A wearable vision detection device based on EEG signals

Abstract

The present invention discloses a wearable vision detection device based on EEG signals, including a portable EEG acquisition device, an EEG recognition module and a vision detection module; the portable EEG acquisition device is used to play a stimulation paradigm video and collect EEG signals from the subject's brain at the same time; the EEG recognition module is used to control the video playback of the portable EEG acquisition device and to verify and analyze the EEG signals; the vision detection module performs preprocessing, feature analysis and extraction, feature fusion and model detection on the EEG signals; wherein, the fusion features of the subject are detected through a pre-constructed model to obtain the subject's vision range. The present invention is based on an objective evaluation system based on visual-brainwave response, and can establish a visual function evaluation system independent of a vision chart system based on visual resolution-cognition-subjective expression, redefining the basic concept of vision; at the same time, the present invention has the characteristics of being less affected by the external environment, being applicable to a wide range of people, and being portable.

Description

Technical Field

[0001] The present invention relates to vision detection device technology, and in particular to a wearable vision detection device based on electroencephalogram signals. Background Art

[0002] A previous research report by the World Health Organization showed that there are as many as 600 million myopia patients in China. The reasons behind this include low efficiency of myopia detection, untimely prevention and control, and neglect of early vision protection. Faced with such a large number of myopia patients, it is necessary to improve the efficiency of myopia monitoring and timely prevention and control, and to achieve early detection, early diagnosis, and early intervention.

[0003] However, the traditional detection method is mostly letter vision chart, which has the following main shortcomings: 1. It is applicable to a limited number of people. The traditional letter vision chart requires full understanding and cooperation between the examiner and the examinee, while people with communication disorders, infants and young children cannot interact with the examiner in a benign manner; 2. It is applicable to a limited number of occasions. The traditional vision test requires the examinee to go to the hospital and other places for examination under the guidance of professionals, which is not suitable for people with limited mobility and self-testing at home after amblyopia rehabilitation training. In addition, the traditional vision test cannot be completed independently at home due to factors such as brightness and detection distance, and vision testing must also be carried out under the guidance of professionals; 3. The test results are subjective. When using the letter vision chart for testing, on the one hand, it is affected by factors such as the brightness of the detection environment and the size of the sound, and on the other hand, it is affected by the examinee himself. For example, when the letters become blurred, the examinee will begin to guess the direction of the letters, and sometimes even pass the test through memory. As a signal that reflects the objective physiological state of the human brain, the EEG signal contains a lot of physiological and disease information, which also reflects information about human vision.

[0004] In the current existing technology, most vision detection methods still use traditional detection methods. A small number of vision detection methods will be based on intelligent equipment for vision detection, which improves the efficiency of detection and saves time and manpower. For example, CN202021139491.5 is a vision detection instrument that can be adjusted in height, is easy to use, and has strong practicality. CN201922298567 is a self-service vision detection device. This solution can perform detection autonomously without manual processing, reducing the cost of vision detection.

[0005] However, the above detection devices still have certain defects: (1) They are not suitable for special groups of people. The above devices still require cooperation between both parties when testing, and are not suitable for the deaf-mute, infants, pseudo-blind people and other groups; (2) They are not portable. The detection device is large in size and inconvenient to carry, and is not suitable for people with limited mobility; (3) The detection basis is not objective. The existing device detection basis is still based on the letter vision chart, which requires the subjective cooperation of the subjects and will be affected by the test environment and the test subject's own factors, resulting in subjective results. Summary of the invention

[0006] The purpose of the present invention is to provide a wearable vision detection device based on EEG signals to overcome the problems existing in the prior art.

