CN112790774A

CN112790774A - A primitive EEG deep learning classification method and application

Info

Publication number: CN112790774A
Application number: CN202110182451.1A
Authority: CN
Inventors: 车艳秋; 郑海令; 刘靖; 秦迎梅; 韩春晓; 王若凡
Original assignee: Tianjin University of Technology and Education China Vocational Training Instructor Training Center
Current assignee: Tianjin University of Technology and Education China Vocational Training Instructor Training Center
Priority date: 2020-11-10
Filing date: 2021-02-07
Publication date: 2021-05-14
Also published as: CN112244875A

Abstract

The invention discloses an original electroencephalogram deep learning classification method which comprises the steps of extracting a test sample, converting the obtained test sample into a test input image, inputting the obtained test input image into a deep learning model, extracting robustness characteristics to obtain robustness characteristics, and inputting the extracted robustness characteristics into an extreme learning machine classifier for classification analysis. The deep learning model is composed of a convolution neural network and a long-term and short-term memory neural network. The convolutional neural network can utilize the convolutional layer to extract the characteristics, so that proper characteristics do not need to be searched repeatedly, and a large amount of time is saved. Meanwhile, the feature extraction is carried out by utilizing the convolution layer, the accuracy of the electroencephalogram analysis can be improved by changing the size of the convolution kernel, the operation is simple and convenient, the speed is higher, and the accuracy can reach more than 90%.

Description

Original electroencephalogram deep learning classification method and application

Technical Field

The invention relates to the technical field of schizophrenia detection, in particular to a deep learning classification method and application of original brain electricity.

Background

The brain electrical signals are the spontaneous and rhythmic electrical activity of the brain cell population recorded by the electrodes. The electroencephalogram signals contain rich brain activity information, so the electroencephalogram signals are widely applied to electroencephalogram analysis research.

Most of the current electroencephalogram analysis depends on the comparison between normal electroencephalograms and abnormal electroencephalograms or on the experience of an analyst, a specific and reasonable analysis model is not available, subjective influence factors are more, and the accuracy is lower.

Disclosure of Invention

The invention aims to provide an original electroencephalogram deep learning classification method aiming at the technical defect of low accuracy of the original electroencephalogram deep learning classification method in the prior art.

The technical scheme adopted for realizing the purpose of the invention is as follows:

a method for deep learning and classifying original brain electricity comprises the following steps:

step A: extracting an original electroencephalogram time sequence signal of a person to be tested, and carrying out noise filtering processing on the original electroencephalogram time sequence signal, wherein the electroencephalogram time sequence signal of the person to be tested after the noise filtering processing is used as a test sample;

and B: converting the test sample obtained in the step A into a test input image;

and C: b, inputting the test input image obtained in the step B into a deep learning model, and extracting robust features in the deep learning model to obtain robust features;

step D: and C, inputting the robustness features extracted in the step C into an Extreme Learning Machine (ELM) classifier for classification. Reference documents: [1] li, S.Song, and Y.Wan, "Laplacian twist extension learning machine for semi-collaborative classification", neuro-typing, vol.321, pp.17-27,2018.

[2]Y.Yu and Z.Sun,“Sparse coding extreme learning machine for classification”,Neurocomputing,vol.261,pp.50-56,2017.

In the above technical solution, the test sample conversion method in step B is based on a short-time fourier transform algorithm.

In the above technical solution, the short-time fourier transform algorithm is completed by the following formula:

where z (u) is the electroencephalogram time-series signal at time u, g (u-t) is a window function, f is the electroencephalogram time-series signal sampling rate, j is a complex number, j is 1 × i, and i is an imaginary unit.

In the above technical solution, when the obtained test sample is converted into a test input image, a window of 3 seconds is used to perform short-time fourier transform.

In the above technical solution, in step C, the deep learning model is composed of a Convolutional Neural Network (CNN) and a long-short term memory neural network (LSTM). CNN and LSTM are stacked, and the characteristic matrix extracted by CNN is integrated into the matrix which can be input by LSTM. Reference [3] Saintath T N, Vinyals O, Senior A, et al. volumetric, Long Short-Term Memory, full connected Deep Neural Networks [ C ]// ICASSP 2015 + 2015IEEE International Conference on Acoustics, speed and Signal Processing (ICASSP). IEEE,2016.

