Brain wave anomaly warning method based on electroencephalogram signals and tongue image

Abstract

Embodiments of the present disclosure disclose a brain wave anomaly warning method based on electroencephalogram signals and tongue image pictures. A specific embodiment of the method includes: obtaining brain wave test information, a brain wave test image, and a target tongue image; sending the brain wave test information and the brain wave test image to a user terminal to generate an initial electroencephalogram test signal, and receiving the initial electroencephalogram test signal from the user terminal; performing anomaly detection processing on the initial electroencephalogram test signal to obtain target brain wave anomaly feature information; performing recognition processing on the target tongue image to obtain target tongue anomaly feature information; inputting the target brain wave anomaly feature information and the target tongue anomaly feature information into a pre-trained feature fusion model to obtain brain wave anomaly information; and sending the brain wave anomaly information to a warning terminal for performing a warning operation. The embodiment improves the accuracy of anomaly warning.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of computer technology, and more specifically to a method for alerting abnormal brainwaves based on electroencephalogram (EEG) signals and tongue images. Background Technology

[0002] Brain waves reflect the electrophysiological activity of brain nerve cells in the cerebral cortex or scalp. When abnormal brain waves occur, timely alarm operations are required. Currently, the common method for issuing brain wave abnormality alarms is to extract feature information of the electroencephalogram (EEG) signal using existing convolutional neural network models to identify the abnormality and issue an alarm.

[0003] However, the inventors discovered that when using the above method to issue brainwave abnormality alarms, the following technical problems often occur:

[0004] First, the characteristic information of brainwave signals can only characterize short-term (e.g., one day, one hour) brainwave characteristics, and it is difficult to characterize long-term (e.g., one week, one month) brainwave characteristics. This makes it difficult to characterize some abnormal brainwave information, which in turn makes it difficult to alert to some abnormal brainwaves, and consequently reduces the accuracy of abnormal alerts.

[0005] Second, existing convolutional neural network models can only extract feature information from brainwave signals according to a fixed single process, which reduces the accuracy of the extracted brainwave signal feature information, and thus reduces the accuracy of abnormal alarms.

[0006] The information disclosed in this background section is only intended to enhance the understanding of the background of the inventive concept, and therefore may contain information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0008] Some embodiments of this disclosure propose a method and apparatus for alarming abnormal brain waves based on electroencephalogram (EEG) signals and tongue images, in order to solve one or more of the technical problems mentioned in the background section above.

[0009] In a first aspect, some embodiments of this disclosure provide a method for alerting abnormal brainwaves based on electroencephalogram (EEG) signals and tongue images. The method includes: acquiring brainwave test information, brainwave test images, and a target tongue image; sending the brainwave test information and the brainwave test image to a user terminal to generate an initial EEG test signal, and receiving the initial EEG test signal from the user terminal; performing anomaly detection processing on the initial EEG test signal to obtain target brainwave abnormality feature information; performing recognition processing on the target tongue image to obtain target tongue abnormality feature information; inputting the target brainwave abnormality feature information and the target tongue abnormality feature information into a pre-trained feature fusion model to obtain brainwave abnormality information; and sending the brainwave abnormality information to an alarm terminal for executing an alarm operation.

[0010] Secondly, some embodiments of this disclosure provide a brainwave abnormality alarm device based on electroencephalogram (EEG) signals and tongue images. The device includes: an acquisition unit configured to acquire brainwave test information, brainwave test images, and target tongue images; a first transmission unit configured to transmit the brainwave test information and the brainwave test images to a user terminal to generate an initial EEG test signal, and to receive the initial EEG test signal from the user terminal; an abnormality detection unit configured to perform abnormality detection processing on the initial EEG test signal to obtain target brainwave abnormality feature information; an identification unit configured to perform identification processing on the target tongue image to obtain target tongue abnormality feature information; an input unit configured to input the target brainwave abnormality feature information and the target tongue abnormality feature information into a pre-trained feature fusion model to obtain brainwave abnormality information; and a second transmission unit configured to send the brainwave abnormality information to an alarm terminal for performing an alarm operation.

[0011] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any of the implementations of the first aspect above.

[0012] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0013] The above-described embodiments of this disclosure have the following beneficial effects: the accuracy of abnormal alarms can be improved through the brainwave abnormality alarm method based on EEG signals and tongue images according to some embodiments of this disclosure. Specifically, the reason for the reduced accuracy of abnormal alarms is that the feature information of brainwave signals can only characterize short-term (e.g., one day, one hour) brainwave feature information, and it is difficult to characterize long-term (e.g., one week, one month) brainwave feature information, resulting in difficulty in characterizing some brainwave abnormal information, and thus, difficulty in alarming some brainwave abnormalities. Based on this, the brainwave abnormality alarm method based on EEG signals and tongue images according to some embodiments of this disclosure first acquires brainwave test information, brainwave test images, and target tongue images. Second, the brainwave test information and brainwave test images are sent to a user terminal to generate an initial EEG test signal, and the initial EEG test signal is received from the user terminal. Thus, real-time brainwave signals can be obtained for subsequent feature extraction of brainwave signals. Next, abnormality detection processing is performed on the initial EEG test signal to obtain target brainwave abnormality feature information. Therefore, real-time brainwave feature information can be obtained. Then, the target tongue image is processed to obtain abnormal feature information of the target tongue. This allows for the acquisition of category information for the tongue image that can characterize long-term brainwave feature information. Next, the target brainwave abnormal feature information and the target tongue abnormal feature information are input into a pre-trained feature fusion model to obtain abnormal brainwave information. This yields brainwave feature information that fuses long-term and short-term features. Finally, the abnormal brainwave information is sent to an alarm terminal for alarm operation. Thus, based on the brainwave feature information that fuses long-term and short-term features, an alarm can be triggered for abnormal brainwave feature information. Therefore, some brainwave abnormality alarm methods disclosed in this invention not only consider real-time short-term brainwave abnormality information but also long-term abnormal feature information. This allows for the characterization of some brainwave abnormalities, enabling alarms for some brainwave abnormalities and improving the accuracy of abnormality alarms. Attached Figure Description

[0014] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0015] Figure 1 This is a flowchart of some embodiments of the brainwave abnormality alarm method based on electroencephalogram signals and tongue images according to the present disclosure;

[0016] Figure 2 This is a schematic diagram of the structure of some embodiments of the brainwave abnormality alarm device based on electroencephalogram signals and tongue images according to the present disclosure.

