Sleep stimulation method and system based on brain and muscle electrical deep learning fusion

Abstract

The application discloses a sleep stimulation method and system based on brain and myoelectricity deep learning fusion, the method comprising the following steps: acquiring bioelectric signals of a tester, and fusing the acquired bioelectric signals; inputting the fused signals into an artificial intelligence network for feature extraction, and identifying a sleep stage according to the extracted features; correcting correlation according to the sleep stage identified by the artificial intelligence network and the current acquired bioelectric signals; and outputting corresponding sound stimulation according to the correction result. The sleep stage identified by the artificial intelligence network is further detected whether it is in a slow wave sleep stage, corresponding sound stimulation is applied according to the brain electric phase for the slow wave sleep stage, and the deep sleep time of a user is prolonged.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of sleep stimulation, and particularly relates to a sleep stimulation method and system based on brain and muscle electrical deep learning fusion. BACKGROUND

[0002] According to relevant data statistics, the World Health Organization statistics found that about 25% of people in the world have sleep problems. Among them, about 20-30% are caused by different sleep disorders, and the proportion in developed countries and cities may be higher.

[0003] Sleep disorders are particularly evident in the elderly, and as age increases, the incidence of insomnia also increases, and the deep sleep time is getting shorter and shorter. Deep sleep is very important for the brain and memory regeneration, and is also crucial for the cardiovascular system. Previous studies have shown that playing specific sounds through earphones at regular intervals while sleeping can improve slow wave sleep (SWS) that characterizes deep sleep, and has achieved good results in sleep laboratories.

[0004] Current research can divide sleep stages according to electroencephalogram data, however, current electroencephalogram signal research largely depends on complex and expensive laboratory settings, and cannot be widely applied to home and personal sleep monitoring. Therefore, it is necessary to realize a portable device and monitoring method with comparable performance to laboratory systems, which has smaller volume, lighter weight, lower working condition requirements, so as to be applied in daily life under the condition of ensuring the collection accuracy and speed meeting the requirements.

[0005] The existing sleep stimulation scheme is mostly based on the user's electroencephalogram signal, and the sleep stage is recognized by a neural network to directly feedback sleep stimulation adjustment. For example, in the patent document with the application number 201910116278.8 and the name of a sleep stimulation method and device, the electroencephalogram signal is classified by a neural network, and the sound stimulation is directly fed back according to the classification result. However, this method is too dependent on the neural network, and it is difficult to know whether the classification result of the neural network is correct or not. Moreover, the individual differences of the electroencephalogram signal are ignored. When the stimulation is applied, it may fail due to the individual differences of the user or the error of the classification result. The user who is in the sleep latency period or the light sleep period may wake up instead. In the patent document with the patent number 201911191990.0 and the name of a deep learning sound stimulation system and method for sleep regulation, the sleep is detected according to the EEG signal, and the optimal sound stimulation is selected according to the sleep stage detected by the neural network to apply one by one. However, this process requires multiple adaptation tests of the tester, and the test process is complicated. The purpose of sleep stimulation is to help the insomnia population quickly enter deep sleep. The characteristics of human sleep are that deep sleep and light sleep are switched back and forth. Obviously, frequent stimulation cannot effectively guide the sleep process of the tester, but may cause the tester to wake up from the stimulation. Therefore, a sleep stimulation method and system based on brain and muscle electrical deep learning fusion are needed, which can accurately determine the sleep stage according to the obtained signal, and give corresponding sound stimulation to promote the tester to quickly enter deep sleep and prolong the deep sleep time. SUMMARY

[0006] Therefore, the purpose of the present application is to provide a sleep stimulation method and system based on brain and muscle electrical deep learning fusion, which can accurately determine the sleep stage according to the obtained biological signal of the tester, and give corresponding sound stimulation to promote the tester to quickly enter deep sleep and prolong the deep sleep time.

[0007] In order to achieve the above purpose, the sleep stimulation method based on brain and muscle electrical deep learning fusion of the present application comprises the following steps:

[0008] Step 1, obtaining the biological signal of the tester, the biological signal including electroencephalogram signal, electrooculogram signal and electromyogram signal; fusing the obtained biological signal;

[0009] Step 2, inputting the fused signal into an artificial intelligence network for feature extraction, and identifying the sleep stage according to the extracted features;

[0010] Step 3, according to the sleep stage identified by the artificial intelligence network, the correlation correction is carried out in combination with the current acquired bioelectric signal;

[0011] Step 4, output the corresponding sound stimulation according to the correction result.

