Local sleep discrimination and model training method and device, equipment and storage medium

Abstract

The embodiment of the application discloses a local sleep discrimination and model training method, device, equipment and storage medium, mainly comprising: cutting the brain electrical signals of a predetermined length of a user collected in real time into multiple brain electrical signal slices, and stacking each brain electrical signal slice in time sequence to obtain multiple layers of brain electrical signals; performing time-frequency conversion on the multiple layers of brain electrical signals to obtain multiple layers of brain electrical spectrum data, and performing differential entropy conversion on each layer of brain electrical spectrum data to obtain multiple layers of differential entropy matrices; and using a pre-trained local sleep discrimination model to discriminate whether the user is in a local sleep state according to the multiple layers of differential entropy matrices. The embodiment of the application can improve the discrimination sensitivity and accuracy of the local sleep state of the user.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bioelectric signal processing, and particularly relates to a local sleep discrimination method and model training method, device, equipment and storage medium. BACKGROUND

[0002] People will appear in a local sleep state due to fatigue and other reasons. When in a local sleep state, some parts of the brain are found to be in a slow wave sleep state, while the remaining parts are in a wakeful state. When people are in a local sleep state, they usually appear blank, slow reaction and impulsive, which may cause certain harm. For example, when people drive vehicles, it may cause traffic accidents. Therefore, local sleep monitoring is of great significance for safe driving.

[0003] The prior art uses a learning model to analyze electroencephalogram signals to monitor and discriminate the local sleep state. However, when the prior art is used for specific electroencephalogram signal local sleep monitoring tasks, the input of the learning model is the original electroencephalogram signal. However, direct input of the original electroencephalogram signal cannot well analyze the weak state mutation of the electroencephalogram signal. SUMMARY

[0004] The embodiments of the present application provide a local sleep discrimination method and model training method, device, equipment and storage medium, which can capture the weak state change of the electroencephalogram signal, improve the discrimination sensitivity and accuracy of the user's local sleep state, take alarm measures in time, and avoid the harm caused by the user in the local state.

[0005] In a first aspect, the embodiments of the present application provide a local sleep discrimination method, comprising: cutting a predetermined length of electroencephalogram signal of a user collected in real time into a plurality of electroencephalogram signal slices, and stacking each electroencephalogram signal slice in time sequence to obtain a plurality of layers of electroencephalogram signals; performing time-frequency conversion on the plurality of layers of electroencephalogram signals to obtain a plurality of layers of electroencephalogram frequency spectrum data, and performing differential entropy transformation on each layer of electroencephalogram frequency spectrum data to obtain a plurality of layers of differential entropy matrices; and using a pre-trained local sleep discrimination model to discriminate whether the user is in a local sleep state according to the plurality of layers of differential entropy matrices.

[0006] In a second aspect, an embodiment of the present application provides a local sleep discrimination model pre-training method, comprising: cutting a plurality of training electroencephalogram signals of a predetermined length of a subject into a plurality of training electroencephalogram signal slices, and stacking each training electroencephalogram signal slice in time sequence to obtain a plurality of layers of training electroencephalogram signals; performing time-frequency conversion on the plurality of layers of training electroencephalogram signals to obtain a plurality of layers of training electroencephalogram frequency spectrum data, and performing differential entropy conversion on each layer of training electroencephalogram frequency spectrum data to obtain a plurality of layers of training differential entropy matrices; and inputting the plurality of layers of training differential entropy matrices into a local sleep discrimination model to be trained, and guiding the output of the local sleep discrimination model to be trained according to whether the corresponding subject is in a local sleep state, training the local sleep discrimination model to be trained, and obtaining a pre-trained local sleep discrimination model.

[0007] In a third aspect, an embodiment of the present application provides a local sleep discrimination device, comprising: an electroencephalogram signal acquisition, cutting and stacking module, configured to cut a predetermined length of electroencephalogram signal of a user acquired in real time into a plurality of electroencephalogram signal slices, and stack each electroencephalogram signal slice in time sequence to obtain a plurality of layers of electroencephalogram signals; an electroencephalogram signal conversion module, configured to perform time-frequency conversion on the plurality of layers of electroencephalogram signals to obtain a plurality of layers of electroencephalogram frequency spectrum data, and perform differential entropy conversion on each layer of electroencephalogram frequency spectrum data to obtain a plurality of layers of differential entropy matrices; and a discrimination module, configured to use a pre-trained local sleep discrimination model to discriminate whether the user is in a local sleep state according to the plurality of layers of differential entropy matrices.

[0008] In a fourth aspect, an embodiment of the present application provides a local sleep discrimination model pre-training device, comprising: a training electroencephalogram signal acquisition, cutting and stacking module, configured to cut a plurality of training electroencephalogram signals of a predetermined length of a subject into a plurality of training electroencephalogram signal slices, and stack each training electroencephalogram signal slice in time sequence to obtain a plurality of layers of training electroencephalogram signals; a training electroencephalogram signal conversion module, configured to perform time-frequency conversion on the plurality of layers of training electroencephalogram signals to obtain a plurality of layers of training electroencephalogram frequency spectrum data, and perform differential entropy conversion on each layer of training electroencephalogram frequency spectrum data to obtain a plurality of layers of training differential entropy matrices; and a pre-training module, configured to input the plurality of layers of training differential entropy matrices into a local sleep discrimination model to be trained, and guide the output of the local sleep discrimination model to be trained according to whether the corresponding subject is in a local sleep state, train the local sleep discrimination model to be trained, and obtain a pre-trained local sleep discrimination model.

[0009] In a fifth aspect, an embodiment of the present application further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor implements the local sleep discrimination method or the local sleep discrimination model pre-training method according to any one of the embodiments of the present application when executing the program.

