

Abstract

The invention discloses an intra-class distance and inter-class distance combined transfer learning method based on motor imagery classification, which comprises the following steps of: 1) taking cross-session data of other subjects except a current subject as a training set and taking the cross-session data of the current subject as a test set; 2) extracting a target signal by using a Butterworthband-pass filter; 3) extracting features by using a CSP (common spatial pattern); 4) performing feature migration in combination with distribution self-adaptation, intra-class distance and inter-class distance; 5) performing classification by using an ensemble learning method. According to the method, the distribution self-adaption, the intra-class distance and the inter-class distance are combined to perform feature migration, the ensemble learning method is used for classification, and the problem that the classification accuracy is not high due to the fact that the motor imagery data is scarce is effectively solved.

Description

technical field [0001] The invention relates to the technical field of motor imagery classification and transfer learning, in particular to a transfer learning method combining intra-class distance and class distance based on motor imagery classification. Background technique [0002] Existing research results have shown that the use of a BCI (brain-computer interface) system must have a sufficient amount of available training samples in order to learn an efficient feature extraction and classification model, and the acquisition of a large number of training samples will inevitably lead to The long data collection process and resource consumption will seriously affect the convenience of the system. Therefore, when the current subject has only a small number or even no training set, it is of great significance to use the data of other subjects for feature extraction and classification, and this is the problem that transfer learning aims to solve. [0003] CSP (Common Spatial...

AI Technical Summary

Problems solved by technology

When the number of samples is relatively small, the estimated inaccurate covariance matrix cannot well represent the basic neurophysiological process, which will directly affect the accuracy of the spatial domain filter design

Furthermore, the design of subsequent classifiers will also directly affect the classification results.

Images