[0007] In order to achieve the above tasks, the present invention adopts the following technical solutions:

[0008] A wearable vision detection device based on EEG signals, comprising a portable EEG acquisition device, an EEG recognition module and a vision detection module;

[0009] The portable EEG acquisition device is used to play the stimulation paradigm video and collect EEG signals from the subject's brain at the same time; the EEG recognition module is used to control the video playback of the portable EEG acquisition device and to verify and analyze the EEG signals; the vision detection module performs preprocessing, feature analysis and extraction, feature fusion and model detection on the EEG signals;

[0010] The fusion features of the subject are detected through a pre-built model to obtain the subject's visual range.

[0011] Furthermore, the portable EEG acquisition device is composed of two acquisition units, a display unit and a strap; one of the acquisition units is an EEG acquisition electrode, which is placed on the forehead of the subject and close to the forehead to collect the EEG signal of the subject; the other acquisition unit is clamped to the earlobe of the subject as a reference electrode, and the display unit is used to display the stimulation paradigm video. The strap can be retracted to meet the different head sizes of the subjects.

[0012] Furthermore, the EEG recognition module includes a TGAM chip and an STM32 main control chip; wherein the STM32 main control chip controls the display unit to play a stimulation paradigm video, and the stimulation paradigm video stimulates the subject to generate an EEG signal; at the same time, the acquisition unit acquires the EEG signal of the subject and transmits it to the TGAM chip; the TGAM chip filters out the influence of environmental noise on the EEG signal, then amplifies it, performs ADC digital-to-analog conversion, obtains a digitized EEG signal, and then determines whether the signal checksum is correct, if the checksum is correct, continues to analyze the signal, if the checksum is wrong, abandons the signal; finally, the TGAM chip transmits the acquired valid digital EEG signal to the STM32 main control chip; the STM32 main control chip transmits the EEG signal to the vision detection module 3 wirelessly.

[0013] Furthermore, the vision detection module pre-processes the EEG signal, including coarse processing and fine processing;

[0014] In the rough process:

[0015] The received EEG signal is used as the original signal X to filter the high-frequency signal through a bandpass filter, and the low-frequency signal F(x) is extracted; the mode L=[L1,L2] of the extreme values ​​of the low-frequency signal F(x) is calculated, where L1 is the mode of the maximum value and L2 is the mode of the minimum value; then, F(x) is segmented by a sliding window, and it is determined whether the maximum value and the minimum value of each data point in the signal within the segmented sliding window are greater than the preset multiples of L1 and L2 respectively; if greater, the data point is discarded, and if less than or equal to, it is retained;

[0016] In detail processing:

[0017] (a) First, find all the maximum and minimum values ​​of the original signal X, and construct the envelope of the maximum value y1(x) and the envelope of the minimum value y2(x);

[0018] (b) Find the mean a of y1(x) and y2(x), and construct the envelope g(a) of the mean;

[0019] (c) Calculate the mean difference signal h=Xg(a), use a sliding window to divide h, and determine whether the difference of the extreme values ​​in each sliding window after division is equal to 0. If it does not meet the condition of being equal to 0, then use the mean difference signal h as the original signal X to repeat the operations (a) to (b); if it meets the condition of being equal to 0, then subtract the mean difference signal h from the original signal X to obtain the signal N;

[0020] (d) Determine whether the signal N is equal to 0. If not, repeat the operations (a) to (c) with N as the original signal X. If N is equal to 0, it is regarded as an ad component.

[0021] (e) Subtract ad from the original signal X and repeat (a) to (d) as the new original signal until the signal is decomposed to only one maximum value or one minimum value;

[0022] Finally, after the signal is decomposed, n ad components will be obtained; then all ad components are constructed into a set AD = [ad1, ad2, ad3, ..., ad n ]; Then, the set AD and the low-frequency signal F(x) are correlated to find the correlation coefficient R = [r1, r2, r3, ..., r n ];

[0023] (f) Setting the threshold S = R / n, where n represents n correlation coefficients; determining r n Is it less than the threshold S? If it is greater, it is discarded; if it is less, r i Corresponding ad i The component signals are added to reconstruct the output denoised signal DS.

