Dynamic cell regeneration melody generation algorithm and system based on electroencephalogram characteristic transfer learning, namely medium

Abstract

The invention discloses a dynamic cell regeneration melody generation algorithm and system based on electroencephalogram feature transfer learning, namely a medium, and belongs to the crossing field of artificial intelligence method application and bioengineering technology, and the technical scheme is characterized by comprising the following steps: S1, electroencephalogram feature quantification; s2, dynamically mapping sound wave parameters; the method has the advantages that accurate mapping from the electroencephalogram signals to the dynamic cell regeneration melody is achieved for the first time by fusing electroencephalogram feature quantization, transfer learning, reinforcement learning and three-dimensional sound wave modulation technologies, and the blank in the field of physiological signal driven personalized music generation in the prior art is filled.

Description

Technical Field

[0001] This application relates to the intersection of artificial intelligence methods and bioengineering technology, and more specifically, it relates to a dynamic cell regeneration rhythm generation algorithm, system as medium, based on EEG feature transfer learning. Background Technology

[0002] Electroencephalography (EEG), as a key physiological indicator reflecting brain neural activity, is currently mainly used in clinical disease diagnosis (such as epilepsy and sleep disorders) and emotion recognition. Its application in music generation and biological regulation is still in the exploratory stage. Traditional music generation technologies mainly rely on fixed rules (such as music theory knowledge bases) or static machine learning models (such as hidden Markov models), which have the following significant drawbacks:

[0003] 1. Lack of real-time physiological state response capability: Existing algorithms cannot analyze the dynamic characteristics of emotions and attention implied in EEG signals in real time (such as theta wave phase synchronization and alpha wave power changes), resulting in the generated music failing to form a closed-loop interaction with the user's current neural state. For example, traditional sleep aid music often uses fixed-frequency white noise or alpha wave music, which cannot adaptively adjust according to the dynamic changes in EEG characteristics (such as alpha wave attenuation and delta wave enhancement) during the user's transition from wakefulness to deep sleep;

[0004] 2. Insufficient personalization: Existing methods typically pre-set music parameters based on average group data, ignoring the differences in individual EEG characteristics. For example, there are significant differences in the distribution of beta wave power among different individuals in an anxious state, but traditional algorithms cannot eliminate this difference through transfer learning techniques, resulting in a "one-size-fits-all" music generation effect that fails to meet personalized needs.

[0005] 3. Lack of dynamism and diversity: Traditional music generation algorithms are limited by static model architectures, and the generated melodies lack dynamic change mechanisms in the frequency, amplitude, and time dimensions. For example, the frequency and amplitude parameters of existing sleep aid audio are fixed, which cannot simulate the dynamic characteristics of "growth-division-death" in the process of biological cell regeneration, causing users to easily adapt to monotonous stimulation and affecting long-term effects;

[0006] 4. Lack of cross-domain feature transfer technology: EEG features (such as frequency band energy ratio and phase lock value) and music features (such as frequency and amplitude modulation depth) belong to different modalities, and existing technologies lack effective cross-domain mapping methods. For example, the causal relationship between slow waves (δ / θ waves) in EEG signals and low-frequency sound waves (such as 174Hz) that promote cell regeneration has not been systematically modeled, resulting in a high feature transfer error rate.

[0007] In view of this, the inventors have designed a dynamic cell regeneration rhythm generation algorithm based on EEG feature transfer learning, which addresses the shortcomings of insufficient personalization, lack of dynamism, and weak cross-domain transferability. The system is the medium. Summary of the Invention

[0008] To address the shortcomings of existing technologies, the present invention aims to provide a dynamic cell regeneration rhythm generation algorithm and system based on EEG feature transfer learning. Its advantage lies in the fact that by integrating EEG feature quantization, transfer learning, reinforcement learning, and three-dimensional sound wave modulation technology, it achieves for the first time a precise mapping of EEG signals to dynamic cell regeneration rhythms, filling the gap in the field of physiological signal-driven personalized music generation.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a dynamic cell regeneration rhythm generation algorithm based on EEG feature transfer learning, comprising the following steps:

[0010] S1, Quantification of EEG Features:

[0011] S11. Preprocess the raw EEG signal, including 50Hz power frequency filtering, 0.5-100Hz bandpass filtering and independent component analysis (ICA) to remove eye movement artifacts;

[0012] S12. The improved Morlet wavelet transform is used to extract the energy proportions of the five frequency bands δ (1-4Hz), θ (4-8Hz), α (8-12Hz), β (12-30Hz), and γ (30-45Hz). The calculation formula is as follows:

[0013]

[0014] Where sband is the Morlet wavelet scale for the corresponding frequency band;

[0015] S13. Calculate the phase lock value (PLV) between the Fp1 / Fp2 channels in the prefrontal cortex, using the following formula:

[0016]

[0017] Where φ is the instantaneous phase extracted by the Hilbert transform;

[0018] S2, Dynamic mapping of acoustic parameters:

[0019] S21. Based on a preset EEG feature combination threshold, target acoustic parameters are generated through a transfer learning model, wherein the transfer learning model is pre-trained on 10,000+ sets of clinical EEG-cell regeneration effect corresponding data.

[0020] S22. When the θ wave PLV > 0.7 and the γ energy percentage < 15%, a 528Hz ± 2Hz ​​sine wave carrier is generated and superimposed with a 40Hz amplitude modulation signal; when the α wave peak frequency > 10Hz, a 786Hz linear sweep frequency signal (sweep rate ± 5Hz / s) is generated; when the δ / β power ratio > 1.5 for 10s, a 174Hz square wave signal (duty cycle 22%) is generated.

[0021] S3, Closed-loop optimization adjustment:

[0022] S31. Construct a state space S containing PLV, θ / α power ratio, and HRV rate of change, and define the reward function:

[0023] R = w1·ΔCOL1A1 + w2·(1-Anxiety Index)

[0024] Where w1 and w2 are biological effect weighting coefficients;

[0025] S32. The acoustic parameters are updated using a deep Q-network (DQN) algorithm, and the frequency selection strategy is optimized by minimizing the mean square error loss function.

[0026] A further preferred embodiment of the present invention is: the acoustic amplitude modulation algorithm is as follows:

[0027]

[0028] Where Pα(t) is the real-time α-wave power value, α is the resting-state α-wave power baseline value, and Pαbaseline is the resting-state α-wave power baseline value.

[0029] A further preferred embodiment of the present invention is: the function for dynamically adjusting the acoustic envelope curve is:

[0030] E(t) = e -βt ·(1+0.2sin(2πf mod t))

[0031] The attenuation coefficient β is dynamically calculated based on skin impedance, ranging from 0.05 to 0.2, and fmod is the modulation frequency.

[0032] A further preferred embodiment of the present invention: the transfer learning model in S21 employs a Domain Adversarial Neural Network (DANN), which eliminates data bias from different EEG devices through the following steps:

[0033] S211. Construct a network architecture that includes a feature extractor, a domain classifier, and a task classifier;

[0034] S212. Through adversarial training, the EEG features output by the feature extractor satisfy the domain invariance, ensuring that the cross-device parameter mapping accuracy is ≥83.7%.

[0035] In a further preferred embodiment of the present invention, the mapping relationship between the EEG feature combination and the acoustic parameters is verified by Granger causality test, wherein the predictive F value of the theta wave phase synchronization for COL1A1 gene expression is ≥6.32 (p<0.01), and the Pearson correlation coefficient between the α wave power change and VEGF secretion is ≥0.65.

[0036] A further preferred embodiment of the present invention states that the method achieves precise regulation of cell regeneration effects through the following steps:

[0037] When the theta wave PLV > 0.65 and δ / β < 1.2 is detected, 528Hz acoustic stimulation is triggered, which increases the migration speed of fibroblasts by 41% ± 5%.

[0038] By dynamically adjusting the amplitude of the sound wave through alpha wave power, the synthesis of type III collagen can be increased by 2.3 times ± 0.5 times.

[0039] A dynamic cell regeneration rhythm generation system based on EEG feature transfer learning, comprising:

[0040] The EEG feature processing module performs the preprocessing and feature extraction steps described in claim 1, and outputs the energy percentage of the five frequency bands, PLV value and nonlinear features (approximate entropy, Lyapunov exponent);

[0041] The transfer learning mapping module includes a pre-trained deep neural network that outputs target acoustic wave parameters after inputting EEG features. These parameters include carrier frequency, modulation type (AM / FM), amplitude modulation depth, and envelope curve parameters.

