Electroencephalogram cognitive evaluation system based on particle swarm optimization and deep learning model

Abstract

According to the electroencephalogram cognitive evaluation system based on the particle swarm optimization and the deep learning model, an upper computer controls VR glasses to play A VR audios and videos for each of B historical evaluators and controls a head-mounted electroencephalogram acquisition device to continuously acquire electroencephalogram signals of a plurality of channels in a plurality of cognitive-related brain regions when each historical evaluator watches; the electroencephalogram signals of all channels of all brain regions are preprocessed, and historical evaluators correspond to cognitive function labels; performing analysis processing and feature extraction on the preprocessed electroencephalogram signals of each channel of each brain region to construct one-dimensional electroencephalogram features, so as to obtain A-dimensional electroencephalogram features of each historical evaluator; constructing a training sample set and a verification sample set; and optimizing hyper-parameters of the deep learning neural network model by using an improved particle swarm algorithm, training and verifying the deep learning neural network model by respectively using the training sample set and the verification sample set, and constructing an optimal deep learning neural network model through the optimized optimal hyper-parameters.

Description

Technical Field

[0001] The present invention relates to the technical field of electroencephalogram (EEG) cognitive assessment, and in particular to an EEG cognitive assessment system based on a particle swarm algorithm and a deep learning model. Background Art

[0002] The designer of the present invention previously designed an invention patent with application number 2024110987333, entitled A cognitive dysfunction assessment system based on virtual reality and EEG signals. This patent mentions that the evaluator's EEG characteristics can be directly input into a trained cognitive dysfunction assessment model (using a deep learning model) for prediction to obtain the evaluator's cognitive dysfunction level. However, this patent does not mention how to obtain a trained cognitive dysfunction assessment model or what kind of trained cognitive dysfunction assessment model can be used to obtain a more accurate evaluator's cognitive dysfunction level. Based on this, the present invention designs an EEG cognitive assessment system based on a particle swarm algorithm and a deep learning model. Summary of the invention

[0003] In view of the problems and shortcomings of the prior art, the present invention provides an EEG cognitive assessment system based on a particle swarm algorithm and a deep learning model.

[0004] The present invention solves the above technical problems through the following technical solutions:

[0005] The present invention provides an electroencephalogram cognitive evaluation system based on a particle swarm algorithm and a deep learning model, which is characterized in that it includes VR glasses, a head-mounted electroencephalogram acquisition device and a host computer;

[0006] The host computer is used to control the VR glasses to play A VR audio and video for each of the B historical evaluators to watch, the A VR audio and video playing time is the same and there is the same time interval between adjacent videos, and at the same time, the head-mounted EEG acquisition device is controlled to continuously collect EEG signals of multiple channels in multiple brain regions related to cognition when each historical evaluator watches, and the EEG signals of each channel of each brain region collected by each historical evaluator are preprocessed to obtain preprocessed EEG signals, A and B are positive integers and A≥2, B≥2, and each historical evaluator corresponds to a cognitive function label, and the cognitive function label is a label with no cognitive dysfunction or a label with different degrees of cognitive dysfunction;

[0007] The host computer is also used to analyze and process the pre-processed EEG signals of each channel of each brain region corresponding to the ath VR audio and video of the bth historical evaluator to obtain qualified EEG signals of each channel of each brain region, and then analyze and calculate the variance, δ wave power spectral density, θ wave power spectral density, α wave power spectral density, β wave power spectral density and γ wave power spectral density of qualified EEG signals of each channel corresponding to each brain region, and the connectivity coefficient between each brain region, and construct a one-dimensional EEG feature based on the average variance, δ wave power spectral density, θ wave power spectral density, α wave power spectral density, β wave power spectral density, γ wave power spectral density and the connectivity coefficient between each brain region, a is taken as 1-A in turn, so as to obtain the A-dimensional EEG feature corresponding to the bth historical evaluator, and b is taken as 1-B in turn, so as to obtain the A-dimensional EEG feature corresponding to each historical evaluator;