[0010] In a sixth aspect, an embodiment of the present application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the local sleep discrimination method or the local sleep discrimination model pre-training method according to any one of the embodiments of the present application.

[0011] The local sleep discrimination method and model training method, device, equipment and storage medium provided by the embodiments of the present application can extract features of real-time electroencephalogram signals of a user through cutting and stacking and differential entropy transformation operations, and can discriminate whether the user is in a local sleep state according to the extracted electroencephalogram signal features by using a model, so as to capture weak state changes of the electroencephalogram signals caused by local sleep, thereby improving the discrimination sensitivity and accuracy of the local sleep state of the user, and further facilitating timely taking of alarm measures and the like to avoid harm caused by the user in the local state. BRIEF DESCRIPTION OF DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the present application, the following will briefly introduce the drawings needed in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0013] Figure 1 is a flowchart of the local sleep discrimination method provided by the embodiments of the present application;

[0014] Figure 2 is another flowchart of the local sleep discrimination method provided by the embodiments of the present application;

[0015] Figure 3 is another flowchart of the local sleep discrimination method provided by the embodiments of the present application;

[0016] Figure 4 is a flowchart of the local sleep discrimination model pre-training method provided by the embodiments of the present application;

[0017] Figure 5 is a structural diagram of the local sleep discrimination device provided by the embodiments of the present application;

[0018] Figure 6 is a structural diagram of the local sleep discrimination model pre-training device provided by the embodiments of the present application;

[0019] Figure 7 is a structural diagram of the electronic device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0020] The application will be described in further detail below with reference to the drawings and embodiments. It is to be understood that the specific embodiments described herein are merely illustrative of the application and are not intended to limit the application. In addition, it should be noted that only the parts related to the application are shown in the drawings for ease of description.

[0021] Figure 1 A flowchart of a local sleep discrimination method provided by an embodiment of the application is shown in FIG. 1. The method can be performed by a local sleep discrimination device provided by an embodiment of the application, which can be implemented in software and / or hardware. In a specific embodiment, the device can be integrated in an electronic device, such as a computer, a server, a vehicle client, etc. The following embodiments will be described by way of example with the device integrated in an electronic device. Referring to FIG. 1, the method can specifically include the following steps: Figure 1

[0022] At step 101, the predetermined length of the real-time acquired EEG signal of the user is cut into multiple EEG signal slices, and each EEG signal slice is stacked in time sequence to obtain multiple layers of EEG signals. This can facilitate time-frequency conversion of the multiple layers of EEG signals and differential entropy transformation to obtain a multiple layer differential entropy matrix, and further use a model to determine whether the user is in a local sleep state according to the differential entropy matrix.

[0023] Specifically, the user can be a vehicle driver. If the vehicle driver is in a local sleep state, the harm caused to himself and others can be greater, and more people can be affected. Therefore, it is necessary to monitor and determine the local sleep state of the vehicle driver in real time.

[0024] Specifically, the EEG signal can be an EEG signal with a large change in the alpha wave band.

[0025] Optionally, the predetermined length of the EEG signal is 1 minute long.

[0026] Specifically, the EEG signal of the vehicle driver can be collected and processed in real time every minute during the operation of the vehicle.

[0027] Optionally, after collecting the original EEG signal of the user for a predetermined length of time, the electrooculogram and electromyogram artifacts, line noise, and invalid signals caused by invalid or local damage of the electrode are removed to obtain the EEG signal.

[0028] Specifically, the process of removing the electrooculogram and electromyogram artifacts can include: sampling the original EEG signal at a frequency of 1200 Hz, then down-sampling to a frequency of 200 Hz, and removing the eye movement and electromyogram artifacts using a meta-learning adaptive method. ​

[0029] Specifically, the original electroencephalogram signal can be filtered by a digital filter. The digital filter can be a band-pass filter with a frequency range of 0.5-60 Hz.

[0030] Specifically, the number of the plurality of electroencephalogram signal slices can be set according to a predetermined time length and subsequent processing.

[0031] Optionally, the process of cutting the electroencephalogram signal of the predetermined time length of the user collected in real time into a plurality of electroencephalogram signal slices includes cutting the electroencephalogram signal of the 1-minute time length every 10 seconds to obtain 6 electroencephalogram signal slices.

[0032] Optionally, the process of stacking the electroencephalogram signal slices in time sequence to obtain a plurality of layers of electroencephalogram signals includes stacking the 6 electroencephalogram signal slices in time sequence to obtain 6 layers of electroencephalogram signal slices.

[0033] In step 102, the multi-layer electroencephalogram signal is subjected to time-frequency conversion to obtain multi-layer electroencephalogram spectrum data, and each layer of electroencephalogram spectrum data is subjected to differential entropy transformation to obtain a multi-layer differential entropy matrix. The differential entropy matrix obtained contains distribution information of the main transformation of the frequency domain feature, which can provide a more accurate electroencephalogram transformation reference for the local sleep discrimination model.

[0034] Optionally, the process of converting the multi-layer electroencephalogram signal to obtain multi-layer electroencephalogram spectrum data includes subjecting the multi-layer electroencephalogram signal to fast Fourier transform to realize conversion from time domain to frequency domain. Specifically, 256-point fast Fourier transform can be performed.

[0035] Optionally, after the time-frequency conversion and differential entropy transformation of each layer of electroencephalogram signal are completed, a differential entropy matrix of one layer is obtained, and then the next layer is operated.

[0036] Specifically, the dimension of each layer of differential entropy matrix represents the number of electroencephalogram signal acquisition channels, i.e., the number of leads of the acquisition device and the number of frequency points of fast Fourier transform, and the value of each layer of differential entropy matrix represents the differential entropy value of a certain channel at a certain frequency point.

[0037] In step 103, a pre-trained local sleep discrimination model is used to discriminate whether the user is in a local sleep state according to the multi-layer differential entropy matrix. The distribution information of the main transformation of the frequency domain feature contained in the differential entropy matrix can accurately and effectively identify the local sleep state of the user.