[0017] Figure 3 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0018] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0019] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0020] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0021] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0022] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0023] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0024] Figure 1 A flowchart 100 is shown illustrating some embodiments of a brainwave abnormality alarm method based on electroencephalogram (EEG) signals and tongue images according to this disclosure. The brainwave abnormality alarm method includes the following steps:

[0025] Step 101: Obtain brainwave test information, brainwave test images, and target tongue image.

[0026] In some embodiments, the entity executing the brainwave abnormality alarm method based on EEG signals and tongue images can acquire the aforementioned brainwave test information, brainwave test images, and target tongue images from a terminal device via a wired or wireless connection. The aforementioned brainwave test information may be preset test information used to generate brainwave information. The aforementioned brainwave test image may be an image depicting an emotion. The aforementioned target tongue image may be a tongue image captured by a camera.

[0027] As an example, the preset test information mentioned above could be VFT (verbal fluency test) information.

[0028] It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other currently known or future wireless connection methods.

[0029] Step 102: Send the brainwave test information and brainwave test images to the user terminal to generate the initial brainwave test signal, and receive the initial brainwave test signal from the user terminal.

[0030] In some embodiments, the executing entity may send the brainwave test information and the brainwave test image to a user terminal to generate an initial brainwave test signal, and receive the initial brainwave test signal from the user terminal. The initial brainwave test signal may include, but is not limited to, at least one of the following: an initial brainwave signal set. Each initial brainwave signal in the initial brainwave signal set may be a brainwave signal within a preset band of the user terminal.

[0031] As an example, the preset bands mentioned above can be, but are not limited to, at least one of the following: alpha wave, beta wave, gamma wave, delta wave, or theta wave.

[0032] Step 103: Perform abnormal detection processing on the initial EEG test signal to obtain the target brainwave abnormality feature information.

[0033] In some embodiments, the aforementioned execution entity may perform abnormal detection processing on the aforementioned initial EEG test signal to obtain target brainwave abnormality feature information.

[0034] In some optional implementations of certain embodiments, the execution entity performs anomaly detection processing on the initial EEG test signal to obtain target brainwave abnormality feature information, which may include the following steps:

[0035] The first step involves denoising each initial brainwave signal in the initial brainwave signal set included in the initial EEG test signal to generate a denoised brainwave signal, thus obtaining a denoised brainwave signal set. Specifically, a preset denoising filter can be used to denoise each initial brainwave signal in the initial brainwave signal set included in the initial EEG test signal to generate the denoised brainwave signal.

[0036] As an example, the preset noise reduction filter mentioned above can be, but is not limited to, at least one of the following: a bandpass filter or a Butterworth filter.

[0037] The second step is to select denoised brainwave signals that meet the target band conditions from the aforementioned set of denoised brainwave signals, and use these as the target brainwave signals. The target band conditions can be that the denoised brainwave signals are alpha wave brainwave signals.

[0038] The third step involves feature extraction processing of the target brainwave signal to obtain target brainwave feature information. Specifically, the differential entropy of the target brainwave signal can be determined as the target brainwave feature information.

[0039] The fourth step involves inputting the aforementioned target brainwave feature information into a pre-trained brainwave abnormality detection model to obtain the target brainwave abnormality feature information. This pre-trained brainwave abnormality detection model can be a pre-trained neural network model that takes the target brainwave feature information as input and outputs the target brainwave abnormality feature information. The aforementioned target brainwave abnormality feature information can be the brainwave abnormality discrimination result. This brainwave abnormality discrimination result can represent information indicating "abnormal brainwaves" or information indicating "no abnormal brainwaves."

[0040] Optionally, the pre-trained brainwave abnormality detection model described above can be trained through the following steps:

[0041] The first step is to obtain the training sample brainwave information set and the initial brainwave training round values. Each training sample brainwave information set includes: sample brainwave feature information and sample brainwave abnormality feature information. The training sample brainwave information set can be obtained from a sample brainwave database, and the initial brainwave training round values ​​can be obtained from a storage terminal. The storage terminal can be a terminal used for storing information. The sample brainwave feature information can be the differential entropy of the sample brainwave signal. The sample brainwave abnormality feature information can be information indicating "abnormalities in the sample brainwave signal" or information indicating "no abnormalities in the sample brainwave signal".

[0042] As an example, the aforementioned sample brainwave database may be, but is not limited to, at least one of the following: a database that uses physiological signals for emotion analysis or a neural information database. The initial brainwave training round value may be 0.

[0043] The second step involves selecting brainwave information from the aforementioned set of training sample brainwave information and performing the following first training sub-step:

[0044] The first sub-step involves inputting the sample brainwave feature information, including the training sample brainwave information, into the first anomaly detection sub-model of the initial brainwave anomaly detection model to obtain the first initial brainwave anomaly information. The initial brainwave anomaly detection model further includes: a second anomaly detection sub-model, an anomaly feature fusion layer, and an anomaly feature linear layer. Training sample brainwave information can be randomly selected from the aforementioned training sample brainwave information set to execute the first training sub-step. The aforementioned initial brainwave anomaly detection model can be an untrained neural network model that takes the target brainwave feature information as input and the target brainwave anomaly feature information as output.