[0024] Furthermore, the vision detection module performs feature analysis and extraction, including extraction of time domain features Ft and frequency domain features Ff:

[0025] In the time domain feature extraction, DS is segmented by a sliding window, and then a superposition signal DS(x) is constructed, where DS(x) = DS1 + DS2 + ... + DS n ; The signal of the sliding window at the nth second is represented as DS(n); then the maximum value max and the minimum value min of DS(x) are calculated, and the average value (min+max) / 2 is calculated as the time domain feature Ft;

[0026] In frequency domain feature extraction, the denoised signal DS is first subjected to Fourier transform to obtain the signal F(ω), and multiple preset sub-bands are extracted from F(ω) through a bandpass filter; then, the weighted curve area Area of ​​each 1s is calculated for each of the multiple sub-bands;

[0027] Then, the changing frequency value freq of all weighted curve areas Area of ​​each sub-frequency band is calculated. Finally, each frequency band will get n freqs, which are recorded as freq = [freq1, freq2, ..., freq n ], where freq n =1 / (Area n+1 -Area n ), Area n It is expressed as the weighted curve area at the nth second; based on this, all freqs of all sub-bands are taken as frequency domain features Ff.

[0028] Furthermore, after obtaining the time domain features and frequency domain features, a feature matrix is ​​constructed, where:

[0029] A row of the feature matrix corresponds to the time domain features and frequency domain features of a sample, which are listed as feature categories. After constructing the feature matrix, label each sample in the feature matrix according to the measured visual acuity range of the sample. One label corresponds to the corresponding visual acuity range. The labels are 1, 2, 3, and 4. 1-visual acuity is 0 to 2.0D, 2-visual acuity is 2.1 to 4.0D, 3-visual acuity is 4.1 to 6.0D, and 4-visual acuity is 6.1 to 8.0D. Among them, D is the unit of visual acuity.

[0030] For all features of each sample in the feature matrix, a Pearson correlation analysis is performed between each feature and the label corresponding to the sample, and the time domain features and frequency domain features with correlation coefficients less than 0.05 are screened out. The screened features are fused at the feature level to obtain the fused feature F(tf).

[0031] Furthermore, the pre-built model includes:

[0032] First, the standard deviation stdf and the average af of F(tf) with the same label are calculated, thereby obtaining the range matrix [stdf af] of the label; at the same time, the covariance matrix S of F(tf) with the same label is constructed m ; M labels will result in M ​​range matrices and M covariance matrices; subsequently, the M range matrices constitute the small model and the M covariance matrices constitute the large model.

[0033] Furthermore, the fusion features of the subject are detected by a pre-built model to obtain the visual range of the subject, including:

[0034] When a subject measures an EEG signal, the EEG signal of the subject is preprocessed and subjected to feature analysis and extraction, and feature fusion to obtain the fusion feature of the EEG signal of the subject, which is recorded as F; then, the feature F is first put into the small model for matching; if the value of F belongs to a range in a certain range matrix, the visual range corresponding to the label of the range matrix is ​​the visual range detected by the subject;

[0035] If not, put the feature F into the large model and calculate the covariance matrix S of the feature and each m Mahalanobis distance; find the feature F and all covariance matrices S m The covariance matrix S with the smallest Mahalanobis distance m , then the visual acuity range corresponding to the label of the covariance matrix is ​​the visual acuity range detected by the subject.

[0036] Compared with the prior art, the present invention has the following technical features:

[0037] 1. It is applicable to a wide range of people. It can be used for testing the general population, people with communication disorders, infants and young children, etc.

[0038] 2. The device is portable and practical. The device is wearable and can be used in not only conventional occasions, such as hospitals, schools, and communities, but also in unconventional occasions, such as homes;

[0039] 3. Less affected by the outside world. The detection is based on EEG signals and the video playback source of the device is sealed, which is not affected by the ambient light, sound, detection distance, or the subject's own factors. For example, the subject guesses the vision chart through memory.