[0038] Specifically, the local sleep discrimination model can select a neural network model or other models as needed.

[0039] Optionally, the local sleep discrimination model is a meta-learning model, and the meta-learning model can be a long short-term memory network model based on meta-learning.

[0040] Specifically, the meta-learning model can adopt the model strategy shown in Table 1.

[0041] Model Meta Learning Neural Network LSTM Long Short Term Memory Network Activation Function Sigmoid Function Gradient Descent Principle LSTM Forget Gate Gradient Descent Optimization Method Maximum / Minimum Likelihood Method Variable Threshold Offset Coefficient Derivative Calculation Judgment Method Bisection Method

[0042] Table 1

[0043] Specifically, the meta-learning model capable of adaptive learning has unique advantages in processing massive data and different state judgments. By constructing a meta-learning training and testing module, the change state characteristics of the input electroencephalogram information can be obtained, and good judgment effects can be achieved. The use of the meta-learning model for monitoring and discriminating local sleep can further improve the accuracy of local sleep state discrimination.

[0044] The method for discriminating local sleep will be further introduced below, as shown in Figure 2 Step 103 in Figure 1 may include the following steps:

[0045] Step 1031, inputting the multi-layer differential entropy matrix into the local sleep discrimination model, the local sleep discrimination model comprising a first model part and a second model part.

[0046] Preferably, the local sleep discrimination model is a long short-term memory network (LSTM meta-learning model) based on meta-learning, the first model part can be a training part model of the local sleep discrimination meta-learning model, and the second model part can be a testing part model of the local sleep discrimination meta-learning model.

[0047] Step 1032, analyzing the multi-layer differential entropy matrix by using the first model part to obtain a wake electroencephalogram feature label value representing a user's wake state and a local sleep related feature analysis parameter.

[0048] Specifically, the training part model of the pre-trained local sleep discrimination meta-learning model can be used to further train and optimize the local sleep related feature analysis parameter of the pre-trained local sleep discrimination model according to the multi-layer differential entropy matrix, and extract the wake electroencephalogram feature.

[0049] In an optional specific example of the present application, the process of analyzing the multi-layer differential entropy matrix by using the first model part to obtain a wake electroencephalogram feature label value representing a user's wake state and a local sleep related feature analysis parameter comprises: calculating the offset coefficient of each layer differential entropy matrix relative to the corresponding layer differential entropy matrix of the electroencephalogram signal of the previous predetermined time length, and further deriving the offset coefficient to obtain the reciprocal of the offset coefficient of each layer differential entropy matrix.

[0050] Specifically, the process of calculating the offset coefficient of the differential entropy matrix of each layer relative to the corresponding layer differential entropy matrix of the EEG signal of the previous predetermined duration includes: calculating the offset coefficient of each row in the differential entropy matrix of each layer relative to the corresponding row of the corresponding layer of the EEG signal of the previous predetermined duration to obtain the row offset coefficient of each row in the differential entropy matrix of each layer, and then summing the above row offset coefficients of each layer to obtain the offset coefficient of the differential entropy matrix of each layer.

[0051] In an optional specific embodiment of the present invention, firstly, the i-th layer difference entropy matrix Pi in the above-mentioned multi-layer difference entropy matrix is ​​read, and the offset coefficient Φ is calculated for the i-th layer difference entropy matrix. i For Φ i Taking the derivative yields the offset coefficient Φ of the data frame segment from P0 to P2. i '.

[0052] In an optional specific embodiment of the present invention, the process of analyzing the multi-layer differential entropy matrix using the first model part to obtain the awake EEG feature label value representing the user's awake state and the local sleep-related feature analysis parameters includes: obtaining the awake EEG feature label value and the local sleep-related feature analysis parameters based on the inverse of the offset coefficient of each layer of differential entropy matrix.

[0053] In an optional specific embodiment of the present invention, the number of layers in the aforementioned multi-layer differential entropy matrix is ​​an even number greater than 2. Specifically, the number of layers in the aforementioned multi-layer differential entropy matrix is ​​the same as the number of the aforementioned plurality of EEG signal slices, the number of layers in the aforementioned multi-layer EEG signals, and the number of layers in the aforementioned multi-layer EEG spectral data, and can specifically be 6.

[0054] In an optional specific embodiment of the present invention, the process of obtaining the awake EEG feature label value based on the inverse of the offset coefficient of each layer of the differential entropy matrix includes: dividing the multi-layer differential entropy matrix into a first segmented differential entropy matrix and a second segmented differential entropy matrix with the same number of layers; pairing each layer of the first segmented differential entropy matrix with each layer of the second segmented differential entropy matrix in chronological order to obtain multiple differential entropy matrix layer pairs.

[0055] Optional, such as Figure 3 As shown, for a 6-layer differential entropy matrix obtained from a 1-minute EEG signal, it is divided into a first segmented differential entropy matrix, which includes the differential entropy matrices of layers 1-3 corresponding to the first 3 10s of the aforementioned 1 minute, namely P0, P1, and P2, and a second segmented differential entropy matrix, which includes the differential entropy matrices of layers 4-6 corresponding to the last 3 10s of the aforementioned 1 minute, namely P3, P4, and P5.

[0056] In an optional specific embodiment of the present invention, the process of pairing each layer of the first segmented differential entropy matrix with each layer of the second segmented differential entropy matrix in chronological order to obtain multiple differential entropy matrix layer pairs includes: summing the inverse offset coefficients corresponding to the first segmented differential entropy matrix to obtain a first inverse offset sum, and using the first inverse offset sum to perform a first calculation and calibration operation on the LSTM forgetting unit threshold; summing the inverse offset coefficients corresponding to the second segmented differential entropy matrix to obtain a second inverse offset sum, and using the second inverse offset sum to further calculate and calibrate the LSTM forgetting unit threshold based on the first calculation and calibration operation to obtain the calibrated LSTM forgetting unit threshold f. θ ,like Figure 3 As shown, using the above f θ Each layer of the first segmentation difference entropy matrix is ​​paired with each layer of the second segmentation difference entropy matrix.

