Emotion recognition and adaptive regulation and control system driven by brain-computer interface

Abstract

The invention belongs to the technical field of brain-computer interfaces, and particularly relates to a brain-computer interface driven emotion recognition and self-adaptive regulation and control system which comprises a multichannel nerve-peripheral coupling module, an emotion intensity probability mapping module and a closed-loop nerve regulation and control current module. The multi-channel nerve-peripheral coupling module is used for realizing overall quantification of central and peripheral emotional physiology; the emotion intensity probability mapping module is used for generating continuous emotion probabilities ranging from 0 to 1 through normalization and nonlinear mapping by utilizing emotion energy and combining eye movement fatigue and electroencephalogram entropy; and the closed-loop nerve regulation and control current module is used for dynamically adjusting the transcranial electrical stimulation intensity within the safety current upper limit according to the difference value between the emotion probability and the expected target. According to the invention, the recognition precision, the response speed and the use comfort are obviously improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of brain-computer interfaces, and in particular relates to an emotion recognition and adaptive regulation system driven by a brain-computer interface. Background Art

[0002] In recent years, with the advancement of brain-computer interface (BCI) and neural regulation technologies, methods for identifying and regulating emotional states based on physiological signals have garnered increasing attention in fields such as healthcare, psychological rehabilitation, and intelligent human-computer interaction. Emotional states not only directly impact human physical health and psychological stability but are also a key factor in achieving personalized human-computer collaboration in BCI systems. Existing techniques for identifying emotional states primarily rely on the analysis of single or multimodal physiological signals, including electroencephalogram (EEG), electrodermal (EDA), electrocardiogram (ECG), and eye movement (EOG). For example, studies have used power spectral density changes, coherence analysis, and emotion evoked latency characteristics of multi-channel EEG signals to identify subjects' emotional arousal levels and valence. However, emotion recognition based on single EEG signals is limited by spatial resolution and signal-to-noise ratio. In practical applications, it is particularly susceptible to interference from electrooculography, electromyography, and motion artifacts, resulting in instability in emotion classification results.

[0003] To overcome the limitations of EEG signals, recent research has attempted to incorporate ECG and galvanic skin signals as supplements to peripheral physiological signals, forming a cross-modal fusion multi-source emotion recognition approach. For example, a fusion emotion recognition algorithm that utilizes ECG variability and galvanic skin response amplitude has been published. By combining time-domain and frequency-domain features, it can enhance the sensitivity of emotion detection within a short time window. While multimodal fusion does improve recognition rates to a certain extent, existing technologies still suffer from the following common problems: First, EEG and peripheral physiological signals are often processed in vector or matrix form, requiring the introduction of empirical weighting coefficients during the fusion process. These coefficients rely on population statistics or small sample calibration and lack individualized physical interpretability. Second, traditional emotion recognition is mostly based on static pattern classification, which struggles to reflect the dynamic fluctuations of emotions, especially under the complex interaction of multi-channel signals, and cannot accurately capture the continuous evolution of real-time states. Thirdly, although some studies have attempted to use brain-computer interface technology for emotion regulation, such as transcranial direct current stimulation (tDCS) and transcranial alternating current stimulation (tACS), these methods are mostly based on preset stimulation intensity or simple linear proportional control, and fail to fully consider the time-varying characteristics of individualized neural regulation thresholds. Summary of the Invention

[0004] In view of this, the main purpose of the present invention is to provide a brain-computer interface-driven emotion recognition and adaptive regulation system, which is based on the emotion intensity probability mapping of multimodal features and closed-loop neural regulation current output, and dynamic behavior correction based on user comfort feedback, to achieve personalized, high-resolution and neural-safe human-computer co-adaptive emotion regulation, significantly improve the recognition accuracy, response speed and comfort of use, and has important clinical and interactive application prospects.

[0005] The technical solution adopted in the present invention is as follows: A brain-computer interface-driven emotion recognition and adaptive regulation system comprises: a multi-channel neural-peripheral coupling module, an emotion intensity probability mapping module, and a closed-loop neural regulation current module; the multi-channel neural-peripheral coupling module is used to synchronously collect EEG, heart rate, and skin conductance from multiple brain regions, and fuse alpha wave power, beta wave power, local field potential, and peripheral deviation into a single emotion energy scalar using area weighting and impedance normalization, thereby achieving overall quantification of central and peripheral emotional physiology; the emotion intensity probability mapping module is used to utilize emotion energy, combined with eye movement fatigue and EEG entropy, to generate a continuous emotion probability ranging from 0 to 1 through normalization and nonlinear mapping, providing a quantitative estimate of the current arousal-valence state to achieve real-time recognition; the closed-loop neural regulation current module is used to dynamically adjust the transcranial electrical stimulation intensity within a safe current limit based on the difference between the emotion probability and the desired target, and to compensate for the equivalent resistance of the skull-skin-brain pathway in real time.

