A cognitive load assessment method based on EEG signal feature fusion
By preprocessing and feature extraction of EEG signals, combined with the fusion of power spectral density, functional connectivity and micro-state features, the problem of inaccurate cognitive load assessment in the prior art is solved, and the accuracy of cognitive load is achieved Identification improves the security and efficiency of human-computer interaction systems.
Patent Information
- Application Number
- CN202210757622.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-06-30
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2042-06-30
AI Technical Summary
Existing cognitive load assessment methods are difficult to accurately identify different cognitive load levels, especially in human-computer interaction systems, which may lead to operational errors and safety hazards or waste of resources.
By collecting EEG signals, preprocessing is performed, power spectral density, functional connectivity and micro-state features are extracted, and feature fusion is performed, and the trained classification model is input for classification to identify different cognitive load levels.
Accurate identification of cognitive load is achieved, safety hazards caused by excessive load and resource waste caused by excessive load, and the efficiency of human-computer interaction system is improved.
Smart Images

Figure CN115054271B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of cognitive neuroscience, and in particular relates to a cognitive load assessment method based on EEG signal feature fusion. Background Art
[0002] Human-computer interaction technology is a key technology of computer systems, usually implemented in the form of user interfaces. However, with the development of information technology and automation technology, the amount of information exchanged in human-computer interaction has gradually increased, and the difficulty of tasks has also increased. Therefore, cognitive load has become a key factor affecting the performance of operators who play the role of "decision makers" in human-computer interaction systems. When designing human-computer interaction tasks, we can monitor the cognitive load of operators and give them feedback reminders or warnings, etc., to ensure that the cognitive load remains within a reasonable range, thereby reducing the risk of operating errors caused by excessive cognitive load of operators. This is crucial for human-computer interaction systems such as flying airplanes, driving cars, and operating medical equipment, where operating errors caused by excessive cognitive load may cause serious safety hazards. In addition, cognitive load assessment can prevent the waste of cognitive resources caused by too low cognitive load and improve the efficiency of human-computer interaction systems.
[0003] Cognitive load can be understood as a psychological structure under specific tasks and operating environments, reflecting the psychological pressure that operators endure in the process of performing tasks and the requirements for their responsiveness to tasks. There are two measurement methods for cognitive load: subjective and objective. Subjective evaluation usually refers to an individual's self-evaluation of their perception of task complexity through a questionnaire, while objective evaluation focuses on measuring the level of cognitive load through objective indicators. Commonly used objective indicators include neurophysiological signals, such as electroencephalography (EEG) and near-infrared spectroscopy (NIR) signals; physiological signals such as eye movements, electrocardiograms, and heart rate signals; and behavioral signals such as gestures and expressions. Among these objective indicators, EEG has become one of the best means of evaluating the level of cognitive load due to its unique ability to reflect neural activity in the brain; therefore, it is the most commonly used method for objectively evaluating cognitive status. To this end, the present invention proposes a method for measuring cognitive load based on EEG signals.
[0004] EEG recording technology refers to recording low-amplitude electrophysiological activities of the subject's cerebral cortex through a wearable electrode cap, which provides a general reflection of the electrophysiological activities of brain nerve cells on the cerebral cortex or scalp surface. EEG signals contain a large amount of physiological and disease information and are of great value in medical and engineering applications. In recent years, EEG signals have been used to measure cognitive load in various application environments. There are many methods for analyzing EEG signals. In the present invention, the power spectral density, functional connectivity and microstates of EEG signals are extracted, and then the cognitive load is identified through feature fusion. Summary of the invention
[0005] In view of the above technical problems, the present invention provides a cognitive load assessment method based on EEG signal feature fusion.
[0006] The technical solution adopted by the present invention to solve the technical problem is:
[0007] A cognitive load assessment method based on EEG signal feature fusion, the method comprising the following steps:
[0008] Step S100: collecting EEG signals corresponding to different cognitive load levels of the subjects;
[0009] Step S200: preprocessing the EEG signal to obtain a preprocessed EEG signal;
[0010] Step S300: extracting features from the preprocessed EEG signal to obtain power spectrum density, functional connectivity and microstate features;
[0011] Step S400: Perform feature fusion on the power spectral density, functional connectivity and microstate features to obtain fused features at different cognitive load levels, input the fused features at different cognitive load levels into the trained classification model for classification, obtain classification evaluation results, and complete the identification of different cognitive load levels.