[0057] Specifically, the differential entropy matrix layers P0 and P3, P1 and P4, and P2 and P5 can be paired to obtain three differential entropy matrix layer pairs: (P0, P3), (P1, P4), and (P2, P5).

[0058] In an optional specific embodiment of the present invention, the process of obtaining the awake EEG feature label value based on the inverse of the offset coefficient of each layer of the differential entropy matrix includes: analyzing the EEG signal changes corresponding to each differential entropy matrix layer based on the inverse of the offset coefficient of each differential entropy matrix layer to obtain multiple first EEG change label values.

[0059] Specifically, the aforementioned first EEG change label value can represent the magnitude of the change in the differential entropy matrix layer at a later time step relative to the differential entropy matrix layer at the previous time step within each differential entropy matrix layer pair.

[0060] Specifically, the aforementioned first EEG change label value can also represent the similarity score between the differential entropy matrix layer at the later time step and the differential entropy matrix layer at the previous time step in each differential entropy matrix layer pair.

[0061] Specifically, the first EEG change label value L0 can be calculated based on the reciprocal of the offset coefficients v0' and Φ3' of the differential entropy matrix layer pair (P0, P3); the first EEG change label value L1 can be calculated based on the reciprocal of the offset coefficients Φ1' and Φ4' of the differential entropy matrix layer pair (P1, P4); and the first EEG change label value L2 can be calculated based on the reciprocal of the offset coefficients Φ2' and Φ5' of the differential entropy matrix layer pair (P2, P5).

[0062] In optional embodiments of the present application, the process of obtaining the wake EEG feature label value according to the reciprocal of the offset coefficient of each layer of the differential entropy matrix comprises: determining the minimum value in the first EEG change label values as the wake EEG feature label value.

[0063] Specifically, as shown in FIG. 6, the minimum value in the first EEG change label values L0, L1, L2, i.e., the minimum value in MinL0, L1, L2, can be determined as the wake EEG feature label value. Figure 3 i

[0064] Optionally, the process of obtaining the local sleep-related feature analysis parameter according to the reciprocal of the offset coefficient of each layer of the differential entropy matrix comprises: calculating the average value of the reciprocal of the offset coefficient of each layer of the differential entropy matrix, and determining the average value of the reciprocal of the offset coefficient of each layer of the differential entropy matrix as the local sleep-related feature analysis parameter.

[0065] Specifically, the average value of Φ0', Φ1', Φ2', Φ3', Φ4' and Φ5' can be calculated to obtain the local sleep-related feature analysis parameter Φ.

[0066] Step 1033, updating the parameters of the second model part using the local sleep-related feature analysis parameter, and using the second model part with updated parameters to determine whether the user is in a local sleep state according to the multi-layer differential entropy matrix and the wake EEG feature label value.

[0067] Specifically, the training of the local sleep determination meta-learning model is completed by updating the parameters of the second model part using the local sleep-related feature analysis parameter, and then the test part of the trained local sleep determination meta-learning model is used to determine whether the user is in a local sleep state.

[0068] In optional embodiments of the present application, the process of determining whether the user is in a local sleep state according to the multi-layer differential entropy matrix and the wake EEG feature label value comprises: analyzing whether the EEG signal corresponding to each layer of the differential entropy matrix has a change related to local sleep according to the reciprocal of the offset coefficient of each layer of the differential entropy matrix and the local sleep-related feature analysis parameter, to obtain a plurality of second EEG change label values.

[0069] Specifically, the second EEG change label value can represent the magnitude of the change of each layer of the differential entropy matrix related to local sleep.

[0070] Specifically, the second EEG change label value can also represent the similarity score of each layer of the differential entropy matrix with the corresponding differential entropy matrix in the previous predetermined time length.

[0071] ​​Specifically, the second brain electrical change label value L0' corresponding to the offset coefficient reciprocal Φ0' of the differential entropy matrix layer P0 and the local sleep related feature Φ is obtained; the second brain electrical change label value L1' corresponding to the offset coefficient reciprocal Φ1' of the differential entropy matrix layer P1 and the local sleep related feature Φ is obtained; the second brain electrical change label value L2' corresponding to the offset coefficient reciprocal Φ2' of the differential entropy matrix layer P2 and the local sleep related feature Φ is obtained; the second brain electrical change label value L3' corresponding to the offset coefficient reciprocal Φ3' of the differential entropy matrix layer P3 and the local sleep related feature Φ is obtained; the second brain electrical change label value L4' corresponding to the offset coefficient reciprocal Φ4' of the differential entropy matrix layer P4 and the local sleep related feature Φ is obtained; and the second brain electrical change label value L5' corresponding to the offset coefficient reciprocal Φ5' of the differential entropy matrix layer P5 and the local sleep related feature Φ is obtained.

[0072] In optional embodiments of the present application, the process of determining whether the user is in the local sleep state according to the multi-layer differential entropy matrix and the wakefulness brain electrical feature label value comprises: determining the maximum value of the second brain electrical change label values as the local sleep related feature label value.

[0073] Specifically, as shown in Figure 3 , the maximum value MaxL s in the second brain electrical change label values L s , i.e., the maximum value of L0', L1', L2', L3', L4' and L5' can be determined as the local sleep related feature label value.

[0074] In optional embodiments of the present application, the process of determining whether the user is in the local sleep state according to the multi-layer differential entropy matrix and the wakefulness brain electrical feature label value comprises: calculating a label value difference of the local sleep related feature label value minus the wakefulness brain electrical feature label value, and determining whether the user is in the local sleep state according to the label value difference and a preset label value difference threshold.