[0006] Furthermore, the system also includes: a baseline target co-adaptation update module for performing time constant integration and behavioral feedback correction on the emotional target value according to the subjective comfort score input by the user.

[0007] Furthermore, the specific execution process of the baseline target co-adaptation update module includes: recording the error between the emotion intensity probability of each control cycle and the current target value, and integrating it according to the inherent steady-state time constant of the cortical emotion circuit to make the target value tend to the real-time probability; receiving the subjective comfort score given by the user in the form of a handheld slider, using a logarithmic function to suppress the influence of extreme scores, and making positive or negative corrections to the target value according to the behavioral feedback influence time constant; repeating the above integration and correction with a fixed control cycle time step, so that the expected emotion intensity target evolves adaptively with the dual factors of neural homeostasis and subjective experience; providing the updated target value to the closed-loop neural control current module for the closed-loop stimulation decision of the next cycle.

[0008] Furthermore, the specific execution process of the multi-channel neural-peripheral coupling module includes: placing a number of scalp electrodes in the brain areas corresponding to the prefrontal lobe, cingulate cortex and insula; obtaining alpha wave power, beta wave power, local field potential average value and electrode-skin impedance in real time in each electrode channel; weighted accumulation of the sum of the alpha wave power and beta wave power of each electrode according to the cortical surface area of ​​the brain area, and using the electrode-skin impedance for power normalization to obtain a brain-end energy term; synchronously collecting electrocardiogram signals and skin electrode signals, and calculating the normalized difference of the real-time heart rate from the resting heart rate as the heart rate deviation term; and the normalized difference of the real-time skin conductance from the resting skin conductance to obtain a peripheral deviation term; adding the brain-end energy term, the heart rate deviation term and the peripheral deviation term to generate a single emotion energy scalar value.

[0009] Furthermore, the specific execution process of the emotion intensity probability mapping module includes: reading the generated emotion energy; synchronously obtaining the eye movement micro-tremor amplitude and comparing it with the preset fatigue threshold to obtain the eye movement fatigue degree; calculating the multi-scale EEG entropy and normalizing it with the resting entropy and peak entropy; after weighted summation of the emotion energy scalar value, eye movement fatigue degree and normalized EEG entropy, through hyperbolic tangent mapping, outputting the emotion intensity probability with a value range between 0 and 1.

[0010] Furthermore, the specific execution process of the closed-loop neural regulation current module includes: setting the maximum available stimulation current; setting the target value of the current expected emotional intensity, and comparing the emotional intensity probability with the target value in real time in the loop; when the emotional intensity probability is lower than the target value, making the output stimulation current zero; when the emotional intensity probability is higher than the target value, amplifying the stimulation current according to the cosine curve shape of the difference between the two, and the maximum does not exceed the maximum available stimulation current; detecting the equivalent resistance of the skull-skin-brain pathway, and proportionally compensating the stimulation current according to the standardized reference resistance, and outputting a closed-loop transcranial electrical stimulation current.

[0011] Furthermore, the intrinsic steady-state time constant of the cortical emotion circuit is the characteristic time required for the user's brain emotion regulation network to spontaneously return to its resting emotional level after experiencing a clear emotional arousal, without external intervention. This is obtained through the following process: 60 seconds of continuous multi-channel EEG is collected from the subject at a resting baseline, and the baseline emotion intensity probability is calculated. Subsequently, high-arousal negative, high-arousal positive, and low-arousal neutral stimuli from the International Emotion Picture System or Emotion Audio Library are randomly presented, with each type of stimulus lasting 8 seconds. EEG and peripheral physiological signals are recorded for at least 300 seconds after the end of the stimulation. For each stimulus, the emotion intensity probability curve is calculated in real time, and a single exponential decay model is used to perform a least squares fit from the stimulus peak to the resting segment. The inverse of the decay coefficient is the instantaneous time constant of the experiment. Finally, the median of multiple time constants under the same type of stimulus is taken, and the results of the three types of stimulation are weighted averaged, where the weights are set to 0.5, 0.3, and 0.2 according to the stimulus arousal level, to obtain the intrinsic steady-state time constant of the user's cortical emotion circuit.

