Brain Specific Visual Cognitive State Judgment Method Based on Sparse Nonnegative Tensor Decomposition

Abstract

The invention discloses a method for determining a specific visual cognition state of brain based on sparse nonnegative tensor factorization (SNTF). The method is innovatively characterized by comprising the following steps of taking functional magnetic resonance imaging (fMRI) cognitive data as a large tensor by an SNTF algorithm, and establishing a high-order nonnegative tensor model from the tensor level; then, performing feature dimension reduction on the fMRI cognitive data in each dimensionality, thereby obtaining a sparse nonnegative feature tensor with low dimensionality; and finally, effectively determining the specific visual cognition state of brain by combining characteristics of a support vector machine. According to the method provided by the invention, dimension reduction and feature extraction are performed by the SNTF, so that potential structural information in the original data is extracted in a multidirectional multi-angle mode; due to L1-norm regularization and nonnegative constraint, the extracted correlative components are sparse and conform to intuitive experience of the brain perception, then the characteristics of the support vector machine are combined, and thus the accuracy of classification and discrimination is improved.

Description

technical field [0001] The invention belongs to the field of biological feature extraction and brain cognitive state discrimination, and relates to task-related functional magnetic resonance imaging (functional magnetic resonance images, namely fMRI) preprocessing and sparse nonnegative tensor factorization (sparse nonnegative tensor factorization, namely SNTF) Feature extraction and discriminant classification of support vector machine (SVM), specifically related to a brain-specific visual cognitive state judgment method based on sparse non-negative tensor decomposition, which can be used for biological feature extraction, dimension reduction, field of pattern recognition. Background technique [0002] In recent years, with the development of functional Magnetic Resonance Imaging (fMRI) technology, people's ability to conduct brain research has been greatly enhanced, and at the same time a large amount of data has been generated. The storage, query and calculation of the d...

AI Technical Summary

Problems solved by technology

[0004] Correlation analysis uses prior knowledge of the experimental task to determine the activation regions and their intensities. This method is disadvantageous if the model of the experimental fMRI signal changes is unknown or not constant, for example, the subject is a brain disease. patients or were performing a complex learning task, the technique was unable to detect activation regions

But this method has some limitations:

However, in the case of multitasking, we cannot assume that the brain regions activated by different tasks are spatially independent, so the use of ICA methods for functional localization is bound to be limited.

[0008] (2) The independent components decomposed by the ICA method often have negative components, that is, it does not necessarily guarantee that the decomposed components are non-negative

[0010] However, in modern scientific technology and engineering problems, a large number of large-scale data processing problems are often encountered. When dealing with large-scale high-dimensional data, non-negative matrix factorization only vectorizes the image data and then according to the eigenvalues ​​and features The vector is used for feature extraction, while ignoring the factor that the processing object is often multi-dimensional data, resulting in unsatisfactory dimension reduction effect caused by projection in only one direction and destroying the structure and correlation of the original image, and the original image cannot be completely maintained. Issues such as redundancy and structure in

Images