An artificial intelligence-based electroencephalogram signal image recognition method and system

Abstract

The present invention relates to the field of image processing technology, and in particular to an artificial intelligence-based EEG signal image recognition method and system, the method comprising: sequentially denoising, cleaning, filtering, and feature extraction of collected EEG signal data to output EEG signal features; using the labeled EEG signal data and the EEG signal features to train a deep learning model and update it in real time, using the trained deep learning model to recognize EEG signal images to output recognition results; determining the recognition accuracy of the EEG signal image based on the area ratio of the distorted region of the EEG signal image; if the recognition accuracy of the EEG signal image does not meet the requirements, adjusting the sampling and holding time of the EEG signal data; if the recognition effectiveness of the EEG signal image does not meet the requirements, adjusting the cutoff frequency of the filter. The present invention improves the recognition accuracy of EEG signal images.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an artificial intelligence-based electroencephalogram (EEG) signal image recognition method and system. Background Art

[0002] With the rapid development of neuroscience, brain-computer interface technology, and artificial intelligence, the accurate analysis and interpretation of electroencephalogram (EEG) signals have become increasingly important. As the electrophysiological manifestation of brain neural activity, EEG signals contain a wealth of information, such as the brain's cognitive state, emotional changes, sleep stages, and pathological characteristics. Traditional EEG signal processing methods mainly rely on manual feature extraction and simple statistical analysis. However, these methods have many limitations when faced with complex and changing EEG signals. In addition, traditional methods lack effective adaptive mechanisms for dealing with problems such as noise interference and signal distortion, which significantly affects the accuracy and reliability of EEG signal analysis.

[0003] Chinese Patent Publication No.: CN118097281A discloses a method and system for recognizing EEG motion images based on optimal deep learning. The method includes the following steps: obtaining EEG motion images; inputting the EEG motion images into a stacked sparse autoencoder model for feature extraction, and outputting a feature matrix of the EEG signals; inputting the feature matrix into a deep wavelet neural network classification model for classification, and optimizing the weight coefficients and bias values ​​of the deep wavelet neural network classification model using a chaotic dragonfly algorithm to obtain EEG motion image classification results. It can be seen that the method and system for recognizing EEG motion images based on optimal deep learning have the problem that the gain parameter of the transistor in the amplifier increases, resulting in signal saturation and loss of signal details, thereby reducing the recognition accuracy of EEG signal images. Summary of the Invention

[0004] To this end, the present invention provides an artificial intelligence-based EEG signal image recognition method and system to overcome the problem in the prior art that the gain parameter of the transistor in the amplifier increases, leading to signal saturation and loss of signal details, thereby reducing the recognition accuracy of EEG signal images.

[0005] To achieve the above-mentioned objectives, the present invention provides an artificial intelligence-based EEG signal image recognition method, comprising: denoising, cleaning, filtering and feature extraction of the collected EEG signal data in sequence to output EEG signal features, and labeling the EEG signal data according to known labeling information; using the labeled EEG signal data and the EEG signal features to train and update a deep learning model in real time, and using the trained deep learning model to recognize the EEG signal image to output a recognition result; respectively obtaining the distorted area of ​​the recognized EEG signal image and the total area of ​​the EEG signal image; determining the recognition accuracy of the EEG signal image based on the proportion of the distorted area of ​​the EEG signal image; if the recognition accuracy of the EEG signal image does not meet the requirements, adjusting the sampling and holding time of the EEG signal data, or determining the recognition validity of the EEG signal image based on the packet loss rate of the EEG signal data; if the recognition validity of the EEG signal image does not meet the requirements, adjusting the cutoff frequency of the filter, or adjusting the learning rate of the deep learning model based on the accuracy of the deep learning model in recognizing the EEG signal image.

[0006] Furthermore, determining the recognition accuracy of the EEG signal image includes:

[0007] Comparing the area ratio of the distorted region of the EEG signal image with a preset first ratio;

[0008] If the distorted region area ratio of the EEG signal image is greater than the preset first ratio, it is determined that the recognition accuracy of the EEG signal image does not meet the requirements.

[0009] Furthermore, determining the recognition validity of the EEG signal image includes:

[0010] Comparing the area ratio of the distorted region of the EEG signal image with the preset first ratio and the preset second ratio respectively;

[0011] If the area ratio of the distorted region of the EEG signal image is greater than the preset first ratio and less than or equal to the preset second ratio, it is preliminarily determined that the recognition validity of the EEG signal image does not meet the requirements, and whether the recognition validity of the EEG signal image meets the requirements is determined based on the packet loss rate of the EEG signal data.