[0008] The host computer is also used to construct a training sample set and a verification sample set, wherein the training sample set includes training samples composed of the A-dimensional EEG features and corresponding cognitive function labels of most historical evaluators, and the verification sample set includes verification samples composed of the A-dimensional EEG features and corresponding cognitive function labels of the remaining historical evaluators;

[0009] The host computer is also used to build a deep learning neural network model based on the convolutional neural network, use the improved particle swarm algorithm to optimize the hyperparameters of the deep learning neural network model, and use the training sample set and the verification sample set to train and verify the deep learning neural network model respectively, and build the optimal deep learning neural network model through iteratively optimized optimal hyperparameters.

[0010] Furthermore, the host computer is also used to control the VR glasses to play A VR audio and video for the current person to be evaluated to watch, and at the same time control the head-mounted EEG acquisition device to continuously collect EEG signals of multiple channels in multiple brain areas related to cognition when the current person to be evaluated is watching, and pre-process the EEG signals of each channel in each brain area collected by the current person to be evaluated to obtain pre-processed EEG signals;

[0011] The host computer is also used to analyze and process the pre-processed EEG signals of each channel of each brain region corresponding to the ath VR audio and video of the current person to be evaluated to obtain qualified EEG signals of each channel of each brain region, and then analyze and calculate the variance, δ wave power spectrum density, θ wave power spectrum density, α wave power spectrum density, β wave power spectrum density and γ wave power spectrum density of the qualified EEG signals of each channel corresponding to each brain region, and the connectivity coefficient between each brain region, and construct a one-dimensional EEG feature based on the average variance, δ wave power spectrum density, θ wave power spectrum density, α wave power spectrum density, β wave power spectrum density, γ wave power spectrum density and the connectivity coefficient between each brain region, a is taken as 1-A in turn, so as to obtain the A-dimensional EEG feature corresponding to the current person to be evaluated;

[0012] The host computer is also used to input the A-dimensional EEG features corresponding to the current person to be evaluated into the optimal deep learning neural network model for evaluation, so as to output the cognitive function of the current person to be evaluated, and the cognitive function includes no cognitive dysfunction, mild cognitive dysfunction, moderate cognitive dysfunction and severe cognitive dysfunction.

[0013] The positive and progressive effects of the present invention are:

[0014] The present invention utilizes an improved particle swarm algorithm to optimize the hyperparameters of a deep learning neural network model serving as a cognitive dysfunction assessment model, and constructs an optimal deep learning neural network model through iterative optimization of the optimal hyperparameters, thereby obtaining a deep learning neural network model with higher accuracy (optimal deep learning neural network model), thereby being able to obtain a cognitive function assessment result with higher accuracy for the current subject to be assessed. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a block diagram of an implementation of an EEG cognitive assessment system according to a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0017] like Figure 1 As shown, this embodiment provides an EEG cognitive assessment system based on a particle swarm algorithm and a deep learning model, which includes VR glasses, a head-mounted EEG acquisition device and a host computer.

[0018] EEG signal acquisition and noise reduction filtering preprocessing of historical evaluators: The host computer is used to control the VR glasses to play A VR audio and video for each of the B historical evaluators for viewing. The playback time of the A VR audio and video is the same and there is the same time interval between adjacent videos. At the same time, the head-mounted EEG acquisition device is controlled to continuously collect EEG signals of multiple channels in the frontal lobe, parietal lobe, occipital lobe and temporal lobe of each historical evaluator when watching. The EEG signals of each channel in each brain area collected from each historical evaluator are preprocessed by noise reduction filtering to obtain preprocessed EEG signals. A and B are positive integers and A≥2, B≥2. Each historical evaluator corresponds to a cognitive function label, which is a label with no cognitive dysfunction or a label with different degrees of cognitive dysfunction.