[0075] Specifically, the preset label value difference threshold can be set according to the use scenario and the specific user.

[0076] Optionally, if the label value difference is greater than the preset label value difference threshold, it is determined that the user is in the local sleep state.

[0077] In optional embodiments of the present application, the preset label value difference threshold can be 0, i.e., as shown in Figure 3 , if MaxL s > MinL0, it is determined that the user is in the local sleep state, and if MaxL s≤MinL0, that is, it is determined that the user is not in a local sleep state.

[0078] In an optional embodiment of the present application, after the local sleep state of the user is determined, a measure to alleviate the local sleep state is taken, for example, an alarm message is sent.

[0079] Specifically, as shown in Figure 3 after the driver of the vehicle is determined to be in a local sleep state, the voice can be turned on to alarm and attract the attention of the driver, or the fragrance can be turned on to alleviate the local sleep state of the driver.

[0080] The embodiment of the present application first uses the multi-layer differential entropy matrix to further train the long short-term memory network based on meta-learning and extract the wakeful electroencephalogram features, and then uses the trained long short-term memory network of meta-learning to determine the local sleep state of the multi-layer differential entropy matrix, which can be based on the distribution information of the frequency domain features mainly transformed by the differential entropy matrix, can acutely capture the weak state changes of the electroencephalogram signal caused by the local sleep, and further improve the discrimination sensitivity and accuracy of the local sleep state of the user.

[0081] Figure 4 A flowchart of a pre-training method of a local sleep determination model provided by an embodiment of the present application is provided. The method can be executed by a local sleep determination model pre-training device provided by an embodiment of the present application. The device can be realized in the form of software and / or hardware. In a specific embodiment, the device can be integrated in an electronic device, such as a computer, a server, etc. The following embodiments will be described by taking the device integrated in an electronic device as an example. Referring to Figure 4 , the method can specifically include the following steps:

[0082] Step 401, the training electroencephalogram signals of the subject collected and acquired for a predetermined time length are cut into a plurality of training electroencephalogram signal slices, and each training electroencephalogram signal slice is stacked in time sequence to obtain a plurality of layers of training electroencephalogram signals.

[0083] Step 402, time-frequency conversion is performed on the multi-layer training electroencephalogram signals to obtain multi-layer training electroencephalogram frequency spectrum data, and differential entropy transformation is performed on each layer of training electroencephalogram frequency spectrum data to obtain a multi-layer training differential entropy matrix.

[0084] Step 403, the multi-layer training differential entropy matrix is input into the local sleep determination model to be trained, and the output of the local sleep determination model to be trained is guided by whether the corresponding subject is in a local sleep state, the local sleep determination model to be trained is trained, and a pre-trained local sleep determination model is obtained.

[0085] In an optional embodiment of the present application, the local sleep discrimination model is a meta-learning model, which can be a long short-term memory network model based on meta-learning.

[0086] In an optional embodiment of the present application, the local sleep discrimination model comprises a first model part and a second model part.

[0087] Specifically, the first model part can be a training model part of the local sleep discrimination meta-learning model, and the second model part can be a test model part of the local sleep discrimination meta-learning model.

[0088] The process of training the local sleep discrimination model by inputting the multi-layer training differential entropy matrix into the local sleep discrimination model to be trained and guiding the output of the local sleep discrimination model to be trained with the state of whether the subject is in local sleep comprises:

[0089] In the training stage, the training model part of the local sleep discrimination meta-learning model extracts the wake EEG feature label value representing the wake state of the subject from the multi-layer training differential entropy matrix, and trains the local sleep related feature analysis parameters.

[0090] In the test stage, the local sleep related feature label value of the subject is obtained according to the training local sleep related feature analysis parameters and the multi-layer differential entropy matrix analysis, and the state of whether the subject is in local sleep is discriminated according to the local sleep related feature label value and the wake EEG feature label value of the subject to obtain a test discrimination result.

[0091] According to the test discrimination result and the state of whether the corresponding subject is in local sleep, the variable threshold and the distribution weight parameter in the model are corrected and optimized based on the stochastic gradient descent minimum principle and the adaptive optimization method until the variable threshold and the distribution weight parameter in the model no longer change.

[0092] The pre-training method of the local sleep discrimination model provided by the embodiment of the present application can train a pre-trained local sleep discrimination model. By using the local sleep discrimination model, the weak state change of the EEG signal caused by local sleep can be captured, thereby improving the discrimination sensitivity and accuracy of the local sleep state of the user, and further facilitating timely taking measures such as alarm to avoid the harm caused by the user in the local state.

[0093] Figure 5 is a structural diagram of the local sleep discrimination device provided by the embodiment of the present application. The device is suitable for executing the local sleep discrimination method provided by the embodiment of the present application. As shown in Figure 5 The device can specifically comprise:

[0094] The electroencephalogram signal collection cutting and stacking module 501 is configured to cut a predetermined length of electroencephalogram signals of a user collected in real time into a plurality of electroencephalogram signal slices, and stack the electroencephalogram signal slices in time sequence to obtain a plurality of layers of electroencephalogram signals. The electroencephalogram signals can be converted in time and frequency and subjected to differential entropy transformation to obtain a plurality of layers of differential entropy matrices, and the user's local sleep state can be further determined by using a model according to the differential entropy matrices.

[0095] The electroencephalogram signal conversion module 502 is configured to convert the electroencephalogram signals in time and frequency to obtain a plurality of layers of electroencephalogram frequency spectrum data, and convert each layer of electroencephalogram frequency spectrum data to obtain a plurality of layers of differential entropy matrices. The differential entropy matrices contain distribution information of the main transformation of the frequency domain features, and can provide more accurate electroencephalogram transformation references for the local sleep discrimination model.