[0012] Furthermore, the sampling and holding time of the EEG signal data is adjusted, including:

[0013] Comparing the area ratio of the distorted region of the EEG signal image with the preset second ratio;

[0014] If the distorted area ratio of the EEG signal image is greater than a preset second ratio, the sampling and holding time of the EEG signal data is reduced.

[0015] Furthermore, the reduction range of the sampling and holding time of the electroencephalogram signal data is determined by the difference between the area ratio of the distorted region of the electroencephalogram signal image and a preset second ratio.

[0016] Furthermore, adjusting the cutoff frequency of the filter includes:

[0017] Comparing the packet loss rate of the EEG signal data with a preset first packet loss rate and a preset second packet loss rate respectively;

[0018] If the packet loss rate of the EEG signal data is greater than the preset first packet loss rate, determining that the recognition validity of the EEG signal image does not meet the requirements;

[0019] If the packet loss rate of the EEG signal data is greater than a preset first packet loss rate and less than or equal to a preset second packet loss rate, reducing the cutoff frequency of the filter;

[0020] If the packet loss rate of the EEG signal data is greater than the preset second packet loss rate, it is preliminarily determined that the training effectiveness of the deep learning model does not meet the requirements, and whether the training effectiveness of the deep learning model meets the requirements is determined based on the accuracy of the deep learning model in identifying the EEG signal image.

[0021] Furthermore, the reduction range of the cutoff frequency of the filter is determined by the difference between the packet loss rate of the EEG signal data and a preset first packet loss rate.

[0022] Furthermore, adjusting the learning rate of the deep learning model includes:

[0023] Compare the accuracy of the deep learning model in recognizing EEG signal images with the preset accuracy;

[0024] If the accuracy of the deep learning model in recognizing the EEG signal image is less than or equal to the preset accuracy, it is determined that the training effectiveness of the deep learning model does not meet the requirements, and the learning rate of the deep learning model is reduced.

[0025] Furthermore, the reduction range of the learning rate of the deep learning model is determined by the difference between a preset accuracy rate and the accuracy rate of the deep learning model in recognizing the EEG signal image.

[0026] The present invention also provides an artificial intelligence-based EEG signal image recognition system, comprising:

[0027] A data acquisition module, used to collect EEG signal data;

[0028] a data processing module connected to the data acquisition module, comprising a preprocessing unit for preprocessing the EEG signal data to output optimized data, a feature extraction unit connected to the preprocessing unit for extracting features from the optimized data to output EEG signal features, and a labeling unit for labeling the EEG signal data according to known labeling information;

[0029] A model training module, connected to the data processing module, comprising a model training unit for training a deep learning model based on the labeled EEG signal data and the EEG signal features, and a model updating unit connected to the model training unit for updating the deep learning model in real time;

[0030] An image recognition module, connected to the model training module, is used to recognize EEG signal images using the trained deep learning model to output recognition results;

[0031] a storage module, which is respectively connected to the data acquisition module, the data processing module, the model training module, and the image recognition module, and is used to store the EEG signal data, the optimization data, the EEG signal features, the known labeling information, the labeled EEG signal data, the deep learning model, and the recognition results;

[0032] A control module is respectively connected to the data acquisition module, the data processing module, the model training module, the image recognition module and the storage module, and is used to determine the sampling and holding time of the EEG signal data according to the area ratio of the distorted region of the EEG signal image, or to determine the cutoff frequency of the filter according to the packet loss rate of the EEG signal data, and to determine the learning rate of the deep learning model according to the accuracy of the deep learning model in identifying the EEG signal image.