[0042] The reinforcement learning optimization module, based on the state space and reward function described in claim 1, adjusts the acoustic parameters in real time through the DQN algorithm to enhance the cell regeneration effect (the migration speed of fibroblasts increases by ≥40% or the synthesis of type III collagen increases by ≥2 times).

[0043] A further preferred embodiment of the present invention: the sound wave generation module supports three-dimensional parameter adjustment, and simultaneously controls the following parameters:

[0044] Frequency dimensions: 528Hz±5Hz sine wave, 786Hz linear sweep (±5Hz / s), 174Hz square wave (duty cycle 22%);

[0045] Amplitude dimension: Dynamic amplitude modulation based on alpha wave power (adjustment range ±30%);

[0046] Time dimension: envelope curve attack time 50-100ms, decay time 200-500ms, duration amplitude 0.8-1.2.

[0047] In a further preferred embodiment of the present invention, the input features of the transfer learning model further include the prefrontal θ-γ cross-frequency coupling strength, and the output parameters include a 40Hz amplitude-modulated signal, which is used to improve the prefrontal neural synchronization (coupling strength is increased by ≥0.3).

[0048] A computer-readable storage medium having a computer program stored thereon, the program being executed by a processor to implement the steps of the algorithm according to any one of claims 1-5, comprising:

[0049] Time-frequency analysis and phase synchronization detection of electroencephalogram (EEG) signals;

[0050] Acoustic parameter generation based on transfer learning;

[0051] Dynamic optimization and adjustment driven by reinforcement learning.

[0052] In summary, the present invention has the following advantages:

[0053] By integrating EEG feature quantization, transfer learning, reinforcement learning, and three-dimensional sound wave modulation technology, the precise mapping of EEG signals to dynamic cell regeneration rhythms has been achieved for the first time, filling the gap in the field of physiological signal-driven personalized music generation. Attached Figure Description

[0054] Figure 1 This is a flowchart illustrating this embodiment;

[0055] Figure 2 This is a schematic diagram illustrating the working principle of this embodiment. Detailed Implementation

[0056] The present invention will be further described in detail below with reference to the accompanying drawings.

[0057] Combination Figure 1 As shown, a dynamic cell regeneration rhythm generation algorithm based on EEG feature transfer learning includes the following steps:

[0058] S1, Quantification of EEG Features:

[0059] S11. Preprocess the raw EEG signal, including 50Hz power frequency filtering, 0.5-100Hz bandpass filtering and independent component analysis (ICA) to remove eye movement artifacts;

[0060] S12. The improved Morlet wavelet transform is used to extract the energy proportions of the five frequency bands δ (1-4Hz), θ (4-8Hz), α (8-12Hz), β (12-30Hz), and γ (30-45Hz). The calculation formula is as follows:

[0061]

[0062] Where sband is the Morlet wavelet scale for the corresponding frequency band;

[0063] S13. Calculate the phase lock value (PLV) between the Fp1 / Fp2 channels in the prefrontal cortex, using the following formula:

[0064]

[0065] Where φ is the instantaneous phase extracted by the Hilbert transform;

[0066] S2, Dynamic mapping of acoustic parameters:

[0067] S21. Based on a preset EEG feature combination threshold, target acoustic parameters are generated through a transfer learning model, wherein the transfer learning model is pre-trained on 10,000+ sets of clinical EEG-cell regeneration effect corresponding data.

[0068] S22. When the θ wave PLV > 0.7 and the γ energy percentage < 15%, a 528Hz ± 2Hz ​​sine wave carrier is generated and superimposed with a 40Hz amplitude modulation signal; when the α wave peak frequency > 10Hz, a 786Hz linear sweep frequency signal (sweep rate ± 5Hz / s) is generated; when the δ / β power ratio > 1.5 for 10s, a 174Hz square wave signal (duty cycle 22%) is generated.