[0096] The discrimination module 503 is configured to determine whether the user is in a local sleep state according to the plurality of layers of differential entropy matrices by using a pre-trained local sleep discrimination model. The distribution information of the main transformation of the frequency domain features in the differential entropy matrices can be used to accurately and effectively identify the local sleep state of the user.

[0097] In optional embodiments of the present application, the local sleep discrimination model includes a first model part and a second model part, and the discrimination module 503 can be specifically configured to input the plurality of layers of differential entropy matrices into the local sleep discrimination model, analyze the plurality of layers of differential entropy matrices by using the first model part to obtain a wake electroencephalogram feature label value representing a wake state of the user and a local sleep related feature analysis parameter, update the parameters of the second model part by using the local sleep related feature analysis parameter, and determine whether the user is in a local sleep state by using the second model part with updated parameters according to the plurality of layers of differential entropy matrices and the wake electroencephalogram feature label value.

[0098] In optional embodiments of the present application, the discrimination module 503 can be specifically configured to calculate an offset coefficient of each layer of differential entropy matrix relative to the corresponding layer of differential entropy matrix of the electroencephalogram signal of the previous predetermined length, and further derive the offset coefficient to obtain the reciprocal of the offset coefficient of each layer of differential entropy matrix; and analyze the wake electroencephalogram feature label value and the local sleep related feature analysis parameter according to the reciprocal of the offset coefficient of each layer of differential entropy matrix.

[0099] In optional embodiments of the present application, the number of layers of the multi-layer differential entropy matrix is an even number greater than 2, and the discrimination module 503 can be specifically configured to: pair each layer of the first segmented differential entropy matrix with each layer of the second segmented differential entropy matrix in the order of the time corresponding thereto to obtain a plurality of differential entropy matrix layer pairs; analyze the change of the electroencephalogram signal corresponding to each differential entropy matrix layer pair according to the reciprocal of the offset coefficient of each differential entropy matrix layer pair to obtain a plurality of first electroencephalogram change label values; and determine the minimum value in the plurality of first electroencephalogram change label values as the wakefulness electroencephalogram feature label value.

[0100] In optional embodiments of the present application, the discrimination module 503 can be specifically configured to: analyze whether the change of the electroencephalogram signal corresponding to each layer of the differential entropy matrix is related to local sleep according to the reciprocal of the offset coefficient of each layer of the differential entropy matrix and the local sleep-related feature analysis parameter to obtain a plurality of second electroencephalogram change label values; determine the maximum value in the plurality of second electroencephalogram change label values as the local sleep-related feature label value; and calculate a label value difference value of the local sleep-related feature label value minus the wakefulness electroencephalogram feature label value, and discriminate whether the user is in a local sleep state according to the label value difference value and a preset label value difference threshold.

[0101] The local sleep discrimination device provided by the embodiments of the present application can extract features of real-time electroencephalogram signals of a user through cutting and stacking and differential entropy transformation operations, and discriminate whether the user is in a local sleep state according to the extracted electroencephalogram signal features by using a model, can capture weak state changes of the electroencephalogram signal caused by local sleep, thereby improving the discrimination sensitivity and accuracy of the local sleep state of the user, and further facilitating timely adoption of alarm measures and the like to avoid harm that may be caused by the user in a local state.

[0102] Figure 6 is a structural diagram of the local sleep discrimination model pre-training device provided by the embodiments of the present application, and the device is suitable for executing the local sleep discrimination model pre-training method provided by the embodiments of the present application. As shown in Figure 6 the device can specifically include:

[0103] The training electroencephalogram signal collection and cutting and stacking module 601 is configured to cut a plurality of training electroencephalogram signals of a subject acquired in a predetermined time length into a plurality of training electroencephalogram slices, and stack each training electroencephalogram slice in a time sequence to obtain a plurality of layers of training electroencephalogram signals.

[0104] The training electroencephalogram signal conversion module 602 is configured to perform time-frequency conversion on the plurality of layers of training electroencephalogram signals to obtain a plurality of layers of training electroencephalogram frequency spectrum data, and perform differential entropy transformation on each layer of training electroencephalogram frequency spectrum data to obtain a plurality of layers of training differential entropy matrices.

[0105] and the pre-training module 603 is used for inputting the multi-layer training differential entropy matrix into the local sleep discrimination model to be trained, and guiding the output of the local sleep discrimination model to be trained with the corresponding state of whether the subject is in local sleep, so as to train the local sleep discrimination model to be trained to obtain the pre-trained local sleep discrimination model.

[0106] The local sleep discrimination model pre-training device provided by the application can train a pre-trained local sleep discrimination model, and the local sleep discrimination model can capture the weak state change of the brain electrical signal caused by local sleep, thereby improving the discrimination sensitivity and accuracy of the local sleep state of the user, and further facilitating timely alarm and other measures to avoid the harm caused by the user in the local state.

[0107] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above functional modules is exemplified, and in actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the above described functions. The specific working process of the above described functional modules can refer to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0108] The embodiment of the application further provides an electronic device, including a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements the local sleep discrimination method or the local sleep discrimination model pre-training method provided in any of the foregoing embodiments when executing the program.

[0109] The embodiment of the application further provides a computer readable medium having a computer program stored thereon, and the program is executed by the processor to implement the local sleep discrimination method or the local sleep discrimination model pre-training method provided in any of the foregoing embodiments.

[0110] The following refers to Figure 7 which shows a structural schematic diagram of a computer system 700 of the electronic device suitable for implementing the embodiment of the application. Figure 7 The electronic device shown is only an example, and should not bring any limitation to the function and use range of the embodiment of the application.

