Method for EEG identity recognition based on dimension reduction with genetic algorithm

Abstract

The invention discloses a method for EEG identity recognition based on dimension reduction with a genetic algorithm. The method is used to conduct identity recognition according to a user's EEG signals and specifically comprises the steps that EEG data is collected; the EEG data is pre-processed; EEG characteristic parameters are extracted by an AR model; a classifier of a BP neural network is established; the genetic algorithm is used for dimension reduction of the EEG parameters; and a classifier is established according to the dimension-reduced EEG parameters and used for the identity recognition. The method uses the genetic algorithm to conduct the dimension reduction of the EEG parameters creatively; the method can effectively reduce data dimensions and reserve information beneficial for EEG recognition at the same time, so a higher recognition rate can be maintained; and the method is also applicable to EEG data of an individual which is not recognized before, namely the method can still keep a very high recognition rate in set opening verification.

Description

technical field [0001] The invention belongs to the field of identity recognition, and in particular relates to an EEG identity recognition method based on genetic algorithm dimension reduction. Background technique [0002] There are still few existing methods of using EEG signals for identity recognition, basically analyzing EEG data in the frequency domain, extracting EEG signals of a certain band, and using time series models (AR, BL) to fit the EEG data. Data, the model parameters after fitting are extracted as the characteristic parameters of the EEG signal, and after PCA dimensionality reduction processing or several electrodes are selected, the EEG can be directly and simply used with a single-structure learning machine such as a support vector machine or a neural network. Identification. [0003] There is following shortcoming in prior art: [0004] 1. For the estimation of the order of AR or BL model, the AIC criterion or empirical estimation is generally used at...

AI Technical Summary

Problems solved by technology

Although the AIC criterion has many advantages in determining the order of the model, the model research results show that if the data is fitted with an autoregressive model, it may overestimate the order p, and at the same time, the order obtained by the AIC criterion is not compatible

[0005] 2. In the traditional method, since PCA dimensionality reduction and feature recognition after dimensionality reduction are carried out in two separate steps, PCA is not closely connected with feature recognition, resulting in EEG data that may be discarded during the PCA dimensionality reduction process. Features that are really beneficial for identification may reduce the identification effect of EEG data

[0006] 3. The noise in the EEG signal varies in size. Although the feature parameter extraction after filtering the EEG data can better identify the individual EEG data, the correct recognition rate is mostly 85% to 90%.

[0007] 4. Most of the current methods still require a large amount of data calculation after dimensionality reduction processing

Selecting the signals of several electrodes that are more related to the stimulation event for processing can effectively reduce the amount of computation and maintain a good recognition effect, but the electrode signals manually selected in this way may not be able to achieve the best training effect of the classifier. Therefore, its recognition rate may not be optimal

[0008] 5. The existing method is closed-set verification when testing the recognition rate. The existing method has a high accuracy rate when testing the closed-set verification, but for the open-set verification, either the error rate is too large, or it is not applicable at all

[0009] 6. When the recognition rate is high, the existing technology has high requirements for the collection of EEG signals. For example, in the 2015 paper "Recognition Research Based on ERP Induced by Fawell Paradigm", the highest recognition rate can reach 98%, but When collecting EEG, the subjects need to focus on multiple characters on the screen, and silently count the number of flashes of the target characters

Images