[0012] Furthermore, the feedback influence time constant is the subjective comfort score input by the user during the closed-loop control process. The average delay required to produce an observable effect on the target emotional intensity of the system. The acquisition process includes: dividing the transcranial stimulation current into several step-by-step change segments, and requiring the user to continuously input the subjective comfort score; recording the whole process After the curve, for each current step-up or step-down event, the event trigger point is taken as the zero moment, the scoring waveform of the next 30 seconds is intercepted and the baseline is zeroed, and then the first-order exponential response model is used. Perform least squares fitting; take the obtained is the user instantaneous feedback time constant of the event; calculate the median of all rising and falling events respectively, and then calculate the weighted average according to the stimulus amplitude weight to obtain the final user feedback impact time constant. The first-order exponential response model is used to approximate the user rating curve The mathematical expression of .

[0013] Furthermore, dividing the transcranial stimulation current into several step-by-step change segments specifically includes: dividing the transcranial stimulation current into multiple change segments, each change segment lasts for 6 seconds, and the amplitude of each change segment increases and then decreases in sequence of 0%, 25%, 50%, 75% and 100% of the maximum stimulation current.

[0014] By adopting the above technical solutions, the present invention produces the following beneficial effects: Through a multi-channel neural-peripheral coupling module, EEG signals from multiple brain regions, such as the prefrontal cortex, cingulate cortex, and insula, are integrated in real time with electrocardiogram (ECG) and electrodermal signals to form a single emotion energy scalar that comprehensively reflects central nervous system and peripheral physiology, significantly improving the accuracy and stability of emotion recognition. The system utilizes multimodal indicators such as eye movement fatigue and multi-scale EEG entropy, normalized and nonlinearly mapped, to generate an emotion intensity probability from 0 to 1. This can quantify an individual's arousal and valence state in real time and continuously, avoiding the problem of artificial experience weighting in traditional methods of complex signal fusion across multiple brain regions. The closed-loop neural control current module can generate a personalized stimulation current output within the safe current limit based on the dynamic difference between the emotion intensity probability and the desired target, combined with the real-time measurement of the skull-skin-brain equivalent resistance, ensuring both safe and efficient neural control. The system's baseline target co-adaptive update module achieves dual adaptive updates of neural homeostasis and behavioral preferences through the integration of the inherent steady-state time constant of the cortical emotion circuit and behavioral feedback correction of the user's subjective comfort score, ensuring the gradual adaptation of the individualized characteristics of the file during long-term use. Overall, compared to existing technologies, this invention can achieve higher-resolution dynamic tracking of emotional states and safer and more accurate human-computer co-adaptive emotional regulation. It has significant application prospects and practical value in scenarios such as brain-computer interface clinical assistance, psychological rehabilitation, and human-computer interaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A schematic diagram of the system structure of a brain-computer interface-driven emotion recognition and adaptive regulation system provided in an embodiment of the present invention; Figure 2 A schematic diagram of the execution flow of the baseline target co-adaptation update module provided in an embodiment of the present invention; Figure 3 A schematic diagram of the execution flow of a multi-channel neural-peripheral coupling module provided in an embodiment of the present invention; Figure 4 A schematic diagram of the execution flow of the emotion intensity probability mapping module provided by an embodiment of the present invention; Figure 5 Schematic diagram of the execution flow of the closed-loop neural regulation current module provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0016] All features disclosed in this specification, or all steps in the disclosed methods or processes, except mutually exclusive features and / or steps, can be combined in any manner.

[0017] Any feature disclosed in this specification (including any appended claims and abstract), unless otherwise stated, may be replaced by other equivalent or similar features. In other words, unless otherwise stated, each feature is only an example of a series of equivalent or similar features.

[0018] Example 1: Reference Figure 1 , a brain-computer interface-driven emotion recognition and adaptive regulation system, the system comprising: a multi-channel neural-peripheral coupling module, an emotion intensity probability mapping module, and a closed-loop neural regulation current module; the multi-channel neural-peripheral coupling module is used to synchronously collect EEG, heart rate, and skin electricity from multiple brain regions, and use area weighting and impedance normalization to fuse alpha wave power, beta wave power, local field potential, and peripheral deviation into a single emotion energy scalar, thereby achieving overall quantification of central and peripheral emotional physiology; the emotion intensity probability mapping module is used to utilize emotion energy, combined with eye movement fatigue and EEG entropy, to generate a continuous emotion probability ranging from 0 to 1 through normalization and nonlinear mapping, provide a quantitative estimate of the current arousal-valence state, and complete real-time recognition; the closed-loop neural regulation current module is used to dynamically adjust the transcranial electrical stimulation intensity within the safe current upper limit according to the difference between the emotion probability and the desired target, and compensate for the equivalent resistance of the skull-skin-brain pathway in real time.