[0012] Preferably, step S200 includes:
[0013] Step S210: After channel positioning, use a 0.1-40 Hz bandpass filter to remove high-frequency noise from the EEG signal;
[0014] Step S220: using a 48-52 Hz step-down filter to remove power frequency interference;
[0015] Step S230: Divide the data into 2s long segments, draw the signal waveform, and manually remove the high noise segment;
[0016] Step S240: Using independent component analysis to remove eye artifacts to obtain a preprocessed EEG signal.
[0017] Preferably, in step S300, feature extraction is performed on the preprocessed EEG signal to obtain a power spectrum density, including:
[0018] Step S310: Convert the preprocessed EEG signal into a frequency domain signal using Fourier transform, specifically:
[0019]
[0020] Where X(w) is the Fourier transform of the signal, j is the prefix of the imaginary number, and w represents the number of 2π;
[0021] Step S320: Analyze the signals in different frequency domains to obtain corresponding frequency domain information, and obtain the power spectrum density of different frequency bands according to the frequency domain information, specifically:
[0022]
[0023] Among them, Δt, N and f s They represent the sampling period, the number of samples in the time period and the sampling frequency respectively, and S(w) represents the power spectral density of the corresponding frequency band.
[0024] Preferably, in step S300, feature extraction is performed on the preprocessed EEG signal to obtain functional connectivity, including:
[0025] Step S330: Calculate the cross-spectral density and auto-spectral density of any two signals in the preprocessed EEG signals, specifically:
[0026]
[0027]
[0028] Where N is the length of the signal sequence, X(f) is the discrete Fourier transform of signal x, and S xx (f) is the autospectral density of signal x, Y(f) is the discrete Fourier transform of signal y, S xy (f) is the cross-spectral density of signals x and y;
[0029] Step S340: Obtain the functional connectivity of the two signals in the frequency domain according to the cross-spectral density and the auto-spectral density of any two signals, specifically:
[0030]
[0031] Among them, Coh xy The value range is
[01] .
[0032] Preferably, in step S300, feature extraction is performed on the preprocessed EEG signal to obtain microstate features, including:
[0033] Step S350: Calculate the standard deviation of the signals on all electrodes to obtain a global field potential curve, specifically:
[0034]
[0035] Among them, v i (t) is the voltage value of electrode i at time t, is the average voltage value of all electrodes at time t, and m is the number of electrodes;
[0036] Step S360: The peak point of the global field potential curve represents the moment of highest signal-to-noise ratio, the potential on each electrode at the peak point of the global field potential curve is recorded, a sequence of original topographic maps is obtained, and the topographic map set is interpreted into four micro-state maps using a clustering algorithm;
[0037] Step S370: By selecting the microstate type with the highest correlation with each topographic map in the original topographic map sequence, the original topographic map at the peak of the global field potential curve is represented as a series of four alternating types of microstates, and the average duration of each microstate, the number of occurrences of each microstate per unit time, the ratio of the coverage time of each microstate to the total time, and the probability of transition between the four microstates are analyzed to obtain.
[0038] Preferably, in step S400, feature fusion of power spectrum density, functional connectivity and microstate features includes:
[0039] Step S410: Calculate the ratio of the PSD of a single frequency band to the sum of the PSDs of all frequency bands to obtain a relative PSD, the value of the relative PSD is in the interval [0, 1], the PSD feature is the relative power spectral density of a preset second number of frequency bands of a preset first number of channels, and the first dimension vector = the preset first number * the preset second number;
[0040] Step S420: The functional connectivity feature is a COH matrix of a preset second number of frequency bands of a preset first number of channels, and the second dimensional vector = preset first number * preset second number * (preset first number - 1) / 2;
[0041] Step S430: The microstate characteristics are the average duration of the four microstates, the number of microstate occurrences per unit time, the ratio of the coverage time of the microstate to the total time, and the probability of transition between the four microstates. The third dimension vector = the number of microstates * the number of indicators of each microstate + the total number of transition probabilities between different microstates.
[0042] Step S440: performing feature fusion according to the first dimensional vector, the second dimensional vector and the third dimensional vector to obtain fused features.
