EUVIMEDEUVIMED entdecken
Crossref

Motor Imagery Brain-Computer Interface Based on Transfer Learning with Application in Exoskeleton Robot System

Zitong Wan

University of Liverpool · 01.01.2027

DOI: 10.17638/03187996 ↗

Kurz gefasst

Fachartikel in der Zeitschrift University of Liverpool, erschienen am 01.01.2027. Verfasst von Zitong Wan.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Brain-computer interfaces (BCI) recognize human intentions via brain neural activities, facilitating direct communication between brains and external devices. In different BCI paradigms, motor imagery-based BCI (MI-BCI) has been investigated in various scenarios, such as neurological rehabilitation and intelligent device control. MI refers to imaging specific body movements without real bodily activities, which can be recorded by electroencephalogram (EEG). MI is one of the most potential EEG paradigms due to its spontaneity independent of external stimulations and the similar neural mechanisms to real movements. However, MI signals have the characteristics of non-stationarity, significant individual differences, and training data deficiency, leading to low decoding accuracy with traditional analysis methods. Transfer learning is employed to increase the accuracy and robustness of MI decoding models by applying the knowledge learned in one domain to a different but related domain, which has become the main idea in MI signal analysis. However, there is still insufficient research and discussions on cross-subject data fading, single-to-single transfer, and negative transfer currently. This PhD project explored these three existing MI decoding problems in transfer learning, aiming to improve the MI signal decoding accuracy in various transfer scenarios and realize a prototype of an MI-BCI-based exoskeleton. This thesis uses the following four studies to solve the existing problems in MI analysis methods based on transfer learning. First, this study proposed a novel classification method based on manifold learning and adaptive neighborhood selection for the data fading problem. Second, the data fading problem was further discussed with cross-subject scenarios, and a cross-subject fading data classification approach with segment alignment was proposed for this study. Third, the negative transfer problem in single-to-single transfer was discussed with a multimodal alignment-based deep learning model. Last, an MI-BCI-based lower-limb exoskeleton robot system based on multiple segments decoding was constructed, which was conducted online using the method of segment alignment. For the existing shortcomings of MI signal analysis in transfer learning, this project investigates three existing problems for the first time, which have positive effects on the application of transfer learning in MI signal analysis to fill the research gaps in this field. The three proposed original algorithm frameworks significantly expand the decoding approaches of MI signals under different transfer learning scenarios, making a major contribution to the further development of MI recognition in BCI. In addition, the research results provide both theoretical and technical support for improving the reliable MI-BCI-based exoskeleton robot system, reflecting important practical reference values for research and application with other intelligent systems controlled by MI-BCI.

Schlagwörter

motor imagerybrain-computer interfacetransfer learningEEG decodingrobotic exoskeletonnegative transfermanifold learningadaptive neighborhood selectionsegment alignmentmultimodal alignment

Zitieren

APA

Wan, Z. (2027). Motor Imagery Brain-Computer Interface Based on Transfer Learning with Application in Exoskeleton Robot System. University of Liverpool. https://doi.org/10.17638/03187996

Vancouver

Wan Z. Motor Imagery Brain-Computer Interface Based on Transfer Learning with Application in Exoskeleton Robot System. University of Liverpool. 2027. doi:10.17638/03187996

BibTeX
@article{Wan2027,
  author = {Wan, Zitong},
  title = {Motor Imagery Brain-Computer Interface Based on Transfer Learning with Application in Exoskeleton Robot System},
  journal = {University of Liverpool},
  year = {2027},
  doi = {10.17638/03187996}
}
Kostenlos starten →