Electroencephalogram signal adversarial domain adaptation method based on optimized spatial-temporal feature extraction

• Main Authors: CHEN TIE, SHE QINGSHAN, MA YULIANG, SUN MINGXU
• Format: Patent
• Language: Chinese; English
• Published: 31.01.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; DIAGNOSIS; ELECTRIC DIGITAL DATA PROCESSING; HUMAN NECESSITIES; HYGIENE; IDENTIFICATION; MEDICAL OR VETERINARY SCIENCE; PHYSICS; SURGERY

The invention relates to an electroencephalogram signal adversarial domain adaptation method based on optimized spatial-temporal feature extraction. The method comprises the following steps: firstly, carrying out band-pass filtering preprocessing, normalization and other operations on collected motor imagery electroencephalogram signals; extracting features of the motor imagery electroencephalogram signals in a spatial dimension and a time dimension by adopting a feature extractor comprising a convolution attention module integrating a channel and a space, a variance time sequence layer and other modules, and aligning data distribution of a source domain subject and a target domain subject by applying an adversarial domain adaptation method based on a Wasserstein distance; according to the method, a solution parameter is optimized through a gradient descent method, so that the Wasserstein distance is reduced, the difference of data distribution among different subjects is reduced, a classifier trained by usin