[0043] Preferably, before step S500, the initial classification model is trained and tested to obtain a trained classification model, including:
[0044] One-way ANOVA was used to select the features with the greatest difference on the training set data, and the features with a feature ratio ranging from 5% to 95% and a step size of 5% were selected for traversal;
[0045] For each ratio, a five-fold cross validation is performed in the training set data to obtain the classification accuracy. When the classification accuracy reaches the highest accuracy, the corresponding feature ratio is used as the final feature ratio after dimensionality reduction, thereby achieving feature dimensionality reduction.
[0046] After feature dimensionality reduction, a classification algorithm is used for classification to obtain a trained classification model.
[0047] Specifically, the classification results include a normal cognitive load assessment, a cognitive load assessment that is above the normal range of cognitive load for the subject, and a cognitive load assessment that is below the normal range of cognitive load for the subject.
[0048] Preferably, the classification algorithm includes at least one of support vector machine, linear discriminant analysis, random forest, and K nearest neighbor method.
[0049] The above-mentioned cognitive load assessment method based on EEG signal feature fusion analyzes the EEG signals, extracts power spectral density, functional connectivity and microstates, and then classifies the fused features through feature fusion, accurately identifies the cognitive load, prevents operational errors caused by excessive cognitive load that may lead to serious safety hazards and waste of cognitive resources caused by too low cognitive load, and effectively improves the efficiency of the human-computer interaction system. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] Figure 1 It is a flow chart of a cognitive load identification method based on EEG feature fusion of the present invention;
[0051] Figure 2 This is a standard diagram of 32 electrodes used for EEG signal collection in the present invention;
[0052] Figure 3 It is a schematic diagram of the microstate analysis principle used in the present invention. DETAILED DESCRIPTION
[0053] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings.
[0054] In one embodiment, Figure 1 As shown, a cognitive load assessment method based on EEG signal feature fusion includes the following steps:
[0055] Step S100: collecting EEG signals corresponding to different cognitive load levels of the subjects.
[0056] Specifically, EEG recording technology refers to recording low-amplitude electrophysiological activities of the subject's cerebral cortex through a wearable electrode cap, which provides a general reflection of the electrophysiological activities of brain nerve cells on the cerebral cortex or scalp surface. EEG signals contain a large amount of physiological and disease information and are of great value in medical and engineering applications. In the implementation process, the subject first induces different levels of cognitive load through an experimental paradigm, and then collects EEG data. The subject wears an electrode cap to collect Figure 2 The EEG signals of 32 positions shown (in accordance with the international 10-20 system) are sampled at a frequency of 1000 Hz and have a passband of 0 to 200 Hz. The reference electrodes T7 and T8 are located behind the earlobes, the ground electrode is FZ, and the electrode impedance is reduced to less than 5 kΩ.
[0057] Step S200: preprocessing the EEG signal to obtain a preprocessed EEG signal.
[0058] In one embodiment, step S200 includes:
[0059] Step S210: After channel positioning, use a 0.1-40 Hz bandpass filter to remove high-frequency noise from the EEG signal;
[0060] Step S220: using a 48-52 Hz step-down filter to remove power frequency interference;
[0061] Step S230: Divide the data into 2s long segments, draw the signal waveform, and manually remove the high noise segment;
[0062] Step S240: Using independent component analysis to remove eye artifacts to obtain a preprocessed EEG signal.
[0063] Specifically, preprocessing the EEG signal can effectively improve the accuracy of subsequent classification.
[0064] Step S300: extracting features from the preprocessed EEG signal to obtain power spectrum density, functional connectivity and microstate features.
[0065] Specifically, the collected EEG signal is the result of sampling the electrical activity (amplitude) of each electrode over time. It is a time domain signal, which can be converted into a frequency domain signal through Fourier transform. In neuroscience research, EEG signals are usually divided into several frequency bands: δ (1-3Hz); θ (4-7Hz); α (8-12Hz); β (13-30Hz) and y (31-80hz). By analyzing the signals in different frequency bands, significant frequency domain information can be obtained. The power spectral density of different frequency bands is usually estimated in the form of PSD. PSD describes the distribution of the signal between different frequencies.