[0045] The aforementioned first anomaly detection sub-model can be divided into nine layers: The first layer can be used to perform depthwise convolution processing on the brainwave feature information of the aforementioned samples. The second layer can be used to perform batch normalization processing on the output of the first layer. The third layer can be used to perform two-dimensional convolution processing on the output of the second layer. The fourth layer can be used to perform normalized pooling processing on the output of the third layer. Here, the fourth layer can include four modules: The first module can be used to perform batch normalization processing on the output of the third layer. The second module can be used to perform correction processing on the output of the first module. The third module can be used to perform average pooling processing on the output of the second module. The fourth module can be used to discard the output of the third module. The fifth layer can be used to perform separable convolution processing on the output of the fourth layer. The sixth layer can be used to perform attention feature extraction processing on the output of the fifth layer. The seventh layer can be used to perform layer normalization processing on the output of the sixth layer. The eighth layer can be used to perform linear transformation processing on the output of the seventh layer. The ninth layer can be used to flatten the output of the eighth layer to obtain the aforementioned first initial brainwave anomaly information.

[0046] The aforementioned second anomaly detection sub-model can be divided into nine layers: The first layer can be used to perform the aforementioned two-dimensional convolution processing on the aforementioned sample brainwave feature information. The second layer can be used to perform the aforementioned batch normalization processing on the output of the first layer. The third layer can be used to perform the aforementioned depthwise convolution processing on the output of the second layer. The fourth layer can be used to perform the aforementioned normalized pooling processing on the output of the third layer. The fifth layer can be used to perform the aforementioned separable convolution processing on the output of the fourth layer. The sixth layer can be used to perform the aforementioned attention feature extraction processing on the output of the fifth layer. The seventh layer can be used to perform the aforementioned layer normalization processing on the output of the sixth layer. The eighth layer can be used to perform the aforementioned linear transformation processing on the output of the seventh layer. The ninth layer can be used to perform the aforementioned flattening processing on the output of the eighth layer to obtain the second initial brainwave anomaly information.

[0047] As an example, the first layer of the aforementioned first anomaly detection sub-model can be a DepthwiseConv (depthwise convolution) layer. The second layer of the aforementioned first anomaly detection sub-model can be a BatchNorm (batch normalization) layer. The third layer of the aforementioned first anomaly detection sub-model can be a conv2D (two-dimensional convolution) layer. The first module of the fourth layer of the aforementioned first anomaly detection sub-model can be a BatchNorm (batch normalization) module. The second module of the fourth layer of the aforementioned first anomaly detection sub-model can be a Prelu (Parametric Rectified Linear Unit) module. The third module of the fourth layer of the aforementioned first anomaly detection sub-model can be an Average Pooling module. The fourth module of the fourth layer of the aforementioned first anomaly detection sub-model can be a Dropout module. The fifth layer of the aforementioned first anomaly detection sub-model can be a SeparableConv (separable convolution) layer. The sixth layer of the aforementioned first anomaly detection sub-model can be a MultiheadAttention layer. The seventh layer in the aforementioned first anomaly detection sub-model can be a LayerNorm (layer normalization) layer. The eighth layer in the aforementioned first anomaly detection sub-model can be a Linear layer. The ninth layer in the aforementioned first anomaly detection sub-model can be a Flatten layer. The abnormal feature fusion layer in the aforementioned initial EEG anomaly detection model can be a Concentrate layer. The abnormal feature linear layer in the aforementioned initial EEG anomaly detection model can be a Linear layer.

[0048] The second sub-step involves inputting the sample brainwave feature information, including the brainwave information of the training samples, into the second abnormality detection sub-model included in the initial brainwave abnormality detection model to obtain the second initial brainwave abnormality information.

[0049] The third sub-step involves inputting the first and second initial brainwave abnormality information into the abnormal feature fusion layer included in the initial brainwave abnormality detection model to obtain the initial fused brainwave abnormality information.

[0050] The fourth sub-step involves inputting the initial fused brainwave abnormality information into the abnormal feature linear layer included in the initial brainwave abnormality detection model to obtain the initial brainwave abnormality feature information.

[0051] The fifth sub-step involves determining the difference in brainwave abnormality between the initial brainwave abnormality feature information and the sample brainwave abnormality feature information included in the training sample brainwave information, based on a preset first loss function.

[0052] As an example, the first loss function mentioned above can be, but is not limited to, at least one of the following: MSE (Mean Square Error), MAE (Mean Absolute Error), or Categorical Cross Entropy Loss.

[0053] The sixth sub-step involves updating the initial brainwave training cycle value to obtain an updated brainwave training cycle value. Specifically, the updated brainwave training cycle value can be determined by summing the initial brainwave training cycle value with a preset update value.

[0054] As an example, the preset update value mentioned above could be 1.

[0055] The seventh sub-step involves determining that the abnormal brainwave difference value is less than a first preset threshold and the updated brainwave training round value is greater than a first target threshold, thereby establishing the initial brainwave abnormality detection model as the brainwave abnormality detection model. The first target threshold can be the product of the number of training sample brainwave information items in the training sample brainwave information set and a first target ratio value.

[0056] As an example, the first preset threshold mentioned above can be 0.01. The first target ratio value mentioned above can be, but is not limited to, at least one of the following: 0.1, 0.2, or 0.3.

[0057] Optionally, the aforementioned executing entity may further adjust relevant parameters in the initial brainwave anomaly detection model in response to determining that the abnormal brainwave difference value is greater than or equal to a first preset threshold or that the updated brainwave training round value is less than or equal to a first target threshold. This involves determining the adjusted initial brainwave anomaly detection model as the initial brainwave anomaly detection model, determining the updated brainwave training round value as the initial brainwave training round value, and selecting training sample brainwave information from the unselected training sample brainwave information set for re-execution of the first training step. The relevant parameters in the initial brainwave anomaly detection model can be adjusted using a preset adjustment algorithm.