[0012] Further preferably, in step 1, when the acquired bioelectric signal is fused, it includes

[0013] The electroencephalogram signal, electromyogram signal and electrooculogram signal are truncated in a fixed time window, and the truncated electroencephalogram, electromyogram and electrooculogram data are combined into a 12-row (8-dimensional electroencephalogram, 2-dimensional electromyogram and 2-dimensional electrooculogram) multi-column (sampling point number of signal within 30 seconds) two-dimensional vector.

[0014] Further preferably, in step 3, the sleep stage includes five categories: S1, awake state; S2, very light sleep period; S3, light sleep period; S4, medium sleep period or deep sleep period; S5, slow wave sleep state.

[0015] Further preferably, it is judged whether the sleep stage identified by the artificial intelligence network is in the medium sleep period or the deep sleep period, if the duration of the medium sleep period or the deep sleep period exceeds the preset duration, the sleep is further judged whether it is in the slow wave sleep stage, if it does not exceed the preset threshold, the bioelectric signal is continuously collected and the sleep stage is classified.

[0016] Further, preferably, the slow wave sleep stage is judged by the following method:

[0017] It is judged whether the current sleep stage is in the medium sleep period or the deep sleep period, if it is in the medium sleep period or the deep sleep period, it is further judged whether the power value of the delta brain wave, electromyogram and electrooculogram in the electroencephalogram signal is greater than the set threshold; otherwise, it is judged as a non-slow wave sleep stage;

[0018] If the power value of the delta brain wave, electromyogram and electrooculogram is greater than the set threshold, the current sleep stage is judged as a slow wave sleep stage, and sound stimulation is applied according to the phase of the current electroencephalogram signal.

[0019] Further preferably, the method of applying sound stimulation according to the phase of the current electroencephalogram signal is as follows:

[0020] The collected electroencephalogram signal is filtered and pretreated by a high-pass filter of 0.1 Hz;

[0021] The pretreated data is input into a soft PLL algorithm for phase detection, wherein the NOC (numerical controlled oscillator) frequency of the soft PLL is set to 1 Hz;

[0022] The soft PLL algorithm detects that the phase of the brain electrical signal is at a positive 45 degrees, and controls the external stimulation module to output a sound stimulation for enhancing slow wave sleep; if the phase is at a negative phase stage, the external stimulation module is controlled to stop sound output.

[0023] The application also provides a sleep stimulation system based on brain and muscle electrical deep learning fusion, which is used to implement the sleep stimulation method based on brain and muscle electrical deep learning fusion.

[0024] The data acquisition module is used to acquire bioelectric signals, including an electroencephalogram acquisition device, an electromyogram acquisition device, and an electrooculogram acquisition device; the electroencephalogram acquisition device is used to acquire 8 electroencephalogram signals in the 10-20 international standard lead system; the electromyogram acquisition device is used to acquire electromyogram signals at the left and right mandibles of the wearer; and the electrooculogram acquisition device is used to acquire horizontal and vertical electrooculogram signals of the wearer.

[0025] The data buffer module is used to buffer the acquired bioelectric signals in the form of a time queue according to the acquisition time in a first-in-first-out manner, and send them to the data processing module one by one in time sequence.

[0026] The data processing module is used to fuse the acquired bioelectric signals, input the fused signals into an artificial intelligence network for feature extraction, and identify the sleep stage according to the extracted features.

[0027] According to the sleep stage identified by the artificial intelligence network, the correlation is corrected in combination with the currently acquired bioelectric signals; and a sound stimulation control instruction is output according to the correction result.

[0028] The stimulation application module is used to input the corresponding sound stimulation according to the received sound stimulation control instruction.

[0029] Further preferably, the sound stimulation applied by the stimulation application module includes the following modes:

[0030] The stimulation mode is adjusted according to the sound stimulation control instruction, and the stimulation mode includes a continuous stimulation mode or an intermittent stimulation mode.

[0031] The sound type and volume are adjusted according to the sleep stage, the sound type includes natural sound or synthesized sound, and the synthesized sound includes individualized sound output by the artificial intelligence network and white noise, pink noise, and brown noise.