[0066] In one embodiment, in step S300, feature extraction is performed on the preprocessed EEG signal to obtain a power spectrum density, including:
[0067] Step S310: Convert the preprocessed EEG signal into a frequency domain signal using Fourier transform, specifically:
[0068]
[0069] Where X(w) is the Fourier transform of the signal, j is the prefix of the imaginary number, and w represents the number of 2π;
[0070] Step S320: Analyze the signals in different frequency domains to obtain corresponding frequency domain information, and obtain the power spectrum density of different frequency bands according to the frequency domain information, specifically:
[0071]
[0072] Among them, Δt, N and f s They represent the sampling period, the number of samples in the time period and the sampling frequency respectively, and S(w) represents the power spectral density of the corresponding frequency band.
[0073] Furthermore, in task states and even in resting states, different brain regions not only have structural connections but also functional connections. Functional connectivity is defined as the degree of correlation between signals from two different brain regions in the temporal dimension. The interaction between brain regions can be quantified by spectral coherence (Coh), which measures the degree of correlation between two signals in the frequency domain.
[0074] In one embodiment, in step S300, feature extraction is performed on the preprocessed EEG signal to obtain functional connectivity, including:
[0075] Step S330: Calculate the cross-spectral density and auto-spectral density of any two signals in the preprocessed EEG signals, specifically:
[0076]
[0077]
[0078] Where N is the length of the signal sequence, X(f) is the discrete Fourier transform of signal x, and S xx (f) is the autospectral density of signal x, Y(f) is the discrete Fourier transform of signal y, S xy (f) is the cross-spectral density of signals x and y;
[0079] Step S340: Obtain the functional connectivity of the two signals in the frequency domain according to the cross-spectral density and the auto-spectral density of any two signals, specifically:
[0080]
[0081] Among them, Coh xy The value range is
[01] . The larger the value is, the greater the correlation between the two signals at frequency f.
[0082] Furthermore, in microstate analysis, multi-channel EEG signals are decomposed into a series of transient potential topologies.
[0083] In one embodiment, in step S300, feature extraction is performed on the preprocessed EEG signal to obtain microstate features, including:
[0084] Step S350: Calculate the standard deviation of the signals on all electrodes to obtain a global field potential curve, specifically:
[0085]
[0086] Among them, v i (t) is the voltage value of electrode i at time t, is the average voltage value of all electrodes at time t, and m is the number of electrodes;
[0087] Step S360: The peak point of the global field potential curve represents the moment of highest signal-to-noise ratio. The potential on each electrode at the peak point of the global field potential curve is recorded to obtain a sequence of original topographic maps. The topographic map set is interpreted into four micro-state maps using a clustering algorithm. Figure 3 shown.
[0088] Step S370: By selecting the microstate type with the highest correlation with each topographic map in the original topographic map sequence, the original topographic map at the peak of the global field potential curve is represented as a series of four alternating types of microstates, and the average duration of each microstate, the number of occurrences of each microstate per unit time, the ratio of the coverage time of each microstate to the total time, and the probability of transition between the four microstates are analyzed to obtain.
[0089] Specifically, by selecting the microstate type with the highest correlation with each topographic map in the original topographic map sequence, the original topographic map at the peak of the global field potential curve is represented as a series of four alternating types of microstates. The start time of a microstate is the time midpoint between the previous microstate and the current microstate, and the end time is the time midpoint between the current microstate and the next microstate, and the above indicators are then analyzed to obtain the above indicators.
[0090] Step S400: Perform feature fusion on the power spectral density, functional connectivity and microstate features to obtain fused features at different cognitive load levels, input the fused features at different cognitive load levels into the trained classification model for classification, obtain classification evaluation results, and complete the identification of different cognitive load levels.
[0091] Specifically, the output of the classification is a category. For example, there are three cognitive levels to be distinguished (low, medium, and high), and the classification output is that the current sample is at a low level, or the sample is at a high level.