[0040] 4. Based on the objective evaluation system based on visual-brainwave response, a visual function evaluation system independent of the visual acuity chart system based on visual resolution-cognition-subjective expression can be established to redefine the basic concept of vision. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a schematic diagram of the system framework of the present invention;

[0042] Figure 2 is a flow chart of vision detection of the present invention;

[0043] Figure 3 It is the internal principle block diagram of the device of the present invention;

[0044] Figure 4 This is a physical picture of the vision detection device;

[0045] Figure 5 It is a processing flow chart of the vision detection module;

[0046] Figure 6 It is a schematic diagram of the signal preprocessing process;

[0047] Figure 7 Schematic diagram of feature analysis and extraction;

[0048] Figure 8 Schematic diagram of the model training process;

[0049] Fig. 9 Schematic diagram of the process flow actually applied by the subjects. DETAILED DESCRIPTION

[0050] The present invention provides a portable and wearable vision detection device based on EEG signals. The subject wears the vision detection device on his head and is prompted to generate corresponding EEG signals by watching a stimulation paradigm video. The EEG signals are initially identified and processed and further transmitted to a host computer, which processes the data to obtain the subject's vision. The device is suitable for all kinds of people, has an objective detection basis, is more portable, and has strong practicality.

[0051] The system block diagram of the present invention is as follows: Figure 1 As shown, it includes a portable EEG acquisition device 1, an EEG recognition module 2 and a vision detection module 3; the portable EEG acquisition device 1 is used to play the stimulation paradigm video and collect EEG signals from the subject's brain at the same time, the EEG recognition module 2 is used to control the video playback of the portable EEG acquisition device, and to verify and analyze the EEG signals, and the vision detection module 3 further analyzes the EEG signals and obtains the subject's vision. Among them, the stimulation paradigm video is composed of black pictures, white pictures, black and white grid pictures, white and black grid pictures, clear black and white grid pictures and fuzzy black and white grid pictures displayed alternately.

[0052] like Figure 4 As shown, the portable EEG acquisition device 1 is composed of acquisition units 11 and 12, a display unit 13 and a strap 14; the acquisition unit 11 is an EEG acquisition electrode, which is placed on the forehead of the subject and is close to the forehead; the acquisition unit 11 is used to acquire the EEG signal of the subject, the acquisition unit 12 is clamped to the earlobe of the subject as a reference electrode, the display unit 13 is used to display the stimulation paradigm video, and the strap 14 can be retracted to meet the different head sizes of the subjects.

[0053] like Figure 3 As shown, the EEG recognition module 2 includes a TGAM chip and an STM32 main control chip; wherein, the power supply supplies power to the TGAM chip and the STM32 main control chip, and the STM32 main control chip controls the display unit 13 to play the stimulation paradigm video, and the stimulation paradigm video stimulates the subject to generate an EEG signal; at the same time, the acquisition unit 11 acquires the EEG signal of the subject and transmits it to the TGAM chip; the TGAM chip filters the EEG signal to remove the influence of environmental noise, then amplifies it, and performs ADC digital-to-analog conversion to obtain a digitized EEG signal, and then determines whether the signal checksum is correct, and if the checksum is correct, the signal continues to be parsed, and if the checksum is wrong, the signal is abandoned. Finally, the TGAM chip transmits the collected effective and reliable digital EEG signal to the STM32 main control chip; the STM32 main control chip transmits the EEG signal to the vision detection module 3 through the Bluetooth module, and performs the next step of processing on the EEG signal, such as preprocessing, feature extraction, and classification. The analysis result is finally output.