[0032] Further preferably, it also includes a data transmission module, a cloud server, and an APP interaction module.

[0033] The data transmission module is configured to acquire the bioelectric signal in the data processing module and the parameters of the current artificial intelligence network, and send them to the cloud server.

[0034] The cloud server is configured to perform big data processing on the acquired bioelectric signal, update the parameters of the artificial intelligence network, and feed back the optimized parameters after the update to the data processing module to generate a user sleep evaluation report.

[0035] The APP interaction module is configured to acquire the basic information filled by the user and the self-evaluation information filled according to the sleep evaluation report.

[0036] Further preferably, the APP interaction module further comprises establishing a user customized music library, and the user customized music library acquires the preferred sound collected by the user.

[0037] The cloud server is configured to acquire the preferred sound in the customized music library in the APP interaction module, extract the rhythm parameter vector group of the self-selected music according to the multiple groups of music selected by the user, and input the rhythm parameter vector group as the action information A into the reinforcement learning network, wherein the rhythm parameter vector group comprises sound type, rhythm, mode, timbre, melody range, and dynamic range.

[0038] The expected sleep state of the user, the current sleep state of the user, and the power information of the electroencephalogram, electromyogram, and electrooculogram are taken as the state information S of the reinforcement learning network.

[0039] According to the set reward value function, the reward value R of the corresponding state information after the execution of the action information is calculated each time the iteration is performed, wherein the reward value function R

[0040] ;

[0041] wherein The sleep stage identified by the artificial intelligence network outputs five types of sleep states i=1, 2, 3, 4, 5. is the desired sleep state, i.e., the slow wave sleep state. is the electroencephalogram power value, electromyogram power value, and electrooculogram power value. is the electroencephalogram threshold value, electromyogram threshold value, and electrooculogram threshold value. 、 is the adjustment coefficient; j=0, 1, 2.

[0042] The rhythm parameter vector group corresponding to the action information with the maximum reward value is selected, a synthesized sound is synthesized, and the synthesized sound is added with the corresponding state information S, action information A, and reward value R label to form the preferred sound in the customized music library.

[0043] The sleep stimulation method and system based on brain and myoelectric deep learning fusion disclosed in the application, according to the biological electric signals of the tester obtained, the sleep stage recognized by the artificial intelligence network, and further detection whether in the slow wave sleep stage, for the slow wave sleep stage, corresponding sound stimulation is applied according to the electroencephalogram phase, guiding the user to gradually enter deep sleep, which has high degree of fit with the user and is not easy to cause the user to wake up.

[0044] The comprehensive biological electric signals are formed by combining the electroencephalogram, myoelectric and electrooculogram signals together, which is more accurate than the judgment of the user entering the sleep stage by relying on the electroencephalogram signal alone. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 The flowchart of the sleep stimulation method based on brain and myoelectric deep learning fusion provided by the application is provided.

[0046] Figure 2 The structure schematic diagram of the sleep stimulation system based on brain and myoelectric deep learning fusion provided by the application is provided.

[0047] Figure 3 The data interaction schematic diagram of the sleep stimulation system based on brain and myoelectric deep learning fusion provided by the application is provided.

[0048] Figure 4 The data processing flowchart of the sleep detection and sound stimulation provided by the application is provided.

[0049] Figure 5 The schematic diagram of the synthesized music generated based on the reinforcement learning mode and the music generation network provided by the application is provided. DETAILED DESCRIPTION

[0050] The application is further described in detail through the drawings and specific embodiments.

[0051] As Figure 1 shown, the sleep stimulation method based on brain and myoelectric deep learning fusion provided by the embodiment of the application comprises the following steps:

[0052] Step 1, obtaining the biological electric signals of the tester, the biological electric signals comprising electroencephalogram signals, electrooculogram signals and myoelectric signals; fusing the obtained biological electric signals;

[0053] Step 2, inputting the fused signals into the artificial intelligence network for feature extraction, and identifying the sleep stage according to the extracted features;

[0054] Step 3, according to the sleep stage recognized by the artificial intelligence network, correcting the correlation in combination with the currently obtained biological electric signals;

[0055] Step 4, outputting the corresponding sound stimulation according to the correction result.