[0092] In one embodiment, the step S400 performs feature fusion on the power spectrum density, functional connectivity and microstate features, including:
[0093] Step S410: Calculate the ratio of the PSD of a single frequency band to the sum of the PSDs of all frequency bands to obtain a relative PSD, the value of the relative PSD is in the interval [0, 1], the PSD feature is the relative power spectral density of a preset second number of frequency bands of a preset first number of channels, and the first dimension vector = the preset first number * the preset second number;
[0094] Step S420: The functional connectivity feature is a COH matrix of a preset second number of frequency bands of a preset first number of channels, and the second dimensional vector = preset first number * preset second number * (preset first number - 1) / 2;
[0095] Step S430: The microstate characteristics are the average duration of the four microstates, the number of microstate occurrences per unit time, the ratio of the coverage time of the microstate to the total time, and the probability of transition between the four microstates. The third dimension vector = the number of microstates * the number of indicators of each microstate + the total number of transition probabilities between different microstates.
[0096] Step S440: performing feature fusion according to the first dimensional vector, the second dimensional vector and the third dimensional vector to obtain fused features.
[0097] Specifically, the value of the relative PSD is within the interval
[01] , thereby ensuring that the model converges more stably. The three features are fused at the feature layer. Taking the 32-channel EEG signal as an example, the PSD feature is the relative power spectral density of the 4 frequency bands of the 32 channels, that is, a 4×32=128-dimensional vector; the functional connectivity feature is the COH matrix under the 32 channels and 4 frequency bands, that is, a 4×32×(32-1) / 2=1984-dimensional vector; the microstate feature is the average duration of the 4 microstates, the number of microstates occurring per unit time, the ratio of the coverage time of the microstate to the total time, and the probability of transition between the four microstates, a total of 4×3+12=24-dimensional vectors. After fusion, it becomes a 128+1984+24=2136-dimensional vector.
[0098] In one embodiment, before step S500, the initial classification model is trained and tested to obtain a trained classification model, including:
[0099] One-way ANOVA was used to select the features with the greatest difference on the training set data, and the features with a feature ratio ranging from 5% to 95% and a step size of 5% were selected for traversal;
[0100] For each ratio, a five-fold cross validation is performed in the training set data to obtain the classification accuracy. When the classification accuracy reaches the highest accuracy, the corresponding feature ratio is used as the final feature ratio after dimensionality reduction, thereby achieving feature dimensionality reduction.
[0101] After feature dimensionality reduction, a classification algorithm is used for classification to obtain a trained classification model.
[0102] Specifically, for example, if the power spectrum density is 128-dimensional, the feature here refers to the 128 values of the power spectrum density. The dimension with the largest difference is selected from them to achieve feature dimensionality reduction.
[0103] In one embodiment, the classification algorithm includes at least one of support vector machine, linear discriminant analysis, random forest, and K nearest neighbor method.
[0104] Specifically, it can be understood that the classification algorithms are not limited to the above ones.
[0105] The above-mentioned cognitive load assessment method based on EEG signal feature fusion analyzes the EEG signals, extracts power spectral density, functional connectivity and microstates, and then classifies the fused features through feature fusion, accurately identifies the cognitive load, prevents operational errors caused by excessive cognitive load that may lead to serious safety hazards and waste of cognitive resources caused by too low cognitive load, and effectively improves the efficiency of the human-computer interaction system.
[0106] The above is a detailed introduction to a cognitive load assessment method based on EEG signal feature fusion provided by the present invention. This article uses specific examples to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the core idea of the present invention. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present invention, the present invention can also be improved and modified, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.