[0019] EEG signal analysis and feature extraction of historical evaluators: The host computer is also used to analyze and process the pre-processed EEG signals of each channel of each brain region corresponding to the ath VR audio and video of the bth historical evaluator to obtain qualified EEG signals of each channel of each brain region; then, based on the qualified EEG signals of each channel of each brain region, the variance, δ wave power spectral density, θ wave power spectral density, α wave power spectral density, β wave power spectral density and γ wave power spectral density of the qualified EEG signals of each channel corresponding to each brain region are calculated, as well as the connectivity coefficient between each brain region; based on the average variance, average δ wave power spectral density, average θ wave power spectral density, average α wave power spectral density, average β wave power spectral density, average γ wave power spectral density and connectivity coefficient between each brain region, a-dimensional EEG features are constructed, a is taken as 1-A in sequence, so as to obtain the A-dimensional EEG features corresponding to the bth historical evaluator, and b is taken as 1-B in sequence, so as to obtain the A-dimensional EEG features corresponding to each historical evaluator.

[0020] Among them, the pre-processed EEG signals of each channel of each brain region corresponding to the ath VR audio and video of the bth historical evaluator are analyzed and processed to obtain qualified EEG signals of each channel of each brain region: determine whether there are abnormal signals at each sampling point in the pre-processed EEG signals of each channel of each brain region. When there are no abnormal signals at each sampling point, the pre-processed EEG signals of each channel of each brain region are used as qualified EEG signals. When there are abnormal signals at the sampling points, abnormal signal analysis and processing are performed to obtain the EEG signals of each channel of each brain region as qualified EEG signals. For the specific implementation process of performing abnormal signal analysis and processing when there are abnormal signals at the sampling points to obtain the EEG signals of each channel of each brain region as qualified EEG signals, refer to the invention patent with application number 2024110987333 and the name of a cognitive dysfunction assessment system based on virtual reality and EEG signals. This embodiment includes the specific implementation process of obtaining qualified EEG signals through abnormal signal analysis and processing in the patent.

[0021] For calculating the connectivity coefficient between each brain region, please refer to the invention patent with application number 2024110987333, entitled A cognitive dysfunction assessment system based on virtual reality and electroencephalogram signals. This embodiment includes the specific calculation method of the connectivity coefficient between each brain region in the patent.

[0022] Construct training sample sets and verification sample sets based on EEG signal features of historical evaluators: The host computer is also used to construct training sample sets and verification sample sets. The training sample set includes training samples composed of A-dimensional EEG features and corresponding cognitive function labels of most historical evaluators, and the verification sample set includes verification samples composed of A-dimensional EEG features and corresponding cognitive function labels of the remaining historical evaluators.

[0023] The improved particle swarm algorithm is used to optimize the model and build the optimal deep learning neural network model: The host computer is also used to build a deep learning neural network model based on the convolutional neural network, and the improved particle swarm algorithm is used to optimize the hyperparameters of the deep learning neural network model. The training sample set and the verification sample set are used to train and verify the deep learning neural network model, and the optimal deep learning neural network model is built through iterative optimization of the optimal hyperparameters. The deep learning neural network model consists of multiple feature extraction layers, a fully connected layer, and a Softmax classifier. Each feature extraction layer includes a convolution layer, a batch normalization layer, and an activation function layer.

[0024] Among them, a deep learning neural network model is constructed based on a convolutional neural network, and an improved particle swarm algorithm is used to optimize the hyperparameters of the deep learning neural network model. The deep learning neural network model is trained and verified using a training sample set and a verification sample set respectively. The optimal deep learning neural network model is constructed through iterative optimization of the optimal hyperparameters, specifically:

[0025] 1) A deep learning neural network model is constructed based on the convolutional neural network, and the hyperparameters of the model are initialized. The input of the deep learning neural network model is the A-dimensional EEG features of each historical evaluator in the training sample set, and the output is the corresponding cognitive function label. The hyperparameters include the number of network layers, the size and number of convolution kernels of the convolution layer of each feature extraction layer, the number of neurons in the fully connected layer, and the learning rate.