[0056] Further preferably, in step 1, when the acquired bioelectric signals are fused, the following is included

[0057] The electroencephalogram signal, electromyogram signal and electrooculogram signal are time-truncated according to a fixed sliding time window by using a fixed time window to truncate data, and the truncated electroencephalogram, electromyogram and electrooculogram data are combined into a 12-row (8-dimensional electroencephalogram, 2-dimensional electromyogram and 2-dimensional electrooculogram) multi-column (sampling point number of signals within 30 seconds) two-dimensional vector.

[0058] The 8-dimensional electroencephalogram includes electroencephalogram signals acquired at 8 electrodes Fpz, Fp2, AF8, F8, FT8, FT10, T10 and A2 defined according to the 10-20 international standard lead system.

[0059] The sleep stage includes three parts of a wake state, a non-rapid eye movement state (NREM) and a rapid eye movement state (REM), and the non-rapid eye movement state is further divided into four stages: the first stage is a very light sleep period (N1), the second stage is a light sleep period (N2), the third stage is a medium sleep period (N3), and the fourth stage is a deep sleep period (N4).

[0060] In the present application, the entire sleep stage is divided into five categories: S1, a wake state; S2, a very light sleep period; S3, a light sleep period; S4, a medium sleep period or a deep sleep period; and S5, a slow wave sleep state.

[0061] Further preferably, whether the sleep stage identified by the artificial intelligence network is in the S4 category is determined, and if the duration of the sleep stage in the S4 category exceeds a preset duration, it is determined whether the sleep is in the slow wave sleep stage, and if the duration does not exceed the preset threshold, bioelectric signals are continuously collected and the sleep stage is classified.

[0062] As shown in the following table, the slow wave sleep stage is determined by the following method: Figure 4

[0063] It is determined whether the current sleep stage is in the S4 category, and if the sleep stage is in the S4 category, it is further determined whether the power value of the delta brain wave in the electroencephalogram signal is greater than a set threshold value, and otherwise, the sleep stage is determined to be a non-slow wave sleep stage.

[0064] If the power value of the delta brain wave is greater than the set threshold value, the current sleep stage is determined to be a slow wave sleep stage, and a sound stimulus is applied according to the phase of the current electroencephalogram signal.

[0065] ​The classification of the electroencephalogram signal is closely related to the sleep stage, for example, the alpha brain wave (ALPHA) sometimes appears in the brain and sometimes disappears, and it does not always exist. For example, there is no alpha wave in deep sleep; if a person is in an excited state, or fear, anger, there is also no alpha brain wave in the brain. Alpha brain wave appears when falling asleep or waking up (i.e. half asleep and half awake), at this time the body is in a relaxed state, and there is a conscious alertness. Delta brain wave (DELTA) only appears in deep sleep. Theta brain wave (THETA) appears in light sleep. Beta brain wave (BETA) appears in wakefulness, accompanied by attention concentration that needs to be achieved with effort.

[0066] The method of applying sound stimulation according to the phase of the current electroencephalogram signal is as follows:

[0067] The collected electroencephalogram signal is filtered and pretreated by a high-pass filter of 0.1 Hz;

[0068] The pretreated data is input into a soft PLL algorithm for phase detection, wherein the NOC (numerically controlled oscillator) frequency of the soft PLL is set to 1 Hz;

[0069] When the soft PLL algorithm detects that the phase of the electroencephalogram signal is in the positive direction 45 degrees, the external stimulation module is controlled to output sound stimulation for enhancing slow wave sleep; if the phase is in the negative phase stage, the external stimulation module is controlled to stop sound output.

[0070] As shown in Figure 2 and Figure 3 The application also provides a sleep stimulation system based on brain and muscle electrical depth learning fusion, which is used to implement the sleep stimulation method based on brain and muscle electrical depth learning fusion, and comprises a portable data acquisition and intervention device, a cloud processor and an APP interaction module; wherein the portable data acquisition and intervention device comprises a data acquisition module, a data caching module, a data processing module and a stimulation application module.

[0071] The data acquisition module is used to acquire bioelectric signals, and comprises an electroencephalogram signal acquisition device, an electromyogram signal acquisition device and an electrooculogram signal acquisition device; the electroencephalogram signal acquisition device is used to acquire 8 electroencephalogram signals in the 10-20 international standard lead system; the electromyogram signal acquisition device is used to acquire electromyogram signals at the left and right mandibles of the wearer; and the electrooculogram signal acquisition device is used to acquire horizontal and vertical electrooculogram signals of the wearer.