[0058] As an example, the preset adjustment algorithm mentioned above may be, but is not limited to, at least one of the following: the Adam optimizer algorithm or the gradient descent algorithm.

[0059] The content related to step 103, as an inventive point of this disclosure, solves the second technical problem mentioned in the background art: "reduced accuracy of abnormal alarms." The factors leading to reduced accuracy of abnormal alarms are often as follows: existing convolutional neural network models can only extract feature information from brainwave signals according to a fixed, single process, resulting in reduced accuracy of the extracted brainwave signal feature information. Solving these factors can improve the accuracy of abnormal alarms. To achieve this, this disclosure can use a custom neural network model to extract features from the differential entropy of brainwave signals using feature extraction sub-models with different focuses. Then, a self-attention mechanism can be used to comprehensively extract features from the outputs of the two feature sub-models. Subsequently, fully connected layers and concatenation layers can be used to complete feature extraction. Finally, normalization can be performed using linear layers to obtain the final brainwave abnormality probability value. Therefore, feature information from brainwave signals can be extracted through two feature extraction sub-models with different processes, thereby improving the accuracy of the extracted brainwave feature information and, consequently, improving the accuracy of abnormal alarms.

[0060] Step 104: Recognize and process the target tongue image to obtain abnormal feature information of the target tongue.

[0061] In some embodiments, the execution entity may perform recognition processing on the target tongue image to obtain abnormal feature information of the target tongue.

[0062] In some optional implementations of certain embodiments, the execution entity performs recognition processing on the target tongue image to obtain abnormal feature information of the target tongue, which may include the following steps:

[0063] The first step is to segment the target tongue image to obtain the target tongue surface image. This can be done using a pre-defined segmentation algorithm.

[0064] As an example, the aforementioned preset segmentation algorithm could be the SAM (segment anything model) algorithm.

[0065] The second step involves segmenting the target tongue image to obtain an enhanced tongue image set. This can be achieved using a preset segmentation method.

[0066] As an example, the above-mentioned preset splitting method can be, but is not limited to, at least one of the following: random image cropping method or image rotation method.

[0067] The third step involves inputting the enhanced tongue surface image set into a pre-trained image anomaly recognition model to obtain the target tongue anomaly feature information. The pre-trained image anomaly recognition model can be a pre-trained neural network model that takes the enhanced tongue surface image set as input and the target tongue anomaly feature information as output.

[0068] Optionally, the pre-trained image anomaly recognition model described above can be trained through the following steps:

[0069] The first step involves acquiring an initial training sample image sequence set, initial pre-training epoch values, a training sample tongue information set, and initial tongue training epoch values. Each training sample tongue information in the aforementioned training sample tongue information set includes: a sample tongue surface image and sample tongue category information. The initial training sample image sequence can be obtained from a training sample image database, and the training sample tongue information set and initial tongue training epoch values ​​can be obtained from the aforementioned storage terminal. The initial training sample images in the aforementioned initial training sample image sequence can be randomly ordered. The sample tongue surface images can be tongue surface images captured by the aforementioned camera. The sample tongue category information characterizes the category of the sample tongue surface image.

[0070] As an example, the training sample image database mentioned above could be the flowers102 classification database. The initial tongue training round value mentioned above could be 0.

[0071] The second step involves training the initial image anomaly detection model based on the aforementioned initial training sample image sequence set and the initial pre-training round values, thereby obtaining the target image anomaly detection model. This initial image anomaly detection model can be an untrained neural network model that takes the initial training sample image sequence as input and outputs the target image anomaly detection model. The initial image anomaly detection model may include, but is not limited to, at least one of the following: an image classification sub-model and a category normalization sub-model.

[0072] As an example, the image classification sub-model described above could be a ViT (Vision Transformer) model. The category normalization sub-model described above could be a linear layer.

[0073] The third step involves selecting tongue information from the aforementioned training sample tongue information set and performing the following third training sub-step:

[0074] The first sub-step involves inputting the sample tongue image, including the tongue information of the training samples, into the target image anomaly recognition model to obtain initial tongue category information.

[0075] The second sub-step involves determining the class difference value between the initial tongue category information and the sample tongue category information included in the training sample tongue information, based on a preset third loss function.

[0076] As an example, the aforementioned preset third loss function could be the classification cross-entropy loss function.

[0077] The third sub-step involves updating the initial tongue training round value to obtain the updated tongue training round value. Specifically, the updated tongue training round value can be determined by summing the initial tongue training round value with the preset update value.

[0078] The fourth sub-step involves determining the target image anomaly recognition model as such in response to the determination that the category difference value is less than a third preset threshold and the updated tongue training round value is greater than a third target threshold. The third target threshold can be the product of the number of training sample fusion information points in the training sample fusion information set and the third target ratio value.

[0079] As an example, the aforementioned third preset threshold can be 0.01. The aforementioned third target ratio value can be, but is not limited to, at least one of the following: 0.1, 0.2, or 0.3.

[0080] Optionally, the aforementioned execution entity may further adjust the relevant parameters in the initial feature fusion model and the relevant parameters in the image anomaly recognition model to be adjusted in response to determining that the fusion anomaly difference value is greater than or equal to a second preset threshold or that the updated tongue training round value is less than or equal to a third target threshold. The adjusted initial feature fusion model and the image anomaly recognition model to be adjusted are then determined as the initial feature fusion model and the image anomaly recognition model to be adjusted, and the updated tongue training round value is determined as the initial tongue training round value. Furthermore, training sample fusion information is selected from the unselected training sample fusion information set for re-execution of the second training step. The relevant parameters in the initial feature fusion model and the relevant parameters in the image anomaly recognition model to be adjusted can be adjusted using the aforementioned preset adjustment algorithm.