[0026] 2) Construct a particle swarm, set the number of particles N, the maximum number of iterations T, and the particle dimension D corresponding to the hyperparameters, and initialize the speed of each particle and location represents the initial velocity of the nth particle at the dth dimension, Represents the initial position of the d-th dimension of the n-th particle.

[0027] 3) Each particle is used as a hyperparameter to construct a deep learning neural network model, and the deep learning neural network model constructed by the particle is trained using the training sample set, and the deep learning neural network model constructed by the particle is verified using the verification sample set. The calculated model accuracy is used as the fitness value of the particle, and the initial fitness value of each particle is obtained based on this calculation. Based on each initial fitness value, the individual optimal position of each particle in the initialized particle group is obtained. and the optimal position of the group

[0028] 4) Iteratively update the speed and position of each particle

[0029] The particle velocity and position update formula is:

[0030]

[0031] Among them, the initial value of t is 1, W t represents the inertia weight of the t-th iteration, represents the velocity of the nth particle at the dth dimension after t-1 iterations, represents the individual optimal position of the d-dimensional position of the n-th particle after t-1 iterations, It represents the optimal position of the particle swarm in the dth dimension after t-1 iterations. represents the position of the d-dimensional position of the n-th particle after t-1 iterations, C1 and C2 represent learning factors, δ represents the acceleration factor, r1 and r2 represent random numbers between (0,1), and λ t represents the iteration step size of the tth iteration, W min Indicates the minimum inertia weight, W max represents the maximum inertia weight, σ represents the inertia adjustment factor, and λ min represents the minimum iteration step size, λ max Indicates the maximum iteration step size.

[0032] The right side of the formula “=” consists of four parts: 1. The inertial part, which represents the tendency of particles to maintain their original speed; 2. The cognitive part, which reflects the particle’s memory or recollection of its own historical experience, and represents the tendency of particles to approach their own historical best position; 3. The social part, which reflects the group’s historical experience of collaborative cooperation and knowledge sharing among particles, and represents the tendency of particles to approach the group’s or neighborhood’s historical best position; 4. The acceleration part, which reflects the acceleration of the particle.

[0033] In this embodiment, improvements are made on the basis of the original particle velocity and position update formulas. A fourth acceleration part is added to the particle velocity formula, and improvements are made to the inertia part, the cognitive part and the acceleration part. An iteration step is added to the particle position formula, thereby improving the optimization speed and optimization ability of the particle swarm algorithm for the deep learning neural network model.

[0034] 5) Update the individual optimal position of each particle and the group optimal position of the particle swarm according to the position of each particle, calculate the fitness value of each particle after the tth iteration, and when the fitness value of the nth particle after the tth iteration is greater than The fitness value of Otherwise, it will not be updated, and the fitness value corresponding to the best position in the individual best position of each particle will be compared with If the fitness value is greater than = the best position among the individual optimal positions, otherwise no update.

[0035] 6) Determine whether the optimal position of the group reaches the set threshold or the number of iterations reaches the maximum number of iterations T. If yes, the algorithm ends, and if no, execute 7).

[0036] 7) Sort the fitness values ​​of each particle after the t-th iteration from large to small, discard the N1 particles with the lowest fitness values, randomly generate N1 particles in the area formed by the positions of the first N2 particles, the randomly generated N1 particles correspond one to one to the discarded N1 particles, the current positions of the randomly generated N1 particles are respectively used as the positions of the N1 particles after the t-th iteration, the calculated speeds of the randomly generated N1 particles are respectively used as the speeds of the N1 particles after the t-th iteration, the calculated speed of a particle among the randomly generated N1 particles = the average of the speed of the particle closest to the particle among the first N2 particles and the speed of the corresponding discarded particle, N1 and N2 are positive integers and 1≤N1≤1 / 3N, N=N1+N2, tt=t+1, and execute 4).