[0054] The detection process of the present invention is as follows Figure 2As shown, after the subject wipes the earlobes and forehead clean, he can wear the vision detection device. The examiner checks whether the device is worn correctly and whether there are any interferences such as sweat and hair at the collection site. After the inspection, the subject watches the stimulation paradigm video and starts to collect EEG signals; after watching, the EEG signal collection stops. The device transmits the collected raw EEG signals to the vision detection module 3 via Bluetooth and analyzes them, and finally obtains the test results.

[0055] The specific process of processing EEG signals by the vision detection module 3 of the present invention is as follows: Figure 5 As shown in the figure, the processing flow of the vision detection module is divided into three steps, namely preprocessing, feature analysis and extraction, and model detection. Preprocessing is to denoise and remove artifacts from the EEG signals sent by the STM32 main control chip of the EEG recognition module 2; feature analysis and extraction is to extract features from the preprocessed EEG signals, and model detection is to classify the extracted features and obtain results, among which:

[0056] Pre-processing process such as Figure 6 As shown, signal preprocessing is divided into coarse processing and fine processing.

[0057] In the rough process:

[0058] The received EEG signal is used as the original signal X to filter the high-frequency signal through a bandpass filter to extract the low-frequency signal F(x); the mode L=[L1,L2] of the extreme values ​​of the low-frequency signal F(x) is calculated, where L1 is the mode of the maximum value and L2 is the mode of the minimum value; then, F(x) is segmented by a sliding window of length 1s, and it is determined whether the maximum and minimum values ​​of each data point in the signal within the segmented sliding window are greater than 1.5 times of L1 and L2 respectively; if greater, the data point is discarded, and if less than or equal to, it is retained.

[0059] In detail processing:

[0060] (a) First, find all the maximum and minimum values ​​of the original signal X, and construct the envelope of the maximum value y1(x) and the envelope of the minimum value y2(x);

[0061] (b) Find the mean a of y1(x) and y2(x), and construct the envelope g(a) of the mean;

[0062] (c) Calculate the mean difference signal h = Xg(a), use a 1s sliding window to divide h, and determine whether the difference of the extreme values ​​(maximum value - minimum value) in each sliding window after division is equal to 0. If it does not meet the condition of being equal to 0, repeat the operations (a) to (b) with the mean difference signal h as the original signal X; if it meets the condition of being equal to 0, subtract the mean difference signal h from the original signal X to obtain the signal N;

[0063] (d) Determine whether the signal N is equal to 0. If not, repeat the operations (a) to (c) with N as the original signal X. If N is equal to 0, it is regarded as an ad component.

[0064] (e) Subtract ad from the original signal X and repeat (a) to (d) as the new original signal until the signal is decomposed to a point where only one maximum or one minimum exists.

[0065] Finally, after the signal is decomposed, n ad components will be obtained; then all ad components are constructed into a set AD = [ad1, ad2, ad3, ..., ad n ]; Then, the set AD and the low-frequency signal F(x) are correlated to find the correlation coefficient R = [r1, r2, r3, ..., r n ]; Correlation coefficient in, is the average value of the low-frequency signal F(x), is the nth AD component signal ad i The average value of .

[0066] (f) Setting the threshold S = R / n, where n represents n correlation coefficients; determining r n Is it less than the threshold S? If it is greater, it is discarded; if it is less, r i Corresponding ad i The component signals are added to reconstruct the output denoised signal DS.

[0067] The specific process of feature analysis and extraction of the present invention is as follows Figure 7 As shown, in feature analysis and extraction, the denoised signal DS is respectively subjected to time domain feature Ft extraction and frequency domain feature Ff extraction:

[0068] In the time domain feature extraction, a sliding window of 1s is set to segment DS, and then construct the superposition signal DS(x), where DS(x)=DS1+DS2+...+DS n ; The signal of the sliding window at the nth second is represented as DS(n); then the maximum value max and the minimum value min of DS(x) are obtained, and its average value (min+max) / 2 is calculated as the time domain feature Ft.