Claims
1. A cognitive load assessment method based on EEG signal feature fusion, It is characterized in that The method comprises the following steps: Step S100: collecting EEG signals corresponding to different cognitive load levels of the subjects; Step S200: preprocessing the EEG signal to obtain a preprocessed EEG signal; Step S300: extracting features from the preprocessed EEG signal to obtain power spectrum density, functional connectivity and microstate features; Step S400: performing feature fusion on the power spectral density, functional connectivity and microstate features to obtain fused features at different cognitive load levels, inputting the fused features at different cognitive load levels into a trained classification model for classification, obtaining classification evaluation results, and completing the identification of different cognitive load levels; In step S300, feature extraction is performed on the preprocessed EEG signal to obtain a power spectrum density, including: Step S310: Convert the preprocessed EEG signal into a frequency domain signal using Fourier transform, specifically: Where X(w) is the Fourier transform of the signal, j is the prefix of the imaginary number, and w represents the number of 2π; Step S320: Analyze signals in different frequency domains to obtain corresponding frequency domain information, and obtain power spectrum densities of different frequency bands according to the frequency domain information, specifically: Among them, Δt, N and f s They represent the sampling period, the number of samples in the time period and the sampling frequency respectively, and S(w) represents the power spectrum density of the corresponding frequency band; In step S300, feature extraction is performed on the preprocessed EEG signal to obtain functional connectivity, including: Step S330: Calculate the cross-spectral density and auto-spectral density of any two signals in the preprocessed EEG signals, specifically: Where N is the length of the signal sequence, X(f) is the discrete Fourier transform of signal x, and S xx (f) is the autospectral density of signal x, Y(f) is the discrete Fourier transform of signal y, S xy (f) is the cross-spectral density of signals x and y; Step S340: Obtaining the functional connectivity of the two signals in the frequency domain according to the cross-spectral density and the auto-spectral density of the two arbitrary signals, specifically: Among them, Coh xy The value range is [01]; In step S300, feature extraction is performed on the preprocessed EEG signal to obtain microstate features, including: Step S350: Calculate the standard deviation of the signals on all electrodes to obtain a global field potential curve, specifically: Among them, v i (t) is the voltage value of electrode i at time t, is the average voltage value of all electrodes at time t, and m is the number of electrodes; Step S360: The peak point of the global field potential curve represents the moment of highest signal-to-noise ratio, the potential on each electrode at the peak point of the global field potential curve is recorded, a sequence of original topographic maps is obtained, and a clustering algorithm is used to interpret the topographic map set into four micro-state maps; Step S370: by selecting the microstate type with the highest correlation with each topographic map in the original topographic map sequence, the original topographic map at the peak of the global field potential curve is represented as a series of four alternating types of microstates, and the average duration of each microstate, the number of occurrences of each microstate per unit time, the ratio of the coverage time of each microstate to the total time, and the probability of conversion between the four microstates are analyzed; In step S400, feature fusion is performed on the power spectrum density, functional connectivity and microstate features, including: Step S410: Calculate the ratio of the PSD of a single frequency band to the sum of the PSDs of all frequency bands to obtain a relative PSD, the value of the relative PSD is in the interval [0, 1], the PSD feature is the relative power spectral density of a preset second number of frequency bands of a preset first number of channels, and the first dimension vector = the preset first number * the preset second number; Step S420: The functional connectivity feature is a COH matrix of a preset second number of frequency bands of a preset first number of channels, and the second dimensional vector = preset first number * preset second number * (preset first number - 1) / 2; Step S430: The microstate characteristics are the average duration of the four microstates, the number of microstate occurrences per unit time, the ratio of the coverage time of the microstate to the total time, and the probability of transition between the four microstates. The third dimension vector = the number of microstates * the number of indicators of each microstate + the total number of transition probabilities between different microstates. Step S440: performing feature fusion according to the first dimensional vector, the second dimensional vector and the third dimensional vector to obtain fused features.
2. The method according to claim 1, It is characterized in that Step S200 includes: Step S210: After channel positioning, high-frequency noise of the EEG signal is removed using a 0.1-40 Hz bandpass filter; Step S220: using a 48-52 Hz step-down filter to remove power frequency interference; Step S230: Divide the data into 2s long segments, draw the signal waveform, and manually remove the high noise segment; Step S240: Using independent component analysis to remove eye artifacts to obtain a preprocessed EEG signal.
3. The method according to claim 2, It is characterized in that The steps before step S500 include training and testing the initial classification model to obtain a trained classification model, including: One-way ANOVA was used to select the features with the greatest difference on the training set data, and the features with a feature ratio ranging from 5% to 95% and a step size of 5% were selected for traversal; For each ratio, a five-fold cross validation is performed in the training set data to obtain the classification accuracy. When the classification accuracy reaches the highest accuracy, the corresponding feature ratio is used as the final feature ratio after dimensionality reduction, thereby achieving feature dimensionality reduction. After feature dimensionality reduction, a classification algorithm is used for classification to obtain a trained classification model.
4. The method according to claim 3, It is characterized in that The classification algorithm includes at least one of support vector machine, linear discriminant analysis, random forest, and K nearest neighbor method.
Citation Information
Cited By
Multi-mode electroencephalogram characteristic cognitive ability evaluation system based on neural network
CN120256918A