[0037] In this embodiment, bad particles are discarded after each iteration, and the positions and velocities of good particles are used to optimize the positions and velocities of particles corresponding to the discarded particles, thereby reducing the number of iterations, improving iteration efficiency, and further improving the optimization speed and optimization ability of the particle swarm algorithm for the deep learning neural network model.

[0038] EEG signal collection and noise reduction filtering preprocessing for the current subject to be evaluated: The host computer is also used to control the VR glasses to play A VR audio and video for the current subject to be evaluated for viewing, and at the same time control the head-mounted EEG acquisition device to continuously collect EEG signals of multiple channels in the frontal lobe, parietal lobe, occipital lobe and temporal lobe brain regions when the current subject to be evaluated is watching, and perform noise reduction filtering preprocessing on the EEG signals of each channel in each brain region collected from the current subject to be evaluated to obtain preprocessed EEG signals.

[0039] EEG signal analysis and feature extraction of the current subject to be evaluated: the host computer is also used to analyze and process the pre-processed EEG signals of each channel of each brain region corresponding to the ath VR audio and video of the current subject to be evaluated to obtain qualified EEG signals of each channel of each brain region, and then calculate the variance, δ wave power spectral density, θ wave power spectral density, α wave power spectral density, β wave power spectral density and γ wave power spectral density of the qualified EEG signals of each channel corresponding to each brain region based on the analysis of the qualified EEG signals of each channel of each brain region, as well as the connectivity coefficient between each brain region; based on the average variance, average δ wave power spectral density, average θ wave power spectral density, average α wave power spectral density, average β wave power spectral density, average γ wave power spectral density and connectivity coefficient between each brain region, a-dimensional EEG features are constructed, and a is taken as 1-A in turn, so as to obtain the A-dimensional EEG features corresponding to the current subject to be evaluated.

[0040] The cognitive function of the current person to be evaluated is evaluated based on the EEG characteristics and the optimal deep learning neural network model: the host computer is also used to input the A-dimensional EEG characteristics corresponding to the current person to be evaluated into the optimal deep learning neural network model for evaluation, so as to output the cognitive function of the current person to be evaluated, and the cognitive functions include no cognitive dysfunction, mild cognitive dysfunction, moderate cognitive dysfunction and severe cognitive dysfunction.

[0041] Although the specific embodiments of the present invention are described above, those skilled in the art should understand that these are only examples, and the protection scope of the present invention is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, but these changes and modifications all fall within the protection scope of the present invention.

Claims

1. An EEG cognitive evaluation system based on particle swarm algorithm and deep learning model, characterized in that: It includes VR glasses, head-mounted EEG acquisition device and host computer; The host computer is used to control the VR glasses to play A VR audio and video for each of the B historical evaluators to watch, the A VR audio and video playing time is the same and there is the same time interval between adjacent videos, and at the same time, the head-mounted EEG acquisition device is controlled to continuously collect EEG signals of multiple channels in multiple brain regions related to cognition when each historical evaluator watches, and the EEG signals of each channel of each brain region collected by each historical evaluator are preprocessed to obtain preprocessed EEG signals, A and B are positive integers and A≥2, B≥2, and each historical evaluator corresponds to a cognitive function label, and the cognitive function label is a label with no cognitive dysfunction or a label with different degrees of cognitive dysfunction; The host computer is also used to analyze and process the pre-processed EEG signals of each channel of each brain region corresponding to the ath VR audio and video of the bth historical evaluator to obtain qualified EEG signals of each channel of each brain region, and then analyze and calculate the variance, δ wave power spectral density, θ wave power spectral density, α wave power spectral density, β wave power spectral density and γ wave power spectral density of qualified EEG signals of each channel corresponding to each brain region, and the connectivity coefficient between each brain region, and construct a one-dimensional EEG feature based on the average variance, δ wave power spectral density, θ wave power spectral density, α wave power spectral density, β wave power spectral density, γ wave power spectral density and the connectivity coefficient between each brain region, a is taken as 1-A in turn, so as to obtain the A-dimensional EEG feature corresponding to the bth historical evaluator, and b is taken as 1-B in turn, so as to obtain the A-dimensional EEG feature corresponding to each historical evaluator; The host computer is also used to construct a training sample set and a verification sample set, wherein the training sample set includes training samples composed of the A-dimensional EEG features and corresponding cognitive function labels of most historical evaluators, and the verification sample set includes verification samples composed of the A-dimensional EEG features and corresponding cognitive function labels of the remaining historical evaluators; The host computer is also used to build a deep learning neural network model based on the convolutional neural network, use the improved particle swarm algorithm to optimize the hyperparameters of the deep learning neural network model, and use the training sample set and the verification sample set to train and verify the deep learning neural network model respectively, and build the optimal deep learning neural network model through iteratively optimized optimal hyperparameters.