[0033] Compared with the prior art, the beneficial effect of the present invention is that the method of the present invention adjusts the sampling and holding time of the EEG signal data according to the area ratio of the distorted area of ​​the EEG signal image. Since the gain parameter of the transistor in the amplifier may increase with the increase of the use time, the gain of the amplifier may increase, resulting in signal saturation and loss of signal details. By reducing the sampling and holding time of the EEG signal, the proportion of the collected saturated signal can be reduced, and the chance of obtaining the effective signal change part can be increased. The cutoff frequency of the filter is adjusted according to the packet loss rate of the EEG signal data. Since in the environment of collecting EEG signals, the electromagnetic signals emitted by other medical equipment or communication equipment may interfere with the transmission of EEG signals, resulting in data errors or loss of some information, resulting in the marking of It is easy to produce inaccurate results. By reducing the cutoff frequency of the filter, high-frequency interference signals can be filtered out, making the waveform of the EEG signal in the low-frequency band clearer, which helps to improve the accuracy of annotation. The learning rate of the deep learning model is adjusted according to the accuracy of the deep learning model in identifying the EEG signal image. During the EEG signal acquisition process, the collected data may be incomplete due to poor contact between the electrode and the scalp, resulting in some data features being mistakenly deleted during the cleaning process, resulting in incomplete model optimization. By reducing the learning rate of the deep learning model, the model can adjust parameters more finely, reduce the learning of erroneous features, avoid overfitting the model to abnormal features caused by incomplete data, and improve the recognition accuracy of EEG signal images.

[0034] Furthermore, the method of the present invention adjusts the sampling and holding time of the EEG signal data by setting a preset first ratio and a preset second ratio. Since the gain parameter of the transistor in the amplifier may increase with the increase of usage time, the gain of the amplifier will increase, resulting in signal saturation and loss of signal details. By reducing the sampling and holding time of the EEG signal, the proportion of the collected saturated signal can be reduced, and the chance of obtaining the effective signal change part is increased, thereby further improving the recognition accuracy of the EEG signal image.

[0035] Furthermore, the method of the present invention adjusts the cutoff frequency of the filter by setting a preset first packet loss rate and a preset second packet loss rate. Since in the environment of collecting EEG signals, electromagnetic signals emitted by other medical equipment or communication equipment may interfere with the transmission of EEG signals, resulting in data errors or loss of some information, which may easily lead to inaccurate results during labeling. By reducing the cutoff frequency of the filter, high-frequency interference signals can be filtered out, making the waveform of the EEG signal in the low-frequency band clearer, which helps to improve the accuracy of labeling and further improves the recognition accuracy of EEG signal images.

[0036] Furthermore, the method of the present invention adjusts the learning rate of the deep learning model by setting a preset accuracy rate. During the EEG signal acquisition process, the collected data may be incomplete due to poor contact between the electrode and the scalp, resulting in some data features being erroneously deleted during the cleaning process, thereby causing incompleteness in the model optimization stage. By reducing the learning rate of the deep learning model, the model can adjust parameters more finely, reduce the learning of erroneous features, and avoid overfitting of the model to abnormal features caused by incomplete data, thereby further improving the recognition accuracy of EEG signal images. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is an overall flow chart of the method for EEG signal image recognition based on artificial intelligence according to an embodiment of the present invention;

[0038] Figure 2 This is a block diagram of the overall structure of the EEG signal image recognition system based on artificial intelligence according to an embodiment of the present invention;

[0039] Figure 3 This is a logic flow chart of an artificial intelligence-based EEG signal image recognition method according to an embodiment of the present invention;

[0040] Figure 4 This is a specific flow chart of the process of adjusting the sampling and holding time of EEG signal data in the EEG signal image recognition method based on artificial intelligence in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are merely used to explain the present invention and are not intended to limit the present invention.

[0042] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0043] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside", and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0044] Furthermore, it should be noted that, in the description of the present invention, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0045] See also Figure 1 、 Figure 2 、 Figure 3 as well as Figure 4 As shown in the figure, they are respectively the overall flow chart, overall structure block diagram, logic flow chart and specific flow chart of the process of adjusting the sampling and holding time of EEG signal data of the method and system based on artificial intelligence of the embodiment of the present invention. The present invention is an EEG signal image recognition method based on artificial intelligence, comprising:

[0046] Step S1, performing denoising, cleaning, filtering, and feature extraction on the collected EEG signal data in sequence to output EEG signal features, and labeling the EEG signal data according to known label information;

[0047] Step S2: using the labeled EEG signal data and the EEG signal features to train a deep learning model and update it in real time, and using the trained deep learning model to recognize the EEG signal image to output a recognition result;

[0048] Step S3, respectively obtaining the distorted area of ​​the identified EEG signal image and the total area of ​​the EEG signal image;

[0049] Step S4, determining the recognition accuracy of the EEG signal image based on the area ratio of the distorted region of the EEG signal image;

[0050] Step S5: If the recognition accuracy of the EEG signal image does not meet the requirements, the sampling and holding time of the EEG signal data is adjusted, or the recognition validity of the EEG signal image is determined based on the packet loss rate of the EEG signal data;

[0051] Step S6: If the recognition effectiveness of the EEG signal image does not meet the requirements, the cutoff frequency of the filter is adjusted, or the learning rate of the deep learning model is adjusted based on the accuracy of the deep learning model in recognizing the EEG signal image.