[0069] In the frequency domain feature extraction, the denoised signal DS is first Fourier transformed to obtain the signal F(ω), and four sub-frequency bands f(δ), f(θ), f(α), and f(β) are extracted from F(ω) through a bandpass filter; among them, the frequency band of f(δ) is 0.5-4Hz, the frequency band of f(θ) is 4-7Hz, the frequency band of f(α) is 8-12Hz, and the frequency band of f(β) is 12-28Hz; then the weighted curve area Area of ​​each 1s is calculated for the four sub-frequency bands, and the weighted curve area of ​​the i-th second is expressed as:

[0070] a takes δ, θ, α, β; i takes 1, 2,..., n;

[0071] std(f(a)) is the preset weight, which is expressed as the standard deviation value of the sub-band f(a). Represents the curve area of ​​the ith second sub-band f(a).

[0072] Then, the changing frequency value freq of all weighted curve areas Area of ​​each sub-frequency band is calculated. Finally, each frequency band will get n freqs, which are recorded as freq = [freq1, freq2, ..., freq n ], where freq n =1 / (Area n+1 -Area n ), Area n It is expressed as the weighted curve area at the nth second; based on this, all the freqs of the four sub-bands are taken as the frequency domain features Ff.

[0073] After obtaining the time domain and frequency domain features, construct the feature matrix, where:

[0074] A row of the feature matrix corresponds to the time domain features and frequency domain features of a sample, which are listed as feature categories. After constructing the feature matrix, each sample in the feature matrix is ​​labeled according to the visual acuity range of the sample. Among them, one label corresponds to the corresponding visual acuity range. The labels are 1, 2, 3, and 4. 1-visual acuity is 0-2.0D, 2-visual acuity is 2.1-4.0D, 3-visual acuity is 4.1-6.0D, and 4-visual acuity is 6.1-8.0D. Among them, D is the unit of visual acuity.

[0075] For all features of each sample in the feature matrix (including time domain features and frequency domain features), a Pearson correlation analysis is performed on each feature and the label corresponding to the sample, and the time domain features and frequency domain features with a correlation coefficient P less than 0.05 are screened out. The screened features are then fused at the feature level, that is, the time domain feature Ft and the frequency domain feature Ff are spliced ​​at the same scale to obtain the fused feature F(tf).

[0076] The specific process of model construction of the present invention is as follows Figure 8 As shown, in model construction:

[0077] First, the standard deviation stdf and the average af of F(tf) with the same label are calculated, thereby obtaining the range matrix [stdf af] of the label; at the same time, the covariance matrix S of F(tf) with the same label is constructed m . The 4 labels will result in 4 range matrices and 4 covariance matrices; then, the 4 range matrices form the small model and the 4 covariance matrices form the large model.

[0078] The specific flow chart of the actual application of the subject of the present invention is as follows Fig. 9 As shown, in practical applications,

[0079] When a subject measures an EEG signal, the subject's EEG signal will undergo preprocessing, feature analysis and extraction, and feature fusion to obtain the fusion feature of the subject's EEG signal, denoted as F; then, the feature F is first placed in the small model for matching; if the value of F belongs to a range in a certain range matrix, the visual range corresponding to the label of the range matrix is ​​the visual range detected by the subject.

[0080] If not, put the feature F into the large model and calculate the covariance matrix S of the feature and the four m The Mahalanobis distance of F is the characteristic of the subject, F af is the mean of feature F, and S is the covariance of feature F. Find the covariance matrix S of feature F and 4 m The covariance matrix S with the smallest Mahalanobis distance m , then the visual acuity range corresponding to the label of the covariance matrix is ​​the visual acuity range detected by the subject.

[0081] The vision detection of the present invention is based on electroencephalogram signals, which are signals that reflect the objective physiological state of the human brain. They contain a large amount of physiological and disease information, including information about human vision, and the video playback source of the device is sealed. Therefore, the device is not affected by the ambient light, sound, and detection distance, nor by factors of the subject himself. The device is also suitable for a wide range of people and is more portable. It can be used for examinations of the general population, as well as for people with communication disorders, infants, and pseudo-blind patients. Therefore, the present invention has the advantages of being widely applicable to a wide range of people, being portable, and being less affected by the external environment.