[0111] As Figure 7As shown, the computer system 700 includes a central processing unit (CPU) 701 which can perform various appropriate actions and processes according to programs stored in a read only memory (ROM) 702 or loaded into a random access memory (RAM) 703 from a storage section 708. In the RAM 703, various programs and data required for the operation of the system 700 are also stored. The CPU 701, the ROM 702, and the RAM 703 are connected to each other through a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0112] Connected to the I / O interface 705 are an input section 706 including a keyboard, a mouse, etc.; an output section 707 including a display device such as a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card, a modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the I / O interface 705 as necessary. A removable recording medium 711 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is attached to the drive 710 as necessary, so that a computer program read therefrom is installed into the storage section 708 as necessary.

[0113] In particular, the processes described above with reference to the flow charts can be implemented as a computer software program according to embodiments of the present disclosure. For example, embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer readable medium, the computer program comprising program code for performing the methods illustrated by the flow charts. In such embodiments, the computer program can be downloaded and installed from a network via the communication section 709, and / or installed from the removable recording medium 711. When the computer program is executed by the central processing unit (CPU) 701, the above-described functions defined in the system of the present disclosure are performed.

[0114] It should be noted that the computer-readable medium shown in the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or component. In the present application, the computer-readable signal medium can include a data signal carried in a baseband or as a part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take various forms, including but not limited to an electromagnetic signal, an optical signal or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, which can send, propagate or transmit a program for use by or in conjunction with an instruction execution system, device or component. The program code contained on the computer-readable medium can be transmitted by any suitable medium, including but not limited to wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0115] The flowcharts and block diagrams in the drawings illustrate the possible implementation architectures, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowcharts or block diagrams can represent a module, a program segment or a part of code containing one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur in different order than that shown in the drawings. For example, two blocks that are shown in succession can actually be executed substantially in parallel, and sometimes in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams or flowcharts, and the combination of blocks in the block diagrams or flowcharts, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0116] The modules and / or units described in the embodiments of the present application can be implemented by software, or by hardware, or by a combination of software and hardware. The described modules and / or units can also be arranged in a processor, for example, can be described as: a processor includes an electroencephalogram signal acquisition cutting and stacking module, an electroencephalogram signal conversion module, and a discrimination module; or can be described as: a processor includes a training electroencephalogram signal acquisition cutting and stacking module, a training electroencephalogram signal conversion module, and a pre-training module. In some cases, the names of these modules do not constitute a limitation on the modules themselves.

[0117] As another aspect, the present application also provides a computer readable medium, which can be included in the device described in the above embodiments, or can exist independently without being assembled into the device. The computer readable medium carries one or more programs, when the one or more programs are executed by the device, the device includes: cutting a predetermined length of electroencephalogram signals of a user acquired in real time into a plurality of electroencephalogram signal slices, and stacking each electroencephalogram signal slice in time sequence to obtain a plurality of layers of electroencephalogram signals; performing time-frequency conversion on the plurality of layers of electroencephalogram signals to obtain a plurality of layers of electroencephalogram frequency spectrum data, and performing differential entropy transformation on each layer of electroencephalogram frequency spectrum data to obtain a plurality of layers of differential entropy matrices; and using a pre-trained local sleep discrimination model to discriminate whether the user is in a local sleep state according to the plurality of layers of differential entropy matrices.

[0118] The above detailed description does not constitute a limitation on the scope of protection of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A local sleep discrimination method characterized by, The method comprises: cutting the brain electrical signals of the user collected in real time for a predetermined length of time into a plurality of brain electrical signal slices, and stacking each brain electrical signal slice in time sequence to obtain a plurality of layers of brain electrical signals; performing time-frequency conversion on the plurality of layers of brain electrical signals to obtain a plurality of layers of brain electrical spectrum data, and performing differential entropy transformation on each layer of brain electrical spectrum data to obtain a plurality of layers of differential entropy matrices; and using a pre-trained local sleep discrimination model to discriminate whether the user is in a local sleep state according to the plurality of layers of differential entropy matrices; the local sleep discrimination model comprises a first model part and a second model part; the process of using the pre-trained local sleep discrimination model to discriminate whether the user is in a local sleep state according to the plurality of layers of differential entropy matrices comprises: inputting the plurality of layers of differential entropy matrices into the local sleep discrimination model; using the first model part to analyze the plurality of layers of differential entropy matrices to obtain a wakefulness brain electrical feature label value representing the user's wakefulness state and a local sleep related feature analysis parameter; and updating the parameters of the second model part using the local sleep related feature analysis parameter, and using the second model part with updated parameters to discriminate whether the user is in a local sleep state according to the plurality of layers of differential entropy matrices and the wakefulness brain electrical feature label value; the process of using the first model part to analyze the plurality of layers of differential entropy matrices to obtain a wakefulness brain electrical feature label value representing the user's wakefulness state and a local sleep related feature analysis parameter comprises: calculating the offset coefficient of each layer of differential entropy matrix relative to the corresponding layer of differential entropy matrix of the brain electrical signal of the previous predetermined length of time, and further deriving the offset coefficient to obtain the reciprocal of the offset coefficient of each layer of differential entropy matrix; and analyzing the reciprocal of the offset coefficient of each layer of differential entropy matrix to obtain the wakefulness brain electrical feature label value and the local sleep related feature analysis parameter.

2. The local sleep discrimination method of claim 1, wherein: the number of layers of the plurality of layers of differential entropy matrices is an even number greater than 2; the process of analyzing the reciprocal of the offset coefficient of each layer of differential entropy matrix to obtain the wakefulness brain electrical feature label value comprises: dividing the plurality of layers of differential entropy matrices into a first divided differential entropy matrix and a second divided differential entropy matrix with the same number of layers; pairing each layer of the first divided differential entropy matrix with each layer of the second divided differential entropy matrix in the order of the corresponding time to obtain a plurality of differential entropy matrix layer pairs; analyzing the brain electrical signal changes corresponding to each differential entropy matrix layer pair according to the reciprocal of the offset coefficient of each differential entropy matrix layer pair to obtain a plurality of first brain electrical change label values; and determining the minimum value of the plurality of first brain electrical change label values as the wakefulness brain electrical feature label value.