2. The EEG cognitive assessment system based on particle swarm algorithm and deep learning as claimed in claim 1, characterized in that: The host computer is also used to control the VR glasses to play A VR audio and video for the current person to be evaluated to watch, and at the same time control the head-mounted EEG acquisition device to continuously collect EEG signals of multiple channels in multiple brain regions related to cognition when the current person to be evaluated is watching, and pre-process the EEG signals of each channel in each brain region collected by the current person to be evaluated to obtain pre-processed EEG signals; The host computer is also used to analyze and process the pre-processed EEG signals of each channel of each brain region corresponding to the ath VR audio and video of the current person to be evaluated to obtain qualified EEG signals of each channel of each brain region, and then analyze and calculate the variance, δ wave power spectrum density, θ wave power spectrum density, α wave power spectrum density, β wave power spectrum density and γ wave power spectrum density of the qualified EEG signals of each channel corresponding to each brain region, and the connectivity coefficient between each brain region, and construct a one-dimensional EEG feature based on the average variance, δ wave power spectrum density, θ wave power spectrum density, α wave power spectrum density, β wave power spectrum density, γ wave power spectrum density and the connectivity coefficient between each brain region, a is taken as 1-A in turn, so as to obtain the A-dimensional EEG feature corresponding to the current person to be evaluated; The host computer is also used to input the A-dimensional EEG features corresponding to the current person to be evaluated into the optimal deep learning neural network model for evaluation, so as to output the cognitive function of the current person to be evaluated, and the cognitive function includes no cognitive dysfunction, mild cognitive dysfunction, moderate cognitive dysfunction and severe cognitive dysfunction.

3. The EEG cognitive assessment system based on particle swarm algorithm and deep learning as claimed in claim 1, characterized in that: The deep learning neural network model consists of multiple feature extraction layers, a fully connected layer and a Softmax classifier. Each feature extraction layer includes a convolution layer, a batch normalization layer and an activation function layer.