In the above technical solution, the convolutional neural network has three layers, the first layer convolution kernel size is 3 × (3-20), preferably 3 × 20, the second layer convolution kernel size is 2 × (2-20), preferably 2 × 20, and the third layer convolution kernel size is 2 × (2-20), preferably 2 × 20. When the size of the convolution kernel in the convolution neural network is in the range, the accuracy rate can be ensured to be more than 90%.

In the above technical solution, the long-term and short-term memory neural network has three layers, the first layer is an input layer, the second layer is a hidden layer, the hidden unit of the hidden layer is 100, and the third layer is an output layer.

In the above technical solution, the training method of the deep learning model includes the following steps:

step 1: extracting original electroencephalogram time sequence signals of the schizophrenia patient and the healthy person, and performing noise filtering processing, wherein the electroencephalogram time sequence signals of the schizophrenia patient and the healthy person after the noise filtering processing are used as training samples;

step 2: dividing the training samples extracted in the step 1 into a training set and a verification set; the number of the electroencephalogram time sequences in the training set is 70% of the total number of the electroencephalogram time sequences in the training sample, and the number of the electroencephalogram time sequences in the verification set is 30% of the total number of the electroencephalogram time sequences in the training sample.

And step 3: based on a short-time Fourier transform algorithm, respectively converting the training set and the verification set divided in the step (2) into a training set input image and a verification set input image which can be identified by a convolutional neural network, wherein the training set input image and the verification set input image form a model input image set;

and 4, step 4: and (4) inputting the model input image set obtained by conversion in the step (3) into the deep learning model, extracting robustness characteristics, and training optimal parameters of the deep learning model.

And 5: and (4) outputting the optimal parameters of the deep learning model trained in the step (4).

Compared with the prior art, the invention has the beneficial effects that:

1. the original electroencephalogram deep learning classification method based on electroencephalogram and deep learning provided by the invention has the advantage that a deep learning model is composed of a convolutional neural network and a long-term and short-term memory neural network. The convolutional neural network can utilize the convolutional layer to extract the characteristics, so that proper characteristics do not need to be searched repeatedly, and a large amount of time is saved.

2. The original electroencephalogram deep learning classification method based on electroencephalogram and deep learning provided by the invention can improve the accuracy of electroencephalogram analysis by changing the size of a convolution kernel for performing feature extraction by using the convolution layer, is simple and convenient to operate and higher in speed, and can ensure that the accuracy is up to more than 90%.

3. According to the original electroencephalogram deep learning classification method based on electroencephalogram and deep learning, the long-term and short-term memory neural network can judge the time dependence on the characteristic extracted by the convolutional layer to further extract the robustness characteristic.

Drawings

FIG. 1 is a flowchart of an original electroencephalogram deep learning classification method based on electroencephalogram and deep learning.

FIG. 2 is a diagram showing the comparison of the brain electrical time series signals of schizophrenia and healthy people.

Figure 3 shows a 3s window short time fourier transform plot of schizophrenia versus healthy persons.

FIG. 4 shows a block diagram of CNN-LSTM-ELM.

Detailed Description

The present invention will be described in further detail with reference to specific examples. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

Example 1

A deep learning model is composed of a convolution neural network and a long-short term memory neural network. The convolutional neural network has three layers, the size of a convolution kernel of the first layer is 3X20, the size of a convolution kernel of the second layer is 2X20, and the size of a convolution kernel of the third layer is 2X 20. The long-term and short-term memory neural network comprises three layers, wherein the first layer is an input layer, the second layer is a hidden layer, a hidden unit arranged in the hidden layer is 100, and the third layer is an output layer.

The training method of the deep learning model comprises the following steps:

step 1: extracting training samples

Extracting original electroencephalogram time sequence signals of 45 schizophrenia patients and 39 healthy persons, and performing noise filtering processing, wherein the electroencephalogram time sequence signals of the schizophrenia patients and the healthy persons after the noise filtering processing are used as training samples.

Step 2: partitioning training samples

Dividing the training sample extracted in the step 1 into a training set and a verification set, wherein the number of the electroencephalogram time sequences in the training set is 70% of the total number of the electroencephalogram time sequences in the training sample, and the number of the electroencephalogram time sequences in the verification set is 30% of the total number of the electroencephalogram time sequences in the training sample.

And step 3: transforming a model input image set

And (3) respectively converting the training set and the verification set divided in the step (2) into a training set input image and a verification set input image which can be identified by the convolutional neural network (the image format which can be identified by the convolutional neural network is a two-dimensional matrix format) based on a short-time Fourier transform algorithm, wherein the training set input image and the verification set input image form a model input image set.

