An improved inception network-based motor imagery electroencephalogram signal classification method

By incorporating residual structures and SE attention mechanisms into the Inception network, combined with the Selu activation function and composite loss function, the problems of gradient vanishing and overfitting were solved, achieving higher accuracy in classifying motor imagery EEG signals.

CN115630314BActive Publication Date: 2026-07-21HARBIN UNIV OF SCI & TECH
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
HARBIN UNIV OF SCI & TECH
Filing Date
2022-09-30
Publication Date
2026-07-21

AI Technical Summary

Technical Problem

Existing methods for classifying motor imagery EEG signals suffer from gradient explosion and gradient vanishing problems. Furthermore, due to the small amount of data, they are prone to overfitting. Single-scale convolutional neural networks do not extract sufficient features, which affects classification accuracy.

Method used

A residual structure and SE attention mechanism module were added to the Inception network, and the Selu activation function was used. Five parallel improved Inception networks were combined for feature extraction. The model was trained using a composite loss function that combined cross-entropy and Dice loss, and the Adam optimization algorithm was used to improve the classification accuracy of the model.

Benefits of technology

By improving the Inception network structure, the problems of gradient vanishing and overfitting are solved, the sufficiency of feature extraction and classification accuracy are improved, and better classification results are achieved than traditional methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115630314B_ABST
    Figure CN115630314B_ABST
Patent Text Reader

Abstract

The application provides a motor imagery electroencephalogram signal classification method based on an improved Inception network, and the scheme is as follows: (1) performing data enhancement processing on a motor imagery electroencephalogram signal data set; (2) introducing an attention mechanism module in the Inception network, assigning weights to the extracted channel features through the attention mechanism module, and introducing a residual to reduce overfitting and network degradation problems caused by small data volume; (3) sending the output of the improved Inception network into a residual network to select effective features; (4) adopting a compound loss function combining cross-entropy loss and Dice loss; and (5) training and testing the improved Inception network. The application has good classification effect on the disclosed motor imagery electroencephalogram signal data set, improves the classification accuracy, does not need to manually extract features, and has high application value.
Need to check novelty before this filing date? Find Prior Art