[0052] Specifically, the EEG signal data includes the amplitude of the EEG signal, the frequency of the EEG signal, and the potential change of the EEG signal.

[0053] Specifically, EEG signal features include time domain features, frequency domain features, and video features.

[0054] Specifically, known marker information includes sleep stage markers, emotional state markers, and brain disease markers.

[0055] Specifically, the deep learning models are convolutional neural networks, recurrent neural networks, and deep belief networks.

[0056] Specifically, the recognition results include sleep stage recognition, emotional state recognition, and brain disease diagnosis results.

[0057] Specifically, the sampling and holding time of the EEG signal data is the length of time during which the signal samples collected at a certain moment in the EEG signal collection process are held and used for subsequent processing.

[0058] Specifically, the filter includes a high-pass filter, a low-pass filter, and a band-stop filter, and the preferred embodiment is a low-pass filter.

[0059] Specifically, the learning rate of a deep learning model is a hyperparameter used to control the step size of model parameter updates.

[0060] In implementation, the method of the present invention adjusts the sampling and holding time of the EEG signal data according to the area ratio of the distorted region of the EEG signal image. As the use time increases, the gain parameter of the transistor in the amplifier may increase, resulting in an increase in the gain of the amplifier, leading to signal saturation and loss of signal details. By reducing the sampling and holding time of the EEG signal, the proportion of the collected saturated signal can be reduced, and the chance of obtaining the effective signal change part can be increased. The cutoff frequency of the filter is adjusted according to the packet loss rate of the EEG signal data. In the environment of collecting EEG signals, the electromagnetic signals emitted by other medical equipment or communication equipment may interfere with the transmission of EEG signals, resulting in data errors or loss of some information, which makes it easy to produce inaccuracies during labeling. The correct result is that by reducing the cutoff frequency of the filter, high-frequency interference signals can be filtered out, making the waveform of the EEG signal in the low-frequency band clearer, which helps to improve the accuracy of annotation. The learning rate of the deep learning model is adjusted according to the accuracy of the deep learning model in identifying the EEG signal image. During the EEG signal acquisition process, the collected data may be incomplete due to poor contact between the electrode and the scalp, resulting in some data features being mistakenly deleted during the cleaning process, resulting in incomplete model optimization stage. By reducing the learning rate of the deep learning model, the model can adjust parameters more finely, reduce the learning of erroneous features, avoid overfitting of the model to abnormal features caused by incomplete data, and improve the recognition accuracy of EEG signal images.

[0061] Specifically, determining the recognition accuracy of the EEG signal image includes:

[0062] Obtaining the distorted area of ​​the identified EEG signal image and the total area of ​​the EEG signal image respectively, and calculating the area ratio of the distorted area of ​​the EEG signal image;

[0063] Comparing the area ratio of the distorted region of the EEG signal image with a preset first ratio;

[0064] If the distorted region area ratio of the EEG signal image is greater than the preset first ratio, it is determined that the recognition accuracy of the EEG signal image does not meet the requirements.

[0065] Specifically, determining the recognition validity of the EEG signal image includes:

[0066] Comparing the area ratio of the distorted region of the EEG signal image with the preset first ratio and the preset second ratio respectively;

[0067] If the area ratio of the distorted region of the EEG signal image is greater than the preset first ratio and less than or equal to the preset second ratio, it is preliminarily determined that the recognition validity of the EEG signal image does not meet the requirements, and whether the recognition validity of the EEG signal image meets the requirements is determined based on the packet loss rate of the EEG signal data.

[0068] It can be understood that the three intervals divided by the preset first proportion and the preset second proportion correspond to three situations respectively:

[0069] The first interval is when the distorted area ratio of the EEG signal image is less than or equal to the preset first ratio, corresponding to the situation that the recognition accuracy of the EEG signal image meets the requirements;

[0070] The second interval is when the distorted area of ​​the EEG signal image accounts for a greater than the preset first ratio and less than or equal to the preset second ratio. This corresponds to the situation where electromagnetic signals emitted by other medical devices or communication devices in the EEG signal collection environment may interfere with the transmission of the EEG signal, resulting in data errors or loss of some information, which may easily lead to inaccurate results during labeling.