[0072] The data caching module is used to cache the acquired bioelectric signals in the form of time queue according to the acquisition time in the principle of first-in first-out, and send them to the data processing module one by one in time sequence.

[0073] The data processing module is configured to fuse the acquired bioelectric signals, input the fused signals into an artificial intelligence network for feature extraction, and identify a sleep stage according to the extracted features.

[0074] According to the sleep stage identified by the artificial intelligence network, correlation correction is performed in combination with the currently acquired bioelectric signals; and a sound stimulation control instruction is output according to a correction result.

[0075] The stimulation applying module is configured to input a corresponding sound stimulation according to the received sound stimulation control instruction.

[0076] Further preferably, the sound stimulation applied by the stimulation applying module includes the following modes:

[0077] According to the sound stimulation control instruction, a stimulation mode is adjusted, the stimulation mode including a continuous stimulation mode or an intermittent stimulation mode.

[0078] According to the sleep stage, a sound type and a volume are adjusted, the sound type including a natural sound or a synthetic sound; the synthetic sound including an individualized sound output by the artificial intelligence network and a white noise, a pink noise, and a brown noise.

[0079] The stimulation applying module receives a command of the data processing module, starts or stops a pre-set sound stimulation signal according to the command, and plays the stimulation sound through a headset or a corresponding peripheral device.

[0080] Further, the external sound stimulation module has a volume size adjusting capability, can output sound information in different ranges according to a command, and the setting range is 46dB-60dB, which is adjusted in steps of 1dB at the minimum.

[0081] Further, the system further includes a data transmission module, which is configured to acquire bioelectric signals in the data processing module and parameters of the current artificial intelligence network, and send them to a cloud server; through a wireless manner, raw data such as electroencephalogram, electromyogram, and electrooculogram collected by the system and result data analyzed by the data processing module, data evaluated by a user are sent to the cloud server system, so as to further perform optimization analysis.

[0082] In order to avoid that the stimulation sound is too large to affect sleep, the system can perform real-time sound adjustment according to individual hearing conditions of a user during use of the system, and adjust appropriate volumes for different users.

[0083] The specific volume adjustment mode is as follows:

[0084] Step 401: Before the system is used, the user can set the stimulation mode, sound type, and volume of the system through the system setting function provided by the APP user interaction system. The user can directly select the comfortable stimulation mode, sound type, and volume size through the earphone. The configuration will be saved as the user-specific sound stimulation configuration in the system.

[0085] Step 402: When the data processing module detects that the trigger sound stimulation condition is reached for the first time, the user-configured sound stimulation is output. In the case where the user does not configure, the default sound stimulation (continuous stimulation mode, sound type white noise, and volume 50dB) is output.

[0086] Step 403: If the data processing module detects that the trigger sound stimulation condition is reached for 10 consecutive times and all according to the default sound stimulation output, the user is not awakened, then the volume is increased by the minimum step value until the volume reaches the maximum value 60dB; if the user is awakened for 3 consecutive times, the volume is reduced by the minimum step value until the volume reaches the minimum value 46dB; according to the above process, the system will adjust the volume suitable for the user, and at this time, the configuration is saved as the parameter for future sound output.

[0087] The cloud server is used for big data processing according to the acquired bioelectric signals, updating the parameters of the artificial intelligence network, and feeding back the optimized parameters after updating to the data processing module to generate a user sleep evaluation report; the cloud server also sets an online training classification network: when the cloud server collects a certain amount of user bioelectric signal raw data, the cloud will retrain the artificial intelligence two-classification network according to the collected user's own bioelectric signals, generate a more personalized artificial intelligence two-classification network model parameter for the user, and the cloud server will continuously train the classification network according to the received data and download the network parameters to the sleep detection and intervention system through the data transmission module, so as to improve the recognition accuracy of the system.

[0088] The APP interaction module is used to obtain the basic information filled by the user and the self-evaluation information filled according to the sleep evaluation report.