[0081] In some optional implementations of certain embodiments, the execution entity may further select an initial training sample image sequence from the initial training sample image sequence set and perform the following pre-training steps:

[0082] In the first pre-training step, for each initial training sample image in the above initial training sample image sequence, the following fourth training sub-step is performed:

[0083] The first training sub-step involves selecting initial training positive and negative sample images from the initial training sample image sequence. This selection can be achieved by: first, using a preset similarity function to determine the image similarity value between the initial training sample image and each other initial training sample image in the initial training sample image sequence, resulting in an image similarity value sequence. Then, the initial training sample image corresponding to the largest image similarity value in the image similarity value sequence can be designated as the initial training positive sample image. Similarly, the initial training sample image corresponding to the smallest image similarity value in the image similarity value sequence can be designated as the initial training negative sample image.

[0084] As an example, the aforementioned preset similarity function can be, but is not limited to, at least one of the following: MSE function, SSIM (structural similarity) function, or cosine similarity function.

[0085] The second training sub-step involves inputting the initial training sample image, the initial training positive sample image, and the initial training negative sample image into the initial image anomaly recognition model to obtain the initial sample category information, the initial positive sample category information, and the initial negative sample category information.

[0086] The third training sub-step involves determining the positive sample similarity vector between the initial sample category information and the initial positive sample category information. This can be achieved using a pre-defined similarity algorithm.

[0087] As an example, the aforementioned preset similarity algorithm could be a cosine similarity algorithm.

[0088] The fourth training sub-step involves determining the negative sample similarity vector between the initial sample category information and the initial negative sample category information. This can be achieved using the aforementioned preset similarity algorithm.

[0089] The fifth training sub-step determines the training loss values ​​for the first preset similarity vector, the second preset similarity vector, the positive sample similarity vector, and the negative sample similarity vector based on the preset fourth loss function.

[0090] As an example, the aforementioned preset fourth loss function can be the aforementioned MSE function. The aforementioned first preset similarity vector can be a vector where all dimensions are 1. The aforementioned second preset similarity vector can be a zero vector.

[0091] The sixth training sub-step involves adjusting the relevant parameters of the initial image anomaly detection model based on the training loss value, and determining the adjusted initial image anomaly detection model as the next initial training sample image and its corresponding initial image anomaly detection model. The relevant parameters of the initial image anomaly detection model can be adjusted using the aforementioned preset adjustment algorithm.

[0092] The second pre-training step involves determining the initial image anomaly recognition model corresponding to the last initial training sample image in the above initial training sample image sequence as the above target image anomaly recognition model.

[0093] The third pre-training step involves updating the initial pre-training epoch values ​​to obtain updated pre-training epoch values. Specifically, the updated pre-training epoch value can be determined by summing the initial pre-training epoch values ​​and the preset update values.

[0094] The fourth pre-training step involves determining the pre-training round value as the target image anomaly recognition model in response to the determination that the updated pre-training round value is greater than the fourth target threshold. The fourth target threshold can be the product of the number of initial training sample image sequences in the initial training sample image sequence set and the fourth target ratio value.

[0095] As an example, the fourth target ratio value mentioned above can be, but is not limited to, at least one of the following: 0.1, 0.2, or 0.3.

[0096] Optionally, the execution entity may also select an initial training sample image sequence from the set of initial training sample image sequences and from the unselected initial training sample image sequences in response to determining that the updated pre-training round value is less than or equal to the fourth target threshold, so as to perform the pre-training step again.

[0097] Therefore, the accuracy of the image anomaly recognition model can be improved by pre-training it. Due to the limited number of tongue images, the training data for the model is insufficient, leading to a decrease in accuracy. Therefore, firstly, based on the similarity between tongue image classification and flower classification, a flower dataset can be used for the first pre-training of the image anomaly recognition model. Then, tongue images can be used alone for the second pre-training. Finally, paired brainwave feature information and tongue images can be used for a third training, resulting in a trained image anomaly recognition model. This improves the accuracy of the image anomaly recognition model, and consequently, the accuracy of the identified tongue image category information.

[0098] Step 105: Input the target brainwave abnormality feature information and the target tongue abnormality feature information into the pre-trained feature fusion model to obtain brainwave abnormality information.

[0099] In some embodiments, the execution entity may input the target brainwave abnormality feature information and the target tongue abnormality feature information into a pre-trained feature fusion model to obtain brainwave abnormality information. The pre-trained feature fusion model may be a pre-trained neural network model that takes the target brainwave abnormality feature information and the target tongue abnormality feature information as input and outputs the brainwave abnormality information.

[0100] Optionally, the pre-trained feature fusion model and the pre-trained image anomaly recognition model described above can be trained through the following steps:

[0101] The first step is to obtain the training sample fusion information set and the initial fusion training round values. Each training sample fusion information in the aforementioned training sample fusion information set includes: sample-paired brainwave information, sample-paired tongue image, and sample brainwave abnormality information. The aforementioned training sample fusion information set and the initial fusion training round values ​​can be obtained from the aforementioned storage terminal. The aforementioned sample-paired brainwave information can be the differential entropy of a brainwave signal. The aforementioned sample-paired tongue image can be a tongue image captured by a camera. The user terminal corresponding to the aforementioned sample-paired brainwave information and the aforementioned sample-paired tongue image can be the same. The aforementioned sample brainwave abnormality information can be information indicating "abnormalities have occurred in the aforementioned training sample fusion information" or information indicating "no abnormalities have occurred in the aforementioned training sample fusion information."