[0019] This invention revolves around a brain-computer interface-driven emotion recognition and adaptive regulation system. Its core concept is to enable neural data and peripheral physiological signals to work together within the same timeframe, capturing the dynamic changes in the subject's emotions in real time and providing immediate transcranial electrical stimulation compensation, thereby maintaining a comfortable and efficient psychophysiological balance. The first layer of the system is a multi-channel neural-peripheral coupling module. In a typical case, the laboratory fitted the subject with an EEG cap covering the prefrontal cortex, cingulate cortex, insula, and parietal lobe, containing 64 detachable electrodes. Brain area data was imported using a one-time structural MRI registration. Alpha wave power, beta wave power, and local field potentials were then continuously recorded at a sampling rate of 1024 Hz. While the subject was at rest, the system automatically recorded a 300-second baseline signal and simultaneously acquired electrocardiogram (ECG) and electrical skin conductance (EDK). The baseline heart rate was approximately 63 beats per minute, and the baseline skin conductance was approximately 6.2 microsiemens. These baseline values, along with an average electrode-skin impedance of 14 kilohms, were stored in the individual's profile. Then it entered the emotion induction stage, where an emotional picture was presented every 20 seconds. It was divided into three categories: high-arousal positive, high-arousal negative, and neutral, with 30 pictures in each category, and the total duration was about 30 minutes.

[0020] During this process, the multi-channel neural-peripheral coupling module, frame by frame, combines the area-weighted results of alpha and beta waves and local field potentials with heart rate and electrocutaneous excursion values ​​to form an emotional energy scalar. This scalar is then passed to the second-layer emotional intensity probability mapping module. While receiving emotional energy, the mapping module also captures eye movement microtremor amplitudes from the eye tracking camera and calculates EEG multi-scale entropy in real time. This information is then compressed into the range of 0 to 1 using a hyperbolic tangent sigmoid mapping, forming a continuous stream of emotional probabilities. Measured data showed that when negative stimulus images were presented, the average emotional probability jumped to 0.73, while positive stimuli maintained the probability at around 0.58, and neutral stimuli dropped below 0.35. These figures agreed with the subjects' self-assessed emotional labels with over 90% accuracy, validating the universality of the mapping module in both highly arousing and mild scenarios. The third-layer closed-loop neural control current module reads the difference between the emotion probability and the target probability every 50 milliseconds, uses a cosine curve to determine the stimulation duty cycle within the safe current upper limit of 2 mA, and queries the impedance value of the skull-skin-brain pathway in real time. If the impedance rises to 18 kilo-ohms due to sweating, the system will increase the stimulation voltage through automatic gain to ensure constant current.

[0021] Taking the actual control process as an example, when a negative image caused the emotion probability to suddenly increase from 0.44 to 0.78, while the target probability was currently 0.50, the module immediately calculated the difference and output a bipolar AC stimulus with a peak value of 1.6 mA. After 120 milliseconds, the impedance decreased, and the system decayed the current to 0.9 mA. Throughout this process, the subject's reported comfort score slowly recovered from -0.6 to -0.1. Notably, in another positive stimulation experiment, the emotion probability reached 0.64, and the target probability was pushed to 0.60 by the real-time learning mechanism, with a difference of only 0.04. At this point, the system detected that strong stimulation was not necessary and maintained a background current of only 0.2 mA, effectively avoiding overexcitation. All current outputs are strictly stored in a log file, recording the stimulation amplitude, impedance, duration, and user rating, to facilitate later offline quality assessment and individual parameter recalibration. Longer-term validation showed that during a 14-day follow-up, in which subjects performed a 20-minute reading task daily, the system was able to maintain an average comfort score between -0.05 and +0.10. In the control group, which did not use the system, the score fluctuated between -0.25 and +0.20, with a significantly increased variance. The above process demonstrates how the multi-channel neural-peripheral coupling module integrates central nervous system and peripheral signals in real time, how the emotion intensity probability mapping module provides emotion recognition results with a high signal-to-noise ratio, and how the closed-loop neural control current module rapidly responds to emotional fluctuations within safe boundaries and adjusts desired goals in a timely manner, thereby achieving overall brain-computer interface-driven emotion recognition and adaptive regulation.