Example 2

step A: extracting an original electroencephalogram time sequence signal, and carrying out noise filtering processing on the original electroencephalogram time sequence signal, wherein the electroencephalogram time sequence signal of a person to be tested after the noise filtering processing is used as a test sample;

and B: converting the test sample obtained in the step A into a test input image which can be identified by a convolutional neural network based on a short-time Fourier transform algorithm;

and C: inputting the test input image obtained in the step B into the deep learning model trained in the embodiment 1, and extracting the robustness characteristics in the trained deep learning model to obtain the robustness characteristics;

step D: and C, inputting the robustness features extracted in the step C into an extreme learning machine classifier for classification analysis.

The method can be used for detecting schizophrenia, and the extreme learning machine classifier detects schizophrenia and outputs a detection result.

Example 3

This example is based on example 2 and describes the detailed method or preference thereof.

In step 3 of example 1 and step B of example 2, the short-time fourier transform algorithm is completed by the following formula:

In step B, when the obtained test sample is converted into a test input image, short-time fourier transform is performed using a window of 3 seconds.

In the step D, the output detection result is that whether the testee has schizophrenia or does not have schizophrenia is judged according to the electroencephalogram time sequence signal.

The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and decorations can be made without departing from the principle of the present invention, and these modifications and decorations should also be regarded as the protection scope of the present invention.

Claims

1. an original EEG deep learning classification method, is characterized in that: comprise the following steps:

Step A: extracting the original EEG time-series signal, and performing noise filtering processing on it, and using the EEG time-series signal after noise filtering as a test sample;

Step B: Convert the test sample obtained in Step A into a test input image;

Step C: input the test input image obtained in step B into the deep learning model through training, and perform robust feature extraction in the deep learning model to obtain robust features;

Step D: Input the robust features extracted in Step C into the extreme learning machine classifier for classification.

2 . The original EEG deep learning classification method according to claim 1 , wherein in step B, the test sample conversion method is based on a short-time Fourier transform algorithm. 3 .

3. The original EEG deep learning classification method as claimed in claim 2, wherein: the short-time Fourier transform algorithm is completed by the following formula:

Among them, z(u) is the EEG time-series signal at time u, g(u-t) is the window function, f is the sampling rate of the EEG time-series signal, j is a complex number, j=1×i, and i is an imaginary unit.

4 . The original EEG deep learning classification method according to claim 3 , wherein: when the obtained test sample is converted into a test input image, a short-time Fourier transform is performed using a 3-second window. 5 .

5. The original EEG deep learning classification method according to claim 1, wherein in step C, the deep learning model is composed of a convolutional neural network and a long short-term memory neural network.

6. The original EEG deep learning classification method according to claim 5, wherein the convolutional neural network has three layers, and the size of the convolution kernel of the first layer is 3×(3-20), preferably 3× 20. The size of the convolution kernel of the second layer is 2×(2-20), preferably 2×20, and the size of the convolution kernel of the third layer is 2×(2-20), preferably 2×20.

7. The original EEG deep learning classification method according to claim 5, wherein: the long short-term memory neural network has three layers, the first is an input layer, the second layer is a hidden layer, and the hidden layer The set of hidden units is 50-200, and the third layer is the output layer.

8. The original EEG deep learning classification method as claimed in claim 7, wherein: the training method of the deep learning model comprises the following steps:

Step 1: Extract the original EEG time series signals of schizophrenia patients and healthy people, and perform noise filtering processing. The EEG time series signals of schizophrenia patients and healthy people after noise filtering are used as training samples;

Step 2: Divide the training samples extracted in Step 1 into a training set and a validation set;

Step 3: Based on the short-time Fourier transform algorithm, the training set and the validation set divided in step 2 are respectively converted into training set input images and validation set input images recognizable by the convolutional neural network. Set input images to form a model input image set;

Step 4: Input the model input image set converted in step 3 into the deep learning model, perform robust feature extraction, and train the optimal parameters of the deep learning model;

Step 5: The optimal parameter output of the deep learning model trained in Step 4 is applied to the deep learning model in the test phase.

9 . The original EEG deep learning classification method according to claim 7 , wherein: in step 2, the number of EEG time series in the training set is the total number of EEG time series in the training sample. 10 . 70%, the number of EEG time series in the validation set is 30% of the total number of EEG time series in the training sample.

10. The application of the original EEG deep learning classification method according to claims 1-9 in EEG analysis.