[0102] As an example, the initial fusion training round value mentioned above can be 0.

[0103] The second step involves selecting training sample fusion information from the training sample fusion information set and performing the following second training sub-step:

[0104] The first sub-step involves inputting the sample-paired brainwave information, including the training sample fusion information, into a pre-trained brainwave anomaly detection model to obtain initial brainwave anomaly pairing feature information. The specific implementation of the training method for the aforementioned pre-trained brainwave anomaly detection model and its resulting technical effects can be found in step 103 of the above embodiments, and will not be repeated here.

[0105] The second sub-step involves inputting the sample-paired tongue images, including the training sample fusion information, into a pre-trained image anomaly recognition model to obtain initial tongue anomaly pairing feature information. The specific implementation of the training method for the aforementioned pre-trained image anomaly recognition model and its resulting technical effects can be found in step 104 of the above embodiments, and will not be repeated here.

[0106] The third sub-step involves inputting the aforementioned initial brainwave abnormality paired feature information and the aforementioned initial tongue abnormality paired feature information into the initial feature fusion model to obtain the initial brainwave abnormality information. This initial feature fusion model can be an untrained neural network model that takes the initial brainwave abnormality paired feature information and the initial tongue abnormality paired feature information as input and outputs the initial brainwave abnormality information. The initial feature fusion model can be divided into three layers: the first layer is used to fuse the aforementioned initial brainwave abnormality paired feature information and the initial tongue abnormality paired feature information; the second layer is used to linearize the output of the first layer; and the third layer is used to normalize the output of the second layer to obtain the aforementioned initial brainwave abnormality information.

[0107] As an example, the first layer of the above initial feature fusion model can be a fusion (multimodal fusion) layer. The second layer of the above initial feature fusion model can be a fully connected layer. The third layer of the above initial feature fusion model can be a tanh (normalization) function.

[0108] The fourth sub-step involves determining the fusion abnormality difference value of the sample brainwave abnormality information included in the initial brainwave abnormality information and the fusion information of the training samples, based on the preset second loss function.

[0109] As an example, the aforementioned preset second loss function may be, but is not limited to, at least one of the following: MSE (Mean Square Error) function, second-order norm loss function, mixed classification loss function, or classification cross-entropy loss function.

[0110] The fifth sub-step involves updating the initial fusion training round value to obtain the updated fusion training round value. Specifically, the updated fusion training round value can be determined by summing the initial fusion training round value with the preset update value.

[0111] The sixth sub-step involves determining the initial feature fusion model as the feature fusion model in response to the determination that the fusion anomaly difference value is less than a second preset threshold and the updated fusion training round value is greater than a second target threshold. The second target threshold can be the product of the number of training sample fusion information items in the training sample fusion information set and the second target ratio value.

[0112] As an example, the second preset threshold mentioned above can be 0.01. The second target ratio value mentioned above can be, but is not limited to, at least one of the following: 0.1, 0.2, or 0.3.

[0113] Optionally, the aforementioned execution entity may further adjust relevant parameters in the initial feature fusion model in response to determining that the fusion anomaly difference value is greater than or equal to a second preset threshold or that the updated fusion training round value is less than or equal to a second target threshold. This involves determining the adjusted initial feature fusion model as the initial feature fusion model, determining the updated fusion training round value as the initial fusion training round value, and selecting training sample fusion information from the aforementioned training sample fusion information set, specifically from the fusion information of each unselected training sample, for re-execution of the aforementioned second training step. The relevant parameters in the initial feature fusion model and the relevant parameters in the image anomaly recognition model to be adjusted can be adjusted using the aforementioned preset adjustment algorithm.

[0114] Therefore, a custom feature fusion model can be used to fuse abnormal EEG and tongue abnormalities through bilinear fusion, self-attention mechanisms, deep canonical correlation analysis, and splicing. This can improve the accuracy of abnormal EEG information, thereby improving the accuracy of abnormality alerts.

[0115] Step 106: Send the abnormal brainwave information to the alarm terminal for alarm operation.

[0116] In some embodiments, the aforementioned executing entity may send the abnormal brainwave information to an alarm terminal for alarm operation. The alarm terminal may execute the alarm operation by displaying warning text or controlling a speaker to emit a prompt sound. The alarm terminal may be a terminal used to execute alarm operations.

[0117] The above-described embodiments of this disclosure have the following beneficial effects: the accuracy of abnormal alarms can be improved through the brainwave abnormality alarm method based on EEG signals and tongue images according to some embodiments of this disclosure. Specifically, the reason for the reduced accuracy of abnormal alarms is that the feature information of brainwave signals can only characterize short-term (e.g., one day, one hour) brainwave feature information, and it is difficult to characterize long-term (e.g., one week, one month) brainwave feature information, resulting in difficulty in characterizing some brainwave abnormal information, and thus, difficulty in alarming some brainwave abnormalities. Based on this, the brainwave abnormality alarm method based on EEG signals and tongue images according to some embodiments of this disclosure first acquires brainwave test information, brainwave test images, and target tongue images. Second, the brainwave test information and brainwave test images are sent to a user terminal to generate an initial EEG test signal, and the initial EEG test signal is received from the user terminal. Thus, real-time brainwave signals can be obtained for subsequent feature extraction of brainwave signals. Next, abnormality detection processing is performed on the initial EEG test signal to obtain target brainwave abnormality feature information. Therefore, real-time brainwave feature information can be obtained. Then, the target tongue image is processed to obtain abnormal feature information of the target tongue. This allows for the acquisition of category information for the tongue image that can characterize long-term brainwave feature information. Next, the target brainwave abnormal feature information and the target tongue abnormal feature information are input into a pre-trained feature fusion model to obtain abnormal brainwave information. This yields brainwave feature information that fuses long-term and short-term features. Finally, the abnormal brainwave information is sent to an alarm terminal for alarm operation. Thus, based on the brainwave feature information that fuses long-term and short-term features, an alarm can be triggered for abnormal brainwave feature information. Therefore, some brainwave abnormality alarm methods disclosed in this invention not only consider real-time short-term brainwave abnormality information but also long-term abnormal feature information. This allows for the characterization of some brainwave abnormalities, enabling alarms for some brainwave abnormalities and improving the accuracy of abnormality alarms.