[0082] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A wearable vision detection device based on EEG signals, characterized in that: Including portable EEG acquisition equipment, EEG recognition module and vision detection module; The portable EEG acquisition device is used to play the stimulation paradigm video and collect EEG signals from the subject's brain at the same time; the EEG recognition module is used to control the video playback of the portable EEG acquisition device and to verify and analyze the EEG signals; the vision detection module performs preprocessing, feature analysis and extraction, feature fusion and model detection on the EEG signals; Among them, the fusion features of the subject are detected through the pre-built model to obtain the subject's visual range; The vision detection module pre-processes the EEG signal, including coarse processing and fine processing; In rough processing: The received EEG signal is used as the original signal X to filter the high-frequency signal through a bandpass filter, and the low-frequency signal F(x) is extracted; the mode L=[L1,L2] of the extreme values ​​of the low-frequency signal F(x) is calculated, where L1 is the mode of the maximum value and L2 is the mode of the minimum value; then, F(x) is segmented by a sliding window, and it is determined whether the maximum value and the minimum value of each data point in the signal within the segmented sliding window are greater than the preset multiples of L1 and L2 respectively; if greater, the data point is discarded, and if less than or equal to, it is retained; In detail processing: (a) First, find all the maximum and minimum values ​​of the original signal X, and construct the envelope of the maximum value y1(x) and the envelope of the minimum value y2(x); (b) Find the mean a of y1(x) and y2(x), and construct the envelope g(a) of the mean; (c) Calculate the mean difference signal h=Xg(a), use a sliding window to divide h, and determine whether the difference of the extreme values ​​in each sliding window after division is equal to 0. If it does not meet the condition of being equal to 0, then use the mean difference signal h as the original signal X to repeat the operations (a) to (b); if it meets the condition of being equal to 0, then subtract the mean difference signal h from the original signal X to obtain the signal N; (d) Determine whether the signal N is equal to 0. If not, repeat the operations (a) to (c) with N as the original signal X. If N is equal to 0, it is regarded as an ad component. (e) Subtract ad from the original signal X and repeat (a) to (d) as the new original signal until the signal is decomposed to only one maximum value or one minimum value; Finally, after the signal is decomposed, n ad components will be obtained; then all ad components are constructed into a set AD = [ad1, ad2, ad3, ..., ad n ]; Then, the set AD and the low-frequency signal F(x) are correlated to find the correlation coefficient R = [r1, r2, r3, ..., r n ]; (f) Setting the threshold S = R / n, where n represents n correlation coefficients; determining r n Is it less than the threshold S? If it is greater, it is discarded; if it is less, r i Corresponding ad i The component signals are added to reconstruct the output denoised signal DS.

2. The wearable vision detection device based on EEG signals according to claim 1 is characterized in that: The portable EEG acquisition device is composed of two acquisition units, a display unit and a strap; one of the acquisition units is an EEG acquisition electrode, which is placed on the forehead of the subject and close to the forehead to collect the EEG signal of the subject; the other acquisition unit is clamped to the earlobe of the subject as a reference electrode, and the display unit is used to display the stimulation paradigm video. The strap can be retracted to meet the different head sizes of the subjects.

3. The wearable vision detection device based on EEG signals according to claim 1 is characterized in that: The EEG recognition module includes a TGAM chip and an STM32 main control chip; wherein the STM32 main control chip controls the display unit to play a stimulation paradigm video, and the stimulation paradigm video stimulates the subject to generate an EEG signal; at the same time, the acquisition unit acquires the EEG signal of the subject and transmits it to the TGAM chip; the TGAM chip filters the EEG signal to remove the influence of environmental noise, then amplifies it, performs ADC digital-to-analog conversion, obtains a digitized EEG signal, and then determines whether the signal checksum is correct, if the checksum is correct, continues to analyze the signal, if the checksum is wrong, abandons the signal; finally, the TGAM chip transmits the acquired valid digital EEG signal to the STM32 main control chip; the STM32 main control chip transmits the EEG signal to the vision detection module wirelessly.

