Emotion electroencephalogram classification method based on CNN

Abstract

The invention discloses an emotion electroencephalogram classification method based on a CNN. According to the invention, a Russell emotion dimension model in a continuous emotion dimension model is used as a reference; a DEAP data set is used as a sample; baseline removal is carried out on the emotion electroencephalogram; data normalization is carried out; Pearson coefficients of three frequencybands of electroencephalogram are extracted; the Pearson coefficients are converted into a 2D picture format, valuable experiments for emotion electroencephalogram classification are screened throughSBS by taking experiments as units, and the screened experimental data are input into CWAGAN-GP for data enhancement so as to supplement a training set; and the data are input into an integrated convolutional neural network in a frame form. According to the method, the emotion electroencephalogram signals can be effectively classified, considerable classification precision is provided, and the operation of inputting the emotion electroencephalogram signals into the integrated convolutional neural network in a frame form can effectively prevent the convolutional neural network from over-fitting.

Description

technical field [0001] The present invention relates to an emotional EEG classification method, in particular to a method for classifying by using CNN after optimizing the preprocessing of emotional EEG. Background technique [0002] Convolutional neural network (CNN) is a feed-forward neural network, which has excellent performance in large-scale image processing, and has been widely used in image classification, positioning and other fields. Compared with other neural network structures, convolutional neural networks require relatively few parameters, making them widely applicable. [0003] The generative confrontation network (GAN) produces quite good output through the mutual game learning of two modules in the framework: the generative model and the discriminative model, and can generate new data that is fake and real based on the original data set. WGAN-GP is proposed in response to the existing problems of WGAN. In the actual experimental process, WGAN still has the ...

AI Technical Summary

Problems solved by technology

WGAN-GP is proposed for the existing problems of WGAN. WGAN still has the problems of difficult training and slow convergence speed in the real experimental process. Compared with traditional GAN, the experimental improvement is not obvious.

The problem with WGAN is that the shear weight is directly used when dealing with continuity constraints, that is, every time the parameters of the discriminator are updated, it is checked whether the absolute value of all the parameters of the discriminator exceeds a threshold, such as 0.01, If so, return these parameters to the range [-0.01,0.01]

In this case, the optimal strategy is to make all parameters as extreme as possible, either take the maximum value (such as 0.01) or take the minimum value (such as -0.01), for the deep neural network, it cannot give full play to the simulation of the deep neural network. In addition, it is also found that the forced shear weight can easily lead to gradient disappearance or gradient explosion. The reason for gradient disappearance and gradient explosion is the selection of the shear range. If the selection is too small, the gradient will disappear. If it is set a little larger, After each layer of the network, the gradient becomes a little bit larger, and a gradient explosion will occur after multiple layers

Images