[0118] Further reference Figure 2 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a brainwave abnormality alarm device based on electroencephalogram (EEG) signals and tongue images. These device embodiments are similar to... Figure 1 Corresponding to the method embodiments shown, this brainwave abnormality alarm device can be specifically applied to various electronic devices.

[0119] like Figure 2As shown, the brainwave abnormality alarm device 200 in some embodiments includes: an acquisition unit 201, a first transmission unit 202, an abnormality detection unit 203, an identification unit 204, an input unit 205, and a second transmission unit 206. The acquisition unit 201 is configured to acquire brainwave test information, brainwave test images, and target tongue images; the first transmission unit 202 is configured to transmit the brainwave test information and brainwave test images to a user terminal to generate an initial brainwave test signal, and to receive the initial brainwave test signal from the user terminal; the anomaly detection unit 203 is configured to perform anomaly detection processing on the initial brainwave test signal to obtain target brainwave abnormality feature information; the recognition unit 204 is configured to perform recognition processing on the target tongue image to obtain target tongue abnormality feature information; the input unit 205 is configured to input the target brainwave abnormality feature information and the target tongue abnormality feature information into a pre-trained feature fusion model to obtain brainwave abnormality information; and the second transmission unit 206 is configured to send the brainwave abnormality information to an alarm terminal for alarm operation.

[0120] It is understandable that the units recorded in the brainwave abnormality alarm device 200 are related to the reference Figure 1 The steps in the described brainwave abnormality alarm method correspond to each other. Therefore, the operation, features, and beneficial effects described above for the brainwave abnormality alarm method also apply to the brainwave abnormality alarm device 200 and the units contained therein, and will not be repeated here.

[0121] The following is for reference. Figure 3 This document illustrates a structural schematic of an electronic device 300 suitable for implementing some embodiments of the present disclosure. The electronic devices in some embodiments of the present disclosure may include, but are not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 3 The terminal device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0122] like Figure 3As shown, the electronic device 300 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the electronic device 300. The processing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0123] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows electronic device 300 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 3 An electronic device 300 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 3 Each box shown can represent a device or multiple devices as needed.

[0124] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 309, or installed from storage device 308, or installed from ROM 302. When the computer program is executed by processing device 301, it performs the functions defined in the methods of some embodiments of this disclosure.

[0125] It should be noted that, in some embodiments of this disclosure, the computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0126] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for alerting abnormal brainwaves, comprising: Acquire brainwave test information, brainwave test images, and target tongue image; The brainwave test information and the brainwave test image are sent to the user terminal to generate an initial brainwave test signal, and the initial brainwave test signal is received from the user terminal. The initial EEG test signal is subjected to anomaly detection processing to obtain target brainwave abnormality feature information. The initial EEG test signal includes an initial brainwave signal set, and the anomaly detection processing of the initial EEG test signal to obtain target brainwave abnormality feature information includes: Each initial brainwave signal in the initial brainwave signal set included in the initial EEG test signal is subjected to noise reduction processing to generate a noise-reduced brainwave signal, thus obtaining a noise-reduced brainwave signal set. Noise-reduced brainwave signals that meet the target band conditions are selected from the set of noise-reduced brainwave signals and used as the target brainwave signals. The target brainwave signal is subjected to feature extraction processing to obtain target brainwave feature information; The target brainwave feature information is input into a pre-trained brainwave abnormality detection model to obtain the target brainwave abnormality feature information; The target tongue image is processed to obtain abnormal feature information of the target tongue; The abnormal brainwave features of the target and the abnormal tongue features of the target are input into a pre-trained feature fusion model to obtain abnormal brainwave information. The abnormal brainwave information is sent to the alarm terminal for alarm operation.

2. The method according to claim 1, wherein, The process of identifying and processing the target tongue image to obtain abnormal feature information of the target tongue includes: The target tongue image is segmented to obtain the target tongue surface image; The target tongue surface image is segmented to obtain an enhanced tongue surface image set; The enhanced tongue surface image set is input into a pre-trained image anomaly recognition model to obtain the abnormal feature information of the target tongue.

3. The method according to claim 1, wherein, The pre-trained brainwave anomaly detection model was obtained through the following steps: Obtain a training sample brainwave information set and an initial brainwave training round value, wherein each training sample brainwave information in the training sample brainwave information set includes: sample brainwave feature information and sample brainwave abnormal feature information. From the set of training sample brainwave information, select training sample brainwave information and perform the following first training step: The sample brainwave feature information, including the brainwave information of the training sample, is input into the first abnormality detection sub-model of the initial brainwave abnormality detection model to obtain the first initial brainwave abnormality information. The initial brainwave abnormality detection model also includes: a second abnormality detection sub-model, an abnormality feature fusion layer, and an abnormality feature linear layer. The sample brainwave feature information, including the brainwave information of the training sample, is input into the second abnormality detection sub-model included in the initial brainwave abnormality detection model to obtain the second initial brainwave abnormality information. The first and second initial brainwave abnormality information are input into the abnormal feature fusion layer included in the initial brainwave abnormality detection model to obtain the initial fused brainwave abnormality information. The initial fused brainwave abnormality information is input into the abnormal feature linear layer included in the initial brainwave abnormality detection model to obtain the initial brainwave abnormality feature information. Based on the preset first loss function, the difference value of brainwave abnormality features between the initial brainwave abnormality feature information and the sample brainwave abnormality feature information included in the training sample brainwave information is determined. The initial brainwave training round values ​​are updated to obtain the updated brainwave training round values. In response to determining that the abnormal difference value of brain waves is less than a first preset threshold and the updated brain wave training round value is greater than a first target threshold, the initial brain wave abnormality detection model is determined as the brain wave abnormality detection model.