[0071] The third interval is when the area ratio of the distorted region of the EEG signal image is greater than the preset second ratio. The corresponding situation is: as the usage time increases, the gain parameter of the transistor in the amplifier may increase, resulting in an increase in the gain of the amplifier, causing signal saturation and loss of signal details.

[0072] In practice, the range of the preset first ratio is generally selected from [0.05, 0.15], and the range of the preset second ratio is generally selected from [0.16, 0.26].

[0073] Preferably, the preferred embodiment of the preset first ratio is 0.1, and the preferred embodiment of the preset second ratio is 0.2.

[0074] Specifically, the area ratio of the distorted region of the EEG signal image is the ratio of the area of ​​the distorted region of the identified EEG signal image to the total area of ​​the EEG signal image.

[0075] In implementation, the method of the present invention determines the recognition accuracy of the EEG signal image by setting a preset first ratio and a preset second ratio, thereby reducing the impact of the decreased recognition stability of the EEG signal image due to inaccurate determination of the recognition accuracy of the EEG signal image, and further improving the recognition accuracy of the EEG signal image.

[0076] Specifically, adjusting the sampling and holding time of the EEG signal data includes:

[0077] Comparing the area ratio of the distorted region of the EEG signal image with the preset second ratio;

[0078] If the distorted area ratio of the EEG signal image is greater than a preset second ratio, the sampling and holding time of the EEG signal data is reduced.

[0079] Specifically, the reduction range of the sampling and holding time of the electroencephalogram signal data is determined by the difference between the area ratio of the distorted region of the electroencephalogram signal image and a preset second ratio.

[0080] Specifically, when the difference between the area ratio of the distorted area of ​​the EEG signal image and the preset second ratio is within 0.2, the sampling and holding time of the EEG signal data is reduced to 0.95 times the original; when the difference between the area ratio of the distorted area of ​​the EEG signal image and the preset second ratio exceeds 0.2, on the basis of being reduced to 0.95 times the original, the sampling and holding time of the EEG signal data is reduced by 20ms for every 0.1 that exceeds. For example, the difference between the area ratio of the distorted area of ​​the EEG signal image and the preset second ratio is 0.4, the current sampling and holding time of the EEG signal data is 500ms, and the reduced sampling and holding time of the EEG signal data is 500×0.95-20×2=435ms.

[0081] In implementation, the method of the present invention adjusts the sampling and holding time of the EEG signal data by setting a preset first ratio and a preset second ratio. Since the gain parameters of the transistors in the amplifier may increase with the increase of usage time, the gain of the amplifier will increase, resulting in signal saturation and loss of signal details. By reducing the sampling and holding time of the EEG signal, the proportion of the collected saturated signal can be reduced, and the chance of obtaining the effective signal change part is increased, thereby further improving the recognition accuracy of the EEG signal image.

[0082] Specifically, adjusting the cutoff frequency of the filter includes:

[0083] Get the packet loss rate of EEG signal data;

[0084] Comparing the packet loss rate of the EEG signal data with a preset first packet loss rate and a preset second packet loss rate respectively;

[0085] If the packet loss rate of the EEG signal data is greater than the preset first packet loss rate, determining that the recognition validity of the EEG signal image does not meet the requirements;

[0086] If the packet loss rate of the EEG signal data is greater than a preset first packet loss rate and less than or equal to a preset second packet loss rate, reducing the cutoff frequency of the filter;

[0087] If the packet loss rate of the EEG signal data is greater than the preset second packet loss rate, it is preliminarily determined that the training effectiveness of the deep learning model does not meet the requirements, and whether the training effectiveness of the deep learning model meets the requirements is determined based on the accuracy of the deep learning model in identifying the EEG signal image.

[0088] It can be understood that the three intervals divided by the preset first packet loss rate and the preset second packet loss rate correspond to three situations respectively:

[0089] The first interval is when the packet loss rate of the EEG signal data is less than or equal to the preset first packet loss rate, corresponding to the situation that: the recognition effectiveness of the EEG signal image meets the requirements;

[0090] The second interval is when the packet loss rate of the EEG signal data is greater than the preset first packet loss rate and less than or equal to the preset second packet loss rate. This corresponds to the situation where electromagnetic signals emitted by other medical devices or communication devices in the environment where the EEG signals are collected may interfere with the transmission of the EEG signals, resulting in data errors or loss of some information, which may easily lead to inaccurate results during labeling.