[0069] S3, Closed-loop optimization adjustment:

[0070] S31. Construct a state space S containing PLV, θ / α power ratio, and HRV rate of change, and define the reward function:

[0071] R = w1·ΔCOL1A1 + w2·(1-Anxiety Index)

[0072] Where w1 and w2 are biological effect weighting coefficients;

[0073] S32. The acoustic parameters are updated using a deep Q-network (DQN) algorithm, and the frequency selection strategy is optimized by minimizing the mean square error loss function.

[0074] A further preferred embodiment of the present invention is: the acoustic amplitude modulation algorithm is as follows:

[0075]

[0076] Where Pα(t) is the real-time α-wave power value, α is the resting-state α-wave power baseline value, and Pαbaseline is the resting-state α-wave power baseline value.

[0077] A further preferred embodiment of the present invention is: the function for dynamically adjusting the acoustic envelope curve is:

[0078] E(t) = e -βt·(1+0.2sin(2πf mod t))

[0079] The attenuation coefficient β is dynamically calculated based on skin impedance, ranging from 0.05 to 0.2, and fmod is the modulation frequency.

[0080] A further preferred embodiment of the present invention: the transfer learning model in S21 employs a Domain Adversarial Neural Network (DANN), which eliminates data bias from different EEG devices through the following steps:

[0081] S211. Construct a network architecture that includes a feature extractor, a domain classifier, and a task classifier;

[0082] S212. Through adversarial training, the EEG features output by the feature extractor satisfy the domain invariance, ensuring that the cross-device parameter mapping accuracy is ≥83.7%.

[0083] In a further preferred embodiment of the present invention, the mapping relationship between the EEG feature combination and the acoustic parameters is verified by Granger causality test, wherein the predictive F value of the theta wave phase synchronization for COL1A1 gene expression is ≥6.32 (p<0.01), and the Pearson correlation coefficient between the α wave power change and VEGF secretion is ≥0.65.

[0084] A further preferred embodiment of the present invention states that the method achieves precise regulation of cell regeneration effects through the following steps:

[0085] When the theta wave PLV > 0.65 and δ / β < 1.2 is detected, 528Hz acoustic stimulation is triggered, which increases the migration speed of fibroblasts by 41% ± 5%.

[0086] By dynamically adjusting the amplitude of the sound wave through alpha wave power, the synthesis of type III collagen can be increased by 2.3 times ± 0.5 times.

[0087] Combination Figure 2 As shown, a dynamic cell regeneration rhythm generation system based on EEG feature transfer learning includes:

[0088] The EEG feature processing module performs the preprocessing and feature extraction steps described in claim 1, and outputs the energy percentage of the five frequency bands, PLV value and nonlinear features (approximate entropy, Lyapunov exponent);

[0089] The transfer learning mapping module includes a pre-trained deep neural network that outputs target acoustic wave parameters after inputting EEG features. These parameters include carrier frequency, modulation type (AM / FM), amplitude modulation depth, and envelope curve parameters.

[0090] The reinforcement learning optimization module, based on the state space and reward function described in claim 1, adjusts the acoustic parameters in real time through the DQN algorithm to enhance the cell regeneration effect (the migration speed of fibroblasts increases by ≥40% or the synthesis of type III collagen increases by ≥2 times).

[0091] A further preferred embodiment of the present invention: the sound wave generation module supports three-dimensional parameter adjustment, and simultaneously controls the following parameters:

[0092] Frequency dimensions: 528Hz±5Hz sine wave, 786Hz linear sweep (±5Hz / s), 174Hz square wave (duty cycle 22%);

[0093] Amplitude dimension: Dynamic amplitude modulation based on alpha wave power (adjustment range ±30%);

[0094] Time dimension: envelope curve attack time 50-100ms, decay time 200-500ms, duration amplitude 0.8-1.2.

[0095] In a further preferred embodiment of the present invention, the input features of the transfer learning model further include the prefrontal θ-γ cross-frequency coupling strength, and the output parameters include a 40Hz amplitude-modulated signal, which is used to improve the prefrontal neural synchronization (coupling strength is increased by ≥0.3).

[0096] A computer-readable storage medium having a computer program stored thereon, the program being executed by a processor to implement the steps of the algorithm according to any one of claims 1-5, comprising:

[0097] Time-frequency analysis and phase synchronization detection of electroencephalogram (EEG) signals;

[0098] Acoustic parameter generation based on transfer learning;

[0099] Dynamic optimization and adjustment driven by reinforcement learning.