4. The method according to claim 3, wherein, The method further includes: In response to determining that the abnormal difference value of brain waves is greater than or equal to a first preset threshold or the updated brain wave training round value is less than or equal to a first target threshold, the relevant parameters in the initial brain wave abnormality detection model are adjusted, the adjusted initial brain wave abnormality detection model is determined as the initial brain wave abnormality detection model, the updated brain wave training round value is determined as the initial brain wave training round value, and training sample brain wave information is selected from the training sample brain wave information set and from the unselected training sample brain wave information for re-executing the first training step.

5. The method according to claim 2, wherein, The pre-trained feature fusion model is obtained through the following steps: Obtain the training sample fusion information set and the initial fusion training round value, wherein each training sample fusion information in the training sample fusion information set includes: sample-paired brainwave information, sample-paired tongue image, and sample brainwave abnormality information; Select training sample fusion information from the training sample fusion information set and perform the following second training step: The training sample fusion information, including the sample-paired brainwave information, is input into the pre-trained brainwave abnormality detection model to obtain the initial brainwave abnormality pairing feature information. The sample-paired tongue images, including the training sample fusion information, are input into a pre-trained image anomaly recognition model to obtain initial tongue anomaly pairing feature information. The initial brainwave abnormality paired feature information and the initial tongue abnormality paired feature information are input into the initial feature fusion model to obtain the initial brainwave abnormality information; Based on the preset second loss function, the fusion abnormality difference value of the sample brainwave abnormal information included in the initial brainwave abnormality information and the training sample fusion information is determined; The initial fusion training round values ​​are updated to obtain the updated fusion training round values; In response to the determination that the fusion anomaly difference value is less than the second preset threshold and the updated fusion training round value is greater than the second target threshold, the initial feature fusion model is determined as the feature fusion model.

6. The method according to claim 5, wherein, The method further includes: In response to determining that the fusion anomaly difference value is greater than or equal to the second preset threshold or the updated fusion training round value is less than or equal to the second target threshold, the relevant parameters in the initial feature fusion model are adjusted, the adjusted initial feature fusion model is determined as the initial feature fusion model, the updated fusion training round value is determined as the initial fusion training round value, and training sample fusion information is selected from the training sample fusion information set and from the unselected training sample fusion information for re-executing the second training step.

7. The method according to claim 2, wherein, The pre-trained image anomaly recognition model is generated through the following steps: Obtain the initial training sample image sequence set, the initial pre-training round value, the training sample tongue information set, and the initial tongue training round value, wherein each training sample tongue information in the training sample tongue information set includes: a sample tongue surface image and sample tongue category information; Based on the initial training sample image sequence set and the initial pre-training round value, the initial image anomaly recognition model is trained to obtain the target image anomaly recognition model. From the set of tongue information of the training samples, select tongue information of training samples and perform the following third training step: The sample tongue images, including the tongue information of the training samples, are input into the target image anomaly recognition model to obtain the initial tongue category information; Based on the preset third loss function, the class difference value of the initial tongue category information and the sample tongue category information included in the training sample tongue information is determined; The initial tongue training round values ​​are updated to obtain the updated tongue training round values. In response to the determination that the category difference value is less than the third preset threshold and the updated tongue training round value is greater than the third target threshold, the target image anomaly recognition model is determined as the image anomaly recognition model.

8. The method according to claim 7, wherein, The method further includes: In response to determining that the category difference value is greater than or equal to the third preset threshold or the updated tongue training round value is less than or equal to the third target threshold, the relevant parameters in the target image anomaly recognition model are adjusted, the adjusted target image anomaly recognition model is determined as the target image anomaly recognition model, the updated tongue training round value is determined as the initial tongue training round value, and the tongue information of the training samples is selected from the tongue information of each unselected training sample in the training sample tongue information set for re-execution of the third training step.

9. A brainwave abnormality alarm device based on electroencephalogram (EEG) signals and tongue images, comprising: The acquisition unit is configured to acquire brainwave test information, brainwave test images, and target tongue images; The first transmitting unit is configured to transmit the brainwave test information and the brainwave test image to a user terminal to generate an initial brainwave test signal, and to receive the initial brainwave test signal from the user terminal. An anomaly detection unit is configured to perform anomaly detection processing on the initial EEG test signal to obtain target brainwave abnormality feature information. The initial EEG test signal includes an initial brainwave signal set. The anomaly detection processing on the initial EEG test signal to obtain the target brainwave abnormality feature information includes: performing noise reduction processing on each initial brainwave signal in the initial brainwave signal set included in the initial EEG test signal to generate a noise-reduced brainwave signal, obtaining a noise-reduced brainwave signal set; selecting a noise-reduced brainwave signal that meets the target waveband condition from the noise-reduced brainwave signal set as the target brainwave signal; performing feature extraction processing on the target brainwave signal to obtain target brainwave feature information; and inputting the target brainwave feature information into a pre-trained brainwave anomaly detection model to obtain the target brainwave abnormality feature information. The recognition unit is configured to perform recognition processing on the target tongue image to obtain abnormal feature information of the target tongue; The input unit is configured to input the target brainwave abnormality feature information and the target tongue abnormality feature information into a pre-trained feature fusion model to obtain brainwave abnormality information; The second sending unit is configured to send the abnormal brainwave information to the alarm terminal for alarm operation.

Images