[0091] The third interval is when the packet loss rate of the EEG signal data is greater than the preset second packet loss rate. The corresponding situation is: during the EEG signal acquisition process, the collected data may be incomplete due to poor contact between the electrode and the scalp, resulting in some data features being mistakenly deleted during the cleaning process, resulting in incompleteness in the model optimization stage.

[0092] In practice, the preset first packet loss rate is generally selected from the range of [0.03, 0.05], and the preset second packet loss rate is generally selected from the range of [0.06, 0.08].

[0093] Preferably, the preferred embodiment of the preset second packet loss rate is 0.04, and the preferred embodiment of the preset second packet loss rate is 0.07.

[0094] In implementation, the method of the present invention determines the recognition effectiveness of the EEG signal image by setting a preset first packet loss rate and a preset second packet loss rate, thereby reducing the impact of the decrease in the recognition accuracy of the EEG signal image due to inaccurate determination of the recognition effectiveness of the EEG signal image, and further improving the recognition accuracy of the EEG signal image.

[0095] Specifically, the reduction range of the cutoff frequency of the filter is determined by the difference between the packet loss rate of the EEG signal data and a preset first packet loss rate.

[0096] Specifically, when the difference between the packet loss rate of the EEG signal data and the preset first packet loss rate is within 0.03, the cutoff frequency of the filter is reduced to 0.92 times the original value; when the difference between the packet loss rate of the EEG signal data and the preset first packet loss rate exceeds 0.03, on the basis of being reduced to 0.92 times the original value, the cutoff frequency of the filter is reduced by 5Hz for every 0.02 that exceeds it. For example, the difference between the packet loss rate of the EEG signal data and the preset first packet loss rate is 0.07, the current cutoff frequency of the filter is 100Hz, and the reduced cutoff frequency of the filter is 100×0.92-5×2=82Hz.

[0097] In implementation, the method of the present invention adjusts the cutoff frequency of the filter by setting a preset first packet loss rate and a preset second packet loss rate. Since in the environment of collecting EEG signals, electromagnetic signals emitted by other medical equipment or communication equipment may interfere with the transmission of EEG signals, resulting in data errors or loss of some information, which may easily lead to inaccurate results during labeling. By reducing the cutoff frequency of the filter, high-frequency interference signals can be filtered out, making the waveform of the EEG signal in the low-frequency band clearer, which helps to improve the accuracy of labeling and further improves the recognition accuracy of EEG signal images.

[0098] Specifically, adjusting the learning rate of the deep learning model includes:

[0099] Obtain the number of times the deep learning model accurately identifies the EEG signal image and the total number of times the EEG signal image is identified, and calculate the accuracy of the deep learning model in identifying the EEG signal image;

[0100] Comparing the accuracy of the deep learning model in recognizing the EEG signal image with a preset accuracy rate;

[0101] If the accuracy of the deep learning model in recognizing the EEG signal image is less than or equal to the preset accuracy, it is determined that the training effectiveness of the deep learning model does not meet the requirements, and the learning rate of the deep learning model is reduced.

[0102] It is understandable that the two intervals of the preset accuracy rate correspond to two situations:

[0103] The first interval is when the accuracy of the deep learning model in recognizing EEG signal images is greater than the preset accuracy, which corresponds to the situation that the training effectiveness of the deep learning model meets the requirements;

[0104] The second interval is when the accuracy of the deep learning model in recognizing EEG signal images is less than or equal to the preset accuracy. The corresponding situation is: during the EEG signal acquisition process, the collected data may be incomplete due to poor contact between the electrode and the scalp, resulting in some data features being mistakenly deleted during the cleaning process, resulting in incompleteness in the model optimization stage.

[0105] In practice, the preset accuracy is generally selected in the range of [94%, 96%].

[0106] Preferably, the preset accuracy rate is 95%.

[0107] Specifically, the accuracy of the deep learning model in recognizing EEG signal images is the ratio of the number of times the deep learning model accurately recognizes EEG signal images to the total number of times the deep learning model recognizes EEG signal images.

[0108] In implementation, the method of the present invention determines the training effectiveness of the deep learning model by setting a preset accuracy rate, thereby reducing the impact of the decline in the recognition accuracy of the EEG signal image due to inaccurate determination of the training effectiveness of the deep learning model, and further improving the recognition accuracy of the EEG signal image.