[0089] Further preferably, the APP interaction module further comprises establishing a user customized music library, and the user customized music library obtains the preferred sound collected by the user;

[0090] Further comprising the cloud server obtaining the preferred sound in the customized music library in the APP interaction module, extracting the rhythm parameter vector group of the self-selected music according to the multiple groups of music selected by the user; and inputting the rhythm parameter vector group as action information A into the reinforcement learning network; wherein the rhythm parameter vector group includes sound type, rhythm, mode, timbre, melody range, dynamic range, or volume range;

[0091] The sleep state expected by the user, the current user sleep state, and the power information of the electroencephalogram, electromyogram, and electrooculogram are taken as the state information S of the reinforcement learning network;

[0092] According to the set reward value function, the reward value R of the corresponding state information after the execution of the action information is calculated each time iteration;

[0093] ;

[0094] Wherein The sleep stage output by the artificial intelligence network outputs five types of sleep states i=1, 2, 3, 4, 5; The sleep state expected by the user is the slow wave sleep state; The electroencephalogram power value, electromyogram power value, and electrooculogram power value; The electroencephalogram threshold value, electromyogram threshold value, and electrooculogram threshold value; 、 The adjustment coefficient is j=0, 1, 2;

[0095] The action information corresponding to the rhythm parameter vector group with the maximum reward value is selected, the sound is synthesized, and the synthesized sound is added with the corresponding state information S, action information A, and reward value R label. The preferred sound in the customized music library is formed.

[0096] The action information corresponding to the rhythm parameter vector group with the maximum reward value is selected, the sound is synthesized, or the corresponding rhythm parameter vector group is input into the music generation network to form synthesized music.

[0097] The synthesized sound is added with the corresponding state information S, action information A, and reward value R label. The preferred sound in the customized music library is formed.

[0098] As Figure 5 shown, when the sound stimulation is performed, the following embodiments are adopted:

[0099] Step 501: The sleep state information output by the sleep stage classification network from the collected electroencephalogram, electromyogram, and electrooculogram of the user, the calculated electroencephalogram, electromyogram, and electrooculogram power information, and the sleep stage expected by the user are taken as the state information S input into the reinforcement learning network;

[0100] Step 502: The preferred sound in the customized music library is selected, and the corresponding action information A is extracted according to the label. The reward value R of the currently generated music parameter is calculated according to the state information. The calculation formula of R is as follows:

[0101] ;

[0102] Wherein The sleep state is The slow wave sleep state; (j=0, 1, 2) are electroencephalogram power values, electromyogram power values, and electrooculogram power values; (j=0, 1, 2) are electroencephalogram threshold values, electromyogram threshold values, and electrooculogram threshold values; , is an adjustment coefficient.

[0103] Step 503: When the reward value R is lower than a preset value, adjust the sound type, rhythm, mode, timbre, melody range, dynamic range, or volume range in the action information A; constantly repeat the above steps to constantly optimize the user-specific customized music library.

[0104] Studies have shown that applying corresponding sound stimulation in deep sleep can improve memory, and therefore, the purpose of the present application is to apply corresponding sound stimulation in middle sleep or deep sleep, which can effectively prolong the time of deep sleep and help improve the memory of the user.

[0105] Obviously, the above embodiments are only examples for clearly illustrating but not limiting the embodiments. Based on the above description, other different forms of changes or variations can be made by those skilled in the art. Here, all the embodiments need not and cannot be exhausted. The obvious changes or variations derived therefrom are still within the protection scope of the present application.