4. The EEG cognitive assessment system based on particle swarm algorithm and deep learning as claimed in claim 3, characterized in that: The host computer is also used to build a deep learning neural network model based on the convolutional neural network, use the improved particle swarm algorithm to optimize the hyperparameters of the deep learning neural network model, and use the training sample set and the verification sample set to learn and train and verify the deep learning neural network model respectively, and build the optimal deep learning neural network model through the optimal hyperparameters optimized by iteration, specifically: 1) Construct a deep learning neural network model based on the convolutional neural network and initialize the model's hyperparameters. The input of the deep learning neural network model is the A-dimensional EEG features of each historical evaluator in the training sample set, and the output is the corresponding cognitive function label. The hyperparameters include the number of network layers, the size and number of convolution kernels of the convolution layer of each feature extraction layer, the number of neurons in the fully connected layer, and the learning rate; 2) Construct a particle swarm, set the number of particles N, the maximum number of iterations T, and the particle dimension D corresponding to the hyperparameters, and initialize the speed of each particle and location represents the initial velocity of the nth particle at the dth dimension, represents the initial position of the d-dimensional position of the n-th particle; 3) Each particle is used as a hyperparameter to construct a deep learning neural network model, and the deep learning neural network model constructed by the particle is trained using the training sample set, and the deep learning neural network model constructed by the particle is verified using the verification sample set. The calculated model accuracy is used as the fitness value of the particle, and the initial fitness value of each particle is obtained based on this calculation. Based on each initial fitness value, the individual optimal position of each particle in the initialized particle group is obtained. and the optimal position of the group 4) Iteratively update the speed and position of each particle; The particle velocity and position update formula is: Among them, the initial value of t is 1, W t represents the inertia weight of the t-th iteration, represents the velocity of the nth particle at the dth dimension after t-1 iterations, represents the individual optimal position of the d-dimensional position of the n-th particle after t-1 iterations, It represents the optimal position of the particle swarm in the dth dimension after t-1 iterations. represents the position of the d-dimensional position of the n-th particle after t-1 iterations, C1 and C2 represent learning factors, δ represents the acceleration factor, r1 and r2 represent random numbers between (0,1), and λ t represents the iteration step size of the tth iteration, W min Indicates the minimum inertia weight, W max represents the maximum inertia weight, σ represents the inertia adjustment factor, and λ min represents the minimum iteration step size, λ max Indicates the maximum iteration step size; 5) Update the individual optimal position of each particle and the group optimal position of the particle swarm according to the position of each particle, calculate the fitness value of each particle after the tth iteration, and when the fitness value of the nth particle after the tth iteration is greater than The fitness value of Otherwise, it will not be updated, and the fitness value corresponding to the best position in the individual best position of each particle will be compared with If the fitness value is greater than The best position among the individual's best positions, otherwise no update; 6) Determine whether the optimal position of the group reaches the set threshold or the number of iterations reaches the maximum number of iterations T. If yes, the algorithm ends, and if no, execute 7); 7) t = t + 1, execute 4).

5. The EEG cognitive assessment system based on particle swarm algorithm and deep learning as claimed in claim 4, characterized in that: In 7), the fitness values ​​of the particles after the t-th iteration are sorted from large to small, the N1 particles with the lowest fitness values ​​are discarded, and N1 particles are randomly generated in the area formed by the positions of the first N2 particles. The randomly generated N1 particles correspond to the discarded N1 particles one by one. The current positions of the randomly generated N1 particles are respectively used as the positions of the N1 particles after the t-th iteration, and the calculation speeds of the randomly generated N1 particles are respectively used as the speeds of the N1 particles after the t-th iteration. The calculation speed of a particle among the randomly generated N1 particles = the average of the speed of the particle closest to the particle among the first N2 particles and the speed of the corresponding discarded particle. N1 and N2 are positive integers and 1≤N1≤1 / 3N, N=N1+N2, t=t+1, and execute 4).

6. The EEG cognitive assessment system based on particle swarm algorithm and deep learning as claimed in claim 1, characterized in that: The host computer is used to control the head-mounted EEG acquisition device to continuously collect EEG signals of multiple channels in the frontal lobe brain area, parietal lobe brain area, occipital lobe brain area and temporal lobe brain area when each historical evaluator is watching, and perform noise reduction filtering preprocessing on the EEG signals of each channel in each brain area collected by each historical evaluator to obtain preprocessed EEG signals.

7. The EEG cognitive assessment system based on particle swarm algorithm and deep learning as claimed in claim 1, characterized in that: The host computer is also used to analyze and process the pre-processed EEG signals of each channel in each brain region corresponding to the a-th VR audio and video of the b-th historical evaluator to obtain qualified EEG signals of each channel in each brain region: determine whether there are abnormal signals at each sampling point in the pre-processed EEG signals of each channel in each brain region; when there are no abnormal signals at each sampling point, the pre-processed EEG signals of each channel in each brain region are used as qualified EEG signals; when there are abnormal signals at the sampling points, abnormal signal analysis and processing are performed to obtain the EEG signals of each channel in each brain region as qualified EEG signals.

Images