3. The local sleep discrimination method of claim 2, wherein: the process of analyzing the reciprocal of the offset coefficient of each layer of differential entropy matrix to obtain the local sleep related feature analysis parameter comprises: calculate an average value of the offset coefficient reciprocal of each layer of the differential entropy matrix, and confirm the average value of the offset coefficient reciprocal of each layer of the differential entropy matrix as the local sleep related feature analysis parameter.

4. The local sleep discrimination method of claim 3, wherein The process of discriminating whether the user is in a local sleep state according to the multi-layer differential entropy matrix and the wakefulness electroencephalogram feature label value includes: According to the offset coefficient reciprocal of each layer of the differential entropy matrix and the local sleep related feature analysis parameter, analyze whether the electroencephalogram signal corresponding to each layer of the differential entropy matrix has a change related to local sleep, and obtain a plurality of second electroencephalogram change label values; Determine the maximum value in the plurality of second electroencephalogram change label values as the local sleep related feature label value; and Calculate a label value difference value of the local sleep related feature label value minus the wakefulness electroencephalogram feature label value, and discriminate whether the user is in a local sleep state according to the label value difference value and a preset label value difference value threshold. 5.A method for pre-training a local sleep discrimination model, the local sleep discrimination model being used to implement the local sleep discrimination method according to claim 1, characterized in that, It includes: Cut the acquired training electroencephalogram signals of the subject for a plurality of predetermined time lengths into a plurality of training electroencephalogram signal slices, and stack each training electroencephalogram signal slice in time sequence to obtain a plurality of layers of training electroencephalogram signals; Perform time-frequency conversion on the multi-layer training electroencephalogram signal to obtain multi-layer training electroencephalogram frequency spectrum data, and perform differential entropy transformation on each layer of the training electroencephalogram frequency spectrum data to obtain a plurality of layers of training differential entropy matrices; and Input the multi-layer training differential entropy matrix into a local sleep discrimination model to be trained, and guide the output of the local sleep discrimination model to be trained according to whether the corresponding subject is in a local sleep state, train the local sleep discrimination model to be trained, and obtain a pre-trained local sleep discrimination model.

6. A local sleep discrimination apparatus characterized by comprising: It includes: An electroencephalogram signal acquisition, cutting and stacking module for cutting the real-time acquired electroencephalogram signals of the user for a predetermined time length into a plurality of electroencephalogram signal slices, and stacking each electroencephalogram signal slice in time sequence to obtain a plurality of layers of electroencephalogram signals; An electroencephalogram signal conversion module for performing time-frequency conversion on the multi-layer electroencephalogram signal to obtain multi-layer electroencephalogram frequency spectrum data, and performing differential entropy transformation on each layer of the electroencephalogram frequency spectrum data to obtain a plurality of layers of differential entropy matrices; And A discrimination module for discriminating whether the user is in a local sleep state according to the multi-layer differential entropy matrix by using the pre-trained local sleep discrimination model; The local sleep discrimination model includes a first model part and a second model part; The discrimination module is specifically configured to input the multi-layer differential entropy matrix into the local sleep discrimination model; Use the first model part to analyze the multi-layer differential entropy matrix to obtain a wakefulness electroencephalogram feature label value representing a user's wakefulness state, and a local sleep related feature analysis parameter; and Use the local sleep related feature analysis parameter to update the parameters of the second model part, and use the second model part with updated parameters to discriminate whether the user is in a local sleep state according to the multi-layer differential entropy matrix and the wakefulness electroencephalogram feature label value; The process of analyzing the multi-layer differential entropy matrix by using the first model part to obtain a wake EEG feature label value representing a user's wake state and a local sleep-related feature analysis parameter includes: calculating the offset coefficient of each layer differential entropy matrix relative to the corresponding layer differential entropy matrix of the EEG signal of the previous predetermined time length, and further deriving the offset coefficient of each layer differential entropy matrix to obtain the reciprocal of the offset coefficient of each layer differential entropy matrix; and According to the offset coefficient reciprocal of each layer differential entropy matrix analysis to obtain the wake EEG feature label value and the local sleep-related feature analysis parameter.

7. A partial sleep discrimination model pre-training device for performing the partial sleep discrimination model pre-training method according to claim 5, wherein It includes: The training EEG signal acquisition cutting and stacking module is used for cutting the acquired training EEG signals of the subject into multiple training EEG signal slices, and stacking each training EEG signal slice in time sequence to obtain a multi-layer training EEG signal; The training EEG signal conversion module is used for converting the multi-layer training EEG signal into time-frequency, obtaining multi-layer training EEG frequency spectrum data, and converting each layer of the training EEG frequency spectrum data into a differential entropy matrix to obtain a multi-layer training differential entropy matrix; And The pre-training module is used for inputting the multi-layer training differential entropy matrix into the local sleep discriminant model to be trained, and guiding the output of the local sleep discriminant model to be trained according to whether the corresponding subject is in a local sleep state, and training the local sleep discriminant model to be trained to obtain a pre-trained local sleep discriminant model.

8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to realize the local sleep discriminant method of any one of claims 1 to 4, or the processor executes the program to realize the local sleep discriminant model pre-training method of claim 5.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to realize the local sleep discriminant method of any one of claims 1 to 4, or the program is executed by the processor to realize the local sleep discriminant model pre-training method of claim 5. The program is executed by the processor to realize the local sleep discriminant method of any one of claims 1 to 4, or the program is executed by the processor to realize the local sleep discriminant model pre-training method of claim 5.

Images