Claims

1. A sleep stimulation method based on brain and muscle electrical depth learning fusion, characterized in that, The method comprises the following steps: Step 1, acquiring biological electrical signals of a tester, wherein the biological electrical signals comprise electroencephalogram signals, electrooculogram signals and electromyogram signals; The acquired biological electrical signals are fused; comprising: The acquired biological electrical signals are fused; comprising: The acquired biological electrical signals are fused; comprising: Step 2, inputting the fused biological electrical signals into an artificial intelligence network to extract features, and identifying a sleep stage according to the extracted features; the sleep stage comprises five types: S1, a wake state; S2, a very light sleep period; S3, a light sleep period; S4, a medium sleep period or a deep sleep period; and S5, a slow wave sleep state; determining whether the sleep stage identified by the artificial intelligence network is in the S4 type of medium sleep period or deep sleep period, if the duration of the S4 type exceeds a preset duration, then determining whether the sleep is in the slow wave sleep stage, and if the preset threshold is not exceeded, then continuing to acquire biological electrical signals and classifying the sleep stage; The slow wave sleep stage is determined by the following method: determining whether the current sleep stage is in the medium sleep period or the deep sleep period, if so, then further determining whether the power values of the delta brain waves, the electromyogram and the electrooculogram in the electroencephalogram signals are greater than a set threshold value; if the power values of the delta brain waves, the electromyogram and the electrooculogram are greater than the set threshold value, then determining that the current sleep stage is the slow wave sleep stage, and applying a sound stimulus according to the phase of the current electroencephalogram signals; otherwise, determining that the current sleep stage is not the slow wave sleep stage; Step 3, correcting the relevance according to the sleep stage identified by the artificial intelligence network and the current acquired biological electrical signals; 2. The sleep stimulation method based on brain and muscle electrical depth learning fusion according to claim 1, characterized in that, Step 4, outputting a corresponding sound stimulus according to the correction result. The method of applying a sound stimulus according to the phase of the current electroencephalogram signals is as follows: filtering and preprocessing the acquired electroencephalogram signals through a high-pass filter of 0.1 Hz; inputting the preprocessed data into a soft PLL algorithm for phase detection, wherein the frequency of the numerical control oscillator of the soft PLL algorithm is set to 1 Hz; 3. A sleep stimulation system based on brain and muscle electrical deep learning fusion, characterized in that, when the soft PLL algorithm detects that the phase of the electroencephalogram signals is at a positive 45 degrees, controlling the stimulus application module to output a sound stimulus for enhancing slow wave sleep; if the phase of the electroencephalogram signals is at a negative phase stage, then controlling the stimulus application module to stop sound output. The method for implementing the sleep stimulation method based on brain and electromyogram deep learning fusion according to any one of claims 1-2 comprises a data acquisition module, a data caching module, a data processing module and a stimulus application module; The data acquisition module is used to acquire biological electrical signals, comprising an electroencephalogram signal acquisition device, an electromyogram signal acquisition device and an electrooculogram signal acquisition device; The electroencephalogram signal acquisition device is used to acquire 8 electroencephalogram signals in the 10-20 international standard lead system; the electromyogram signal acquisition device is used to acquire electromyogram signals at the left and right lower jaws of the wearer; The electrooculogram signal acquisition device is used to acquire horizontal and vertical electrooculogram signals of the wearer; The data caching module is used for caching the collected bioelectric signals in the form of time queue according to the time of collection in the principle of first-in first-out, and sending the bioelectric signals to the data processing module one by one in time sequence; The data processing module is used for fusing the acquired bioelectric signals, inputting the fused bioelectric signals into an artificial intelligence network for feature extraction, and identifying the sleep stage according to the extracted features; According to the sleep stage identified by the artificial intelligence network, the correlation is corrected in combination with the current acquired bioelectric signals; and a sound stimulation control instruction is output according to the correction result; The stimulation applying module is used for inputting the corresponding sound stimulation according to the received sound stimulation control instruction.

4. The sleep stimulation system based on brain and muscle electrical depth learning fusion according to claim 3, characterized in that, The sound stimulation applied by the stimulation applying module includes the following modes: Adjusting the stimulation mode according to the sound stimulation control instruction, the stimulation mode including continuous stimulation mode or intermittent stimulation mode; Adjusting the sound type and volume according to the sleep stage, the sound type including natural sound or synthetic sound; the synthetic sound including personalized sound output by the artificial intelligence network and noise class sound such as white noise, pink noise or brown noise.

5. The brain muscle electrical depth learning fusion based sleep stimulation system according to claim 3, wherein, Further comprising a data transmission module, a cloud server and an APP interaction module; The data transmission module is used for acquiring the bioelectric signals in the data processing module and the current parameters of the artificial intelligence network, and sending the bioelectric signals and the parameters to the cloud server; The cloud server is used for performing big data processing according to the acquired bioelectric signals, updating the parameters of the artificial intelligence network, and feeding back the optimized parameters after the update to the data processing module to generate a user sleep evaluation report; The APP interaction module is used for acquiring the basic information filled by the user and the self-evaluation information filled according to the sleep evaluation report.

Images