[0109] Specifically, the reduction range of the learning rate of the deep learning model is determined by the difference between a preset accuracy rate and the accuracy rate of the deep learning model in recognizing the EEG signal image.

[0110] Specifically, when the difference between the preset accuracy rate and the accuracy rate of the deep learning model in recognizing EEG signal images is within 5%, the learning rate of the deep learning model is reduced to 0.9 times the original rate; when the difference between the preset accuracy rate and the accuracy rate of the deep learning model in recognizing EEG signal images exceeds 5%, on the basis of being reduced to 0.9 times the original rate, the learning rate of the deep learning model is reduced by 0.005 for every 2% that exceeds it. For example, the difference between the preset accuracy rate and the accuracy rate of the deep learning model in recognizing EEG signal images is 9%, the current learning rate of the deep learning model is 0.1, and the reduced learning rate of the deep learning model is 0.1×0.9-0.005×2=0.08.

[0111] In implementation, the method of the present invention adjusts the learning rate of the deep learning model by setting a preset accuracy rate. During the EEG signal acquisition process, the collected data may be incomplete due to poor contact between the electrode and the scalp, resulting in some data features being erroneously deleted during the cleaning process, which leads to incomplete model optimization. By reducing the learning rate of the deep learning model, the model can adjust parameters more finely, reduce the learning of erroneous features, and avoid overfitting of the model to abnormal features caused by incomplete data, thereby further improving the recognition accuracy of EEG signal images.

[0112] An artificial intelligence-based EEG signal image recognition system, comprising:

[0113] A data acquisition module, used to collect EEG signal data;

[0114] a data processing module connected to the data acquisition module, comprising a preprocessing unit for preprocessing the EEG signal data to output optimized data, a feature extraction unit connected to the preprocessing unit for extracting features from the optimized data to output EEG signal features, and a labeling unit for labeling the EEG signal data according to known labeling information;

[0115] A model training module, connected to the data processing module, comprising a model training unit for training a deep learning model based on the labeled EEG signal data and the EEG signal features, and a model updating unit connected to the model training unit for updating the deep learning model in real time;

[0116] An image recognition module, connected to the model training module, is used to recognize EEG signal images using the trained deep learning model to output recognition results;

[0117] a storage module, which is respectively connected to the data acquisition module, the data processing module, the model training module, and the image recognition module, and is used to store the EEG signal data, the optimization data, the EEG signal features, the known labeling information, the labeled EEG signal data, the deep learning model, and the recognition results;

[0118] A control module is respectively connected to the data acquisition module, the data processing module, the model training module, the image recognition module and the storage module, and is used to determine the sampling and holding time of the EEG signal data according to the area ratio of the distorted region of the EEG signal image, or to determine the cutoff frequency of the filter according to the packet loss rate of the EEG signal data, and to determine the learning rate of the deep learning model according to the accuracy of the deep learning model in identifying the EEG signal image.

[0119] Specifically, preprocessing includes denoising, cleaning, and filtering operations.

[0120] Specifically, the optimized data includes the EEG signal data amplitude with reduced noise interference, complete time series data, and EEG signal data with irrelevant frequency interference removed.