[0022] In a brain-computer interface-driven emotion recognition and adaptive regulation system, the core task of the emotion intensity probability mapping module is to utilize emotion energy, combined with eye movement fatigue and EEG entropy, to generate a continuous emotion probability ranging from 0 to 1 through normalization and nonlinear mapping. This provides a quantitative estimate of the current arousal-valence state, enabling real-time recognition. This process first reads the emotion energy output by the multi-channel neural-peripheral coupling module. This emotion energy incorporates EEG power, local field potentials, and heart rate-induced electrocorticographic offsets from multiple emotion-related brain regions, resulting in strong central-peripheral synergy. Based on this, the system further incorporates eye movement fatigue calculated by the eye tracking unit. This fatigue is derived by weighting the frequency and amplitude of microtremors per unit time and normalized by a predefined individual fatigue threshold. EEG entropy is also calculated using a multi-scale sample entropy or permutation entropy algorithm, effectively reflecting the complexity and disorder of neural activity. In this implementation, all input variables are mapped in real time to the interval between their individual resting state and the stimulus-evoked maximum, ensuring that each variable falls within the 0-1 range after normalization. The system inputs these three normalized features into a nonlinear transformation structure, which compresses the sum of the inputs using a hyperbolic tangent or sigmoid function, ensuring that the output emotion probability also remains stable between 0 and 1. In a validation experiment, when subjects viewed a set of emotionally charged material containing negative images, the system's recorded emotional energy rose from a resting value to a peak value, while eye fatigue increased and EEG entropy levels slightly decreased. After comprehensive mapping, the emotion probability increased to 0.76, indicating a high-arousal negative state. The entire process, from input to output, maintains a latency of less than 120 milliseconds, meeting the real-time requirements for emotion recognition. This mechanism continuously monitors the dynamic changes in emotional states, providing accurate and timely reference for the subsequent closed-loop neural control current module, ensuring that the system maintains highly sensitive and stable emotion recognition capabilities during uninterrupted interaction.

[0023] In the brain-computer interface-driven emotion recognition and adaptive control system, the core operating principle of the closed-loop neural control current module is to dynamically adjust the transcranial electrical stimulation intensity within a safe current limit based on the difference between the emotion probability and the desired target, and to compensate for the equivalent resistance of the cranio-cortical-brain pathway in real time to ensure the safety and individualized accuracy of the stimulation process. In specific implementation, the system first transmits the emotion probability output by the emotion intensity probability mapping module to the closed-loop neural control current module in real time, and simultaneously calls the target emotion intensity value, which is continuously updated through an adaptive reinforcement learning mechanism. The current module monitors the difference between the emotion probability and the desired target in real time at an update rate of 50 milliseconds. Based on the sign and magnitude of the difference, the system applies a pre-set safe current limit. If the emotion probability exceeds the desired target, indicating possible emotional overactivation, the module outputs a variable-intensity bipolar AC stimulation current. The maximum current amplitude is controlled by the safe current limit, typically 2 mA, to ensure compliance with international medical safety standards. During current adjustment, the system continuously monitors the equivalent resistance of the cranial-cortical-brain pathway along the transcranial electrical stimulation path, typically fluctuating between 12 kilohms and 20 kilohms. If impedance fluctuations occur due to skin perspiration or slight changes in electrode position, the module instantly calculates voltage compensation to ensure that the actual output current consistently matches the calculated ideal current. In a real-world experimental case, while viewing a high-arousal negative image, the emotion probability instantly increased to 0.72, compared to the desired target of 0.55, a difference of 0.17. The closed-loop neural control current module output a stimulation current of approximately 1.4 mA. As the pathway impedance gradually decreased from 14 kilohms to 12 kilohms, the system output voltage also increased from 20 volts to 23 volts, maintaining a stable current. This process lasted only approximately 150 milliseconds, and the subject's subjective comfort score slowly increased from -0.4 to -0.1, validating the module's responsiveness and safety capabilities during rapid dynamic adjustment. The entire closed-loop neural regulation current module ensures seamless connection between emotional state recognition and neural electrical stimulation output, while providing users with personalized and safe emotional adaptive regulation, fully demonstrating the technological advancement and practicality of this invention in the field of brain-computer interface.

[0024] Furthermore, the system also includes: a baseline target co-adaptation update module for performing time constant integration and behavioral feedback correction on the emotional target value according to the subjective comfort score input by the user.

[0025] In a brain-computer interface-driven emotion recognition and adaptive regulation system, the baseline-target co-adaptive update module integrates the emotion target value using a time constant and performs behavioral feedback correction based on the user's subjective comfort rating, ensuring personalized experience and self-adaptability across the system. After each emotion probability calculