Deep learning vision classifying method based on electroencephalogram data

Abstract

The invention provides a deep learning vision classifying method based on electroencephalogram data. The deep learning vision classifying method based on electroencephalogram data mainly comprises electroencephalogram data acquisition, electroencephalogram learning, electroencephalogram feature extracting and automatic classifying. The process of the method comprises the following steps: learning a brain activity vision classifying manifold with recognition capability by electroencephalogram data induced by a visual object stimulating factor and a recurrent neural network; then training a regressor based on a convolutional neural network; mapping an image to a learned manifold; and finally, carrying out automated vision classifying task through a computer by features based on human brain to obtain an image classifying result. Compared with a convolutional neural network method, the deep learning vision classifying method based on electroencephalogram data has the advantages that the classifying ability and the generalization ability are competitive; an image sign based on brain in a novel mode is used, and significative insight which is associated with a human vision perceptual system is provided; and images can be effectively projected to a new manifold based on biology, so that the development mode of an object classifier can be changed fundamentally.

Description

technical field [0001] The invention relates to the fields of computer vision, machine learning and cognitive neuroscience, in particular to a deep learning visual classification method based on EEG data. Background technique [0002] With the rapid development of science and technology in the information age, the requirements for computer visual processing capabilities are getting higher and higher. Humans have shown excellent performance in interpreting visual scenes, which is still beyond the reach of machines. Although recently rediscovered convolutional neural networks have led to significant improvements in the performance of automatic visual classification, their generalization capabilities are not at human levels because they learn a discriminative feature space that is strictly dependent on the training techniques employed. dataset rather than a more general one, reflecting the difficulty of cognitive-based automated approaches to perform visual classification tasks...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problem that the cognitive-based automation method is difficult to perform visual classification tasks, the purpose of the present invention is to provide a deep learning visual classification method based on EEG data. First, it uses the combination of EEG data induced by visual object stimuli Recurrent neural network to learn a discriminative brain activity visual classification manifold, then train a convolutional neural network-based regressor, which maps images to the learned manifold, and finally the computer uses features based on the human brain to perform automation Visual classification task, get image classification results

Images