[0121] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

Claims

1. A method for recognizing electroencephalogram (EEG) signals based on artificial intelligence, characterized in that: include: De-noising, cleaning, filtering, and feature extraction are performed on the collected EEG signal data in order to output EEG signal features, and the EEG signal data is labeled according to known label information; Using the labeled EEG signal data and the EEG signal features to train and update the deep learning model in real time, and using the trained deep learning model to recognize the EEG signal image to output a recognition result; respectively obtaining the distorted area of ​​the identified EEG signal image and the total area of ​​the EEG signal image; Determine the recognition accuracy of the EEG signal image based on the area ratio of the distorted region of the EEG signal image; If the recognition accuracy of the EEG signal image does not meet the requirements, the sampling and holding time of the EEG signal data is adjusted, or the recognition validity of the EEG signal image is determined based on the packet loss rate of the EEG signal data; If the recognition effectiveness of the EEG signal image does not meet the requirements, the cutoff frequency of the filter is adjusted, or the learning rate of the deep learning model is adjusted based on the accuracy of the deep learning model in recognizing the EEG signal image; Determining the recognition accuracy of the EEG signal image includes: Comparing the area ratio of the distorted region of the EEG signal image with a preset first ratio; If the distorted area ratio of the EEG signal image is greater than the preset first ratio, it is determined that the recognition accuracy of the EEG signal image does not meet the requirement; Determining the recognition validity of the EEG signal image includes: Comparing the area ratio of the distorted region of the EEG signal image with the preset first ratio and the preset second ratio respectively; If the proportion of the distorted area of ​​the EEG signal image is greater than the preset first proportion and less than or equal to the preset second proportion, it is preliminarily determined that the recognition validity of the EEG signal image does not meet the requirements, and whether the recognition validity of the EEG signal image meets the requirements is determined based on the packet loss rate of the EEG signal data; Adjusting the sampling and holding time of the EEG signal data includes: Comparing the area ratio of the distorted region of the EEG signal image with the preset second ratio; If the distorted area ratio of the EEG signal image is greater than a preset second ratio, reducing the sampling and holding time of the EEG signal data; The reduction range of the sampling and holding time of the electroencephalogram signal data is determined by the difference between the area ratio of the distorted region of the electroencephalogram signal image and a preset second ratio.

2. The method for EEG signal image recognition based on artificial intelligence according to claim 1, characterized in that: Adjusting the cutoff frequency of the filter includes: Comparing the packet loss rate of the EEG signal data with a preset first packet loss rate and a preset second packet loss rate respectively; If the packet loss rate of the EEG signal data is greater than the preset first packet loss rate, determining that the recognition validity of the EEG signal image does not meet the requirements; If the packet loss rate of the EEG signal data is greater than a preset first packet loss rate and less than or equal to a preset second packet loss rate, reducing the cutoff frequency of the filter; If the packet loss rate of the EEG signal data is greater than the preset second packet loss rate, it is preliminarily determined that the training effectiveness of the deep learning model does not meet the requirements, and whether the training effectiveness of the deep learning model meets the requirements is determined based on the accuracy of the deep learning model in identifying the EEG signal image.

3. The method for EEG signal image recognition based on artificial intelligence according to claim 2, characterized in that: The reduction range of the cutoff frequency of the filter is determined by the difference between the packet loss rate of the EEG signal data and a preset first packet loss rate.

4. The method for EEG signal image recognition based on artificial intelligence according to claim 3, characterized in that: Adjusting the learning rate of the deep learning model includes: Compare the accuracy of the deep learning model in recognizing EEG signal images with the preset accuracy; If the accuracy of the deep learning model in recognizing the EEG signal image is less than or equal to the preset accuracy, it is determined that the training effectiveness of the deep learning model does not meet the requirements, and the learning rate of the deep learning model is reduced.

5. The method for EEG signal image recognition based on artificial intelligence according to claim 4, characterized in that: The reduction range of the learning rate of the deep learning model is determined by the difference between a preset accuracy rate and the accuracy rate of the deep learning model in recognizing the electroencephalogram signal image.

6. An artificial intelligence-based EEG signal image recognition system using the artificial intelligence-based EEG signal image recognition method according to any one of claims 1 to 5, characterized in that: include: A data acquisition module, used to collect EEG signal data; a data processing module connected to the data acquisition module, comprising a preprocessing unit for preprocessing the EEG signal data to output optimized data, a feature extraction unit connected to the preprocessing unit for extracting features from the optimized data to output EEG signal features, and a labeling unit for labeling the EEG signal data according to known labeling information; A model training module, connected to the data processing module, comprising a model training unit for training a deep learning model based on the labeled EEG signal data and the EEG signal features, and a model updating unit connected to the model training unit for updating the deep learning model in real time; An image recognition module, connected to the model training module, is used to recognize EEG signal images using the trained deep learning model to output recognition results; a storage module, which is respectively connected to the data acquisition module, the data processing module, the model training module, and the image recognition module, and is used to store the EEG signal data, the optimization data, the EEG signal features, the known labeling information, the labeled EEG signal data, the deep learning model, and the recognition results; A control module is respectively connected to the data acquisition module, the data processing module, the model training module, the image recognition module and the storage module, and is used to determine the sampling and holding time of the EEG signal data according to the area ratio of the distorted region of the EEG signal image, or to determine the cutoff frequency of the filter according to the packet loss rate of the EEG signal data, and to determine the learning rate of the deep learning model according to the accuracy of the deep learning model in identifying the EEG signal image.

Images