[0100] Specifically, it includes the following steps:

[0101] Step 1: EEG signal acquisition and preprocessing

[0102] Multi-channel signal acquisition

[0103] A 21-channel EEG cap was used to acquire signals from the Fp1 and Fp2 channels of the prefrontal cortex at a sampling rate of 1000Hz, and HRV (heart rate variability) data was recorded simultaneously.

[0104] A resting-state EEG was collected for 120 seconds as a baseline, and individual-specific EEG baseline values ​​(such as the mean alpha wave power Pαbaseline and the theta wave PLV baseline value) were extracted.

[0105] Signal preprocessing process

[0106] Power frequency and noise suppression: Power supply interference is eliminated through a 50Hz band-stop filter, and the effective frequency band of EEG is preserved through a 0.5-100Hz band-pass filter.

[0107] Artifact removal: Independent component analysis (ICA) is used to separate electrooculography artifacts, and interfering components are removed by identifying spatial distribution characteristics (such as strongly correlated components in the Fp1 / Fp2 channels).

[0108] Step 2: Quantification of EEG Features and Extraction of Causal Features

[0109] Time-frequency energy distribution analysis

[0110] An improved Morlet wavelet transform is applied to the Fp1 channel signal to calculate the energy proportions of the five frequency bands: δ (1-4Hz), θ (4-8Hz), α (8-12Hz), β (12-30Hz), and γ (30-45Hz).

[0111] The wavelet scale ranges from 10 to 128, covering the 1-45 Hz frequency band. The relative energy of each frequency band is obtained by power spectral density normalization (e.g., the energy ratio of theta wave = 22%).

[0112] Phase synchronization calculation

[0113] Perform Hilbert transform on the Fp1 and Fp2 channel signals respectively to extract the instantaneous phases φ1(t) and φ2(t), and calculate the phase lock value (PLV) through a sliding window (1 second):

[0114]

[0115] The PLV value is output in real time, reflecting the functional connectivity of the prefrontal cortex (e.g., resting PLV = 0.62, threshold set to >0.65).

[0116] Causal feature screening

[0117] The predictive power of theta wave PLV on COL1A1 gene expression was verified by Granger causality test (F = 6.32, p < 0.01), and theta wave synchronization was identified as a key feature driving cell regeneration.

[0118] Step 3: Dynamic generation of acoustic parameters based on transfer learning

[0119] Cross-individual parameter mapping model

[0120] Pre-trained Domain Adversarial Neural Network (DANN):

[0121] Feature extractor: Extracts device-independent EEG features (such as frequency band energy ratio and PLV value).

[0122] Label predictor: Maps EEG features to acoustic parameters (such as carrier frequency and amplitude modulation depth).

[0123] Domain classifier: It eliminates the differences in data distribution between different devices / individuals through adversarial training, achieving a cross-dataset mapping accuracy of 83.7%.

[0124] Feature-parameter mapping rules

[0125] Preset parameter rules are triggered based on real-time EEG characteristics:

[0126] Theta wave PLV>0.7 and γ energy<15%: Generates a 528Hz±2Hz sine wave + 40Hz amplitude-modulated signal (promotes COL1A1 expression ↑2.1 times).

[0127] Alpha wave peak value > 10 Hz (forward shift): generates a 786 Hz linear sweep frequency (±5 Hz / s), promoting VEGF secretion ↑37%.

[0128] δ / β power ratio > 1.5 for 10 s: generates a 174 Hz square wave (22% duty cycle), suppressing MMP-1 activity by 29%. Dynamic amplitude and envelope modulation.

[0129] Amplitude Modulation: Adjusting the acoustic wave amplitude based on the real-time power Pα(t) of the α wave:

[0130]

[0131] The modulation depth ranges from 25% to 45%, and the amplitude increases to 35% when the alpha wave power increases by 15%.

[0132] Temporal envelope optimization: Dynamically adjust the exponential decay parameter β (0.05-0.2) based on skin impedance.

[0133] E(t) = e -βt ·(1+0.2sin(2πf mod t))

[0134] The attack time is 50ms, the decay time is 200ms, and the sustained intensity is 0.8, achieving precise spatiotemporal matching of energy.

