Multi-source Domain Adaptive Cross-subject EEG Cognitive State Assessment Method Based on Label Alignment

Abstract

The invention discloses a multi-source domain adaptive cross-subject EEG cognitive state evaluation method based on label alignment. The present invention comprises steps: 1: data acquisition; 2: data preprocessing; 3: EEG cognitive state assessment method across subjects based on LA-MSDA model. The present invention uses the shared public feature extractor and non-shared sub-feature extractor in stages to further learn the subject invariant features and specific features of the source domain samples and target domain samples; secondly, considering the relationship and similarity between subjects , and propose methods to align the inter-domain distributions of local and global representations to assess cognitive states across subjects, addressing the difficulty of learning fine-grained class conditional information and adapting to decision boundary samples across subjects. Finally, the present invention effectively avoids the problem of individual differences in EEG signals in the field of brain cognitive computing, is applicable to EEG-based cognitive state recognition under any task, has strong generalization ability, and is well applicable to clinical diagnosis and practical application.

Description

technical field [0001] The invention relates to a neural electrophysiological signal analysis technology in the field of brain cognitive computing, and a multi-source domain adaptation model construction method in the field of unsupervised learning. The method of state assessment can effectively solve the limitations of significant individual differences and low signal-to-noise ratio among different subjects. Background technique [0002] Due to the characteristics of non-invasiveness, portability, and low cost, as well as the advantages of machine learning or deep learning in extracting and classifying features from large amounts of data, EEG-based cognitive state analysis methods have received more and more attention in recent years. s concern. Existing EEG-based analysis usually combines appropriate feature extraction with a classifier to perform classification tasks, among which: common methods for feature extraction include Common Spatial Patterns (CSP), Discrete Wavel...

AI Technical Summary

Problems solved by technology

[0003] Although the signal analysis of these methods has high discriminative performance, there is a certain degree of defect in EEG-based cross-domain predictive analysis: in the real scene of EEG analysis, EEG has significant differences between subjects, which is mainly due to Caused by physical (such as environment and skin electrode impedance) and biological (such as differences in sex, age, and brain activity patterns) factors, in addition, EEG changes over time despite the same subject

However, due to the highly nonlinear and significant individual differences of EEG, it is difficult to extract the same or similar features among different subjects. Therefore, the existing UDA method has the following two limitations: (1) close to the decision boundary The problem of feature confusion cannot be completely solved, the objective function is difficult to achieve optimality, and may fall into a local optimal state; (2) It is difficult to achieve feature-based alignment to extract feature-based domain-invariant features

Images