4. The wearable vision detection device based on EEG signals according to claim 1 is characterized in that: The vision detection module performs feature analysis and extraction, including extraction of time domain features Ft and frequency domain features Ff: In the time domain feature extraction, DS is segmented using a sliding window, and then a superposition signal DS(x) is constructed, where DS(x) = DS1 + DS2 + ... + DS n ; The signal of the sliding window at the nth second is represented as DS(n); then the maximum value max and the minimum value min of DS(x) are calculated, and the average value (min+max) / 2 is calculated as the time domain feature Ft; In frequency domain feature extraction, the denoised signal DS is first subjected to Fourier transform to obtain the signal F(ω), and multiple preset sub-bands are extracted from F(ω) through a bandpass filter; then, the weighted curve area Area of ​​each 1s is calculated for each of the multiple sub-bands; Then, the changing frequency value freq of all weighted curve areas Area of ​​each sub-frequency band is calculated. Finally, each frequency band will get n freqs, which are recorded as freq = [freq1, freq2, ..., freq n ], where freq n =1 / (Area n+1 -Area n ), Area n It is expressed as the weighted curve area at the nth second; based on this, all freqs of all sub-bands are taken as frequency domain features Ff.

5. The wearable vision detection device based on EEG signals according to claim 1 is characterized in that: After obtaining the time domain features and frequency domain features, construct the feature matrix, where: A row of the feature matrix corresponds to the time domain features and frequency domain features of a sample, which are listed as feature categories. After constructing the feature matrix, each sample in the feature matrix is ​​labeled according to the visual acuity range of the sample. One label corresponds to the corresponding visual acuity range. The labels are 1, 2, 3, and 4. 1-visual acuity is 0 to 2.0D, 2-visual acuity is 2.1 to 4.0D, 3-visual acuity is 4.1 to 6.0D, and 4-visual acuity is 6.1 to 8.0D. D is the unit of visual acuity. For all features of each sample in the feature matrix, a Pearson correlation analysis is performed between each feature and the label corresponding to the sample, and the time domain features and frequency domain features with correlation coefficients less than 0.05 are screened out. The screened features are fused at the feature level to obtain the fused feature F(tf).

6. The wearable vision detection device based on EEG signals according to claim 1 is characterized in that: The pre-built model includes: First, the standard deviation stdf and the average af of F(tf) with the same label are calculated, thereby obtaining the range matrix [stdf af] of the label; at the same time, the covariance matrix S of F(tf) with the same label is constructed m ; M labels will result in M ​​range matrices and M covariance matrices; subsequently, the M range matrices constitute the small model and the M covariance matrices constitute the large model.

7. The wearable vision detection device based on EEG signals according to claim 1, characterized in that: The fusion features of the subject are detected by a pre-built model to obtain the visual range of the subject, including: When a subject measures an EEG signal, the EEG signal of the subject is preprocessed and subjected to feature analysis and extraction, and feature fusion to obtain the fusion feature of the EEG signal of the subject, which is recorded as F; then, the feature F is first put into the small model for matching; if the value of F belongs to a range in a certain range matrix, the visual range corresponding to the label of the range matrix is ​​the visual range detected by the subject; If not, put the feature F into the large model and calculate the covariance matrix S of the feature and each m Mahalanobis distance; find the feature F and all covariance matrices S m The covariance matrix S with the smallest Mahalanobis distance m , then the visual acuity range corresponding to the label of the covariance matrix is ​​the visual acuity range detected by the subject.

Images