[0135] Step 4: Closed-loop optimization mechanism based on reinforcement learning

[0136] State space and reward function

[0137] Three-dimensional state space:

[0138]

[0139] It monitors phase synchronization, θ / α energy ratio, and heart rate variability in real time.

[0140] Biologically-oriented reward function:

[0141] R = w1·ΔCOL1A1 + w2·(1-Anxiety Index)

[0142] ΔCOL1A1 is detected in real time by ELISA, and the anxiety index is calculated based on β wave power and the GAD-7 scale.

[0143] Step 5: Sound wave synthesis and verification of biological effects

[0144] Multimodal signal synthesis

[0145] Generate composite sound waves based on mapping parameters:

[0146] S(t) = A(t)·sin(2πf c t)·(1+m·sin(2πf m t))

[0147] carrier frequency f c Amplitude modulation depth m, modulation frequency f m Real-time dynamic adjustment, output via bone conduction headphones.

[0148] Biological effect verification

[0149] Cellular experiments: Fibroblasts showed a 41% increase in migration speed and a 2.3-fold increase in type III collagen synthesis under 528Hz sound wave stimulation (vs. control group).

[0150] Human trials: Anxiety patients showed a 58% reduction in beta wave power, a 4.2-point decrease in GAD-7 score (significant clinical improvement), and a 0.33-point increase in prefrontal theta-γ cross-frequency coupling strength.

[0151] The working process and beneficial effects of this invention are as follows:

[0152] By integrating EEG feature quantification, transfer learning, reinforcement learning, and three-dimensional sound wave modulation technology, this technology has for the first time achieved precise mapping of EEG signals to dynamic cell regeneration rhythms, filling the gap in existing technologies for personalized music generation driven by physiological signals. Its applications include: psychotherapy: generating music that matches the patient's EEG state to assist in the treatment of anxiety, depression, and other psychological problems; brain-computer interface: enabling real-time interaction between EEG signals and the music generation system, providing users with a personalized auditory experience; scientific research: exploring the deep relationship between EEG signals and music perception, promoting interdisciplinary research in neuroscience and artificial intelligence; and artistic creation: providing inspiration for musicians to generate music that matches their brain activity state.

[0153] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the design concept of the present invention should be included within the protection scope of the present invention.

Claims

1. A dynamic cell regeneration rhythm generation algorithm based on EEG feature transfer learning, characterized in that, Includes the following steps: S1, Quantification of EEG Features: S11. Preprocess the raw EEG signal, including 50Hz power frequency filtering, 0.5-100Hz bandpass filtering and independent component analysis (ICA) to remove eye movement artifacts; S12. The improved Morlet wavelet transform is used to extract the energy proportions of the five frequency bands δ (1-4Hz), θ (4-8Hz), α (8-12Hz), β (12-30Hz), and γ (30-45Hz). The calculation formula is as follows: Where s band The Morlet wavelet scale for the corresponding frequency band; S13. Calculate the phase lock value (PLV) between the Fp1 / Fp2 channels in the prefrontal cortex, using the following formula: Where φ is the instantaneous phase extracted by the Hilbert transform; S2, Dynamic mapping of acoustic parameters: S21. Based on a preset EEG feature combination threshold, target acoustic parameters are generated through a transfer learning model, wherein the transfer learning model is pre-trained on 10,000+ sets of clinical EEG-cell regeneration effect corresponding data. S22. When the θ wave PLV > 0.7 and the γ energy percentage < 15%, a 528Hz ± 2Hz ​​sine wave carrier is generated and superimposed with a 40Hz amplitude modulation signal; when the α wave peak frequency > 10Hz, a 786Hz linear sweep frequency signal (sweep rate ± 5Hz / s) is generated; when the δ / β power ratio > 1.5 for 10s, a 174Hz square wave signal (duty cycle 22%) is generated. S3, Closed-loop optimization adjustment: S31. Construct a state space S containing PLV, θ / α power ratio, and HRV rate of change, and define the reward function: R = w1·ΔCOL1A1 + w2·(1-Anxiety Index) Where w1 and w2 are biological effect weighting coefficients; S32. The acoustic parameters are updated using a deep Q-network (DQN) algorithm, and the frequency selection strategy is optimized by minimizing the mean square error loss function.

2. The dynamic cell regeneration rhythm generation algorithm based on EEG feature transfer learning according to claim 1, characterized in that, The acoustic amplitude modulation algorithm is as follows: Where Pα(t) is the real-time α-wave power value, and P is the resting-state α-wave power baseline value. α baseline This represents the baseline value of the resting-state alpha wave power.

3. The dynamic cell regeneration rhythm generation algorithm based on EEG feature transfer learning according to claim 1, characterized in that, The function for dynamically adjusting the acoustic envelope curve is: E(t)=e -βt ·(1+0.2sin(2πf mod t)) The attenuation coefficient β is calculated dynamically based on skin impedance, and ranges from 0.05 to 0.

2. mod The modulation frequency.

4. The dynamic cell regeneration rhythm generation algorithm based on EEG feature transfer learning according to claim 1, characterized in that, The transfer learning model in S21 employs a Domain Adversarial Neural Network (DANN) to eliminate data bias from different EEG devices through the following steps: S211. Construct a network architecture that includes a feature extractor, a domain classifier, and a task classifier; S212. Through adversarial training, the EEG features output by the feature extractor satisfy the domain invariance, ensuring that the cross-device parameter mapping accuracy is ≥83.7%.

5. The dynamic cell regeneration rhythm generation algorithm based on EEG feature transfer learning according to claim 1, characterized in that: The mapping relationship between the EEG feature combination and the acoustic parameters was verified by Granger causality test, in which the predictive F value of the theta wave phase synchronization for COL1A1 gene expression was ≥6.32 (p<0.01), and the Pearson correlation coefficient between the α wave power change and VEGF secretion was ≥0.

65.

6. The dynamic cell regeneration rhythm generation algorithm based on EEG feature transfer learning according to claim 1, characterized in that: The method achieves precise regulation of cell regeneration effects through the following steps: When the theta wave PLV > 0.65 and δ / β < 1.2 is detected, 528Hz acoustic stimulation is triggered, which increases the migration speed of fibroblasts by 41% ± 5%. By dynamically adjusting the amplitude of the sound wave through alpha wave power, the synthesis of type III collagen can be increased by 2.3 times ± 0.5 times.

7. A dynamic cell regeneration rhythm generation system based on EEG feature transfer learning, characterized in that, include: The EEG feature processing module performs the preprocessing and feature extraction steps described in claim 1, and outputs the energy percentage of the five frequency bands, PLV value and nonlinear features (approximate entropy, Lyapunov exponent); The transfer learning mapping module includes a pre-trained deep neural network that outputs target acoustic wave parameters after inputting EEG features. These parameters include carrier frequency, modulation type (AM / FM), amplitude modulation depth, and envelope curve parameters. The reinforcement learning optimization module, based on the state space and reward function described in claim 1, adjusts the acoustic parameters in real time through the DQN algorithm to enhance the cell regeneration effect (the migration speed of fibroblasts increases by ≥40% or the synthesis of type III collagen increases by ≥2 times).

8. The dynamic cell regeneration rhythm generation system based on EEG feature transfer learning according to claim 7, characterized in that, The sound wave generation module supports three-dimensional parameter adjustment and controls the following parameters simultaneously: Frequency dimensions: 528Hz±5Hz sine wave, 786Hz linear sweep (±5Hz / s), 174Hz square wave (duty cycle 22%); Amplitude dimension: Dynamic amplitude modulation based on alpha wave power (adjustment range ±30%); Time dimension: envelope curve attack time 50-100ms, decay time 200-500ms, duration amplitude 0.8-1.

2.

9. A dynamic cell regeneration rhythm generation system based on EEG feature transfer learning according to claim 7, characterized in that: The input features of the transfer learning model also include the prefrontal θ-γ cross-frequency coupling strength, and the output parameters include a 40Hz amplitude-modulated signal, which is used to improve the prefrontal neural synchronization (coupling strength increased by ≥0.3).

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the algorithm according to any one of claims 1-5, including: Time-frequency analysis and phase synchronization detection of electroencephalogram (EEG) signals; Acoustic parameter generation based on transfer learning; Dynamic optimization and adjustment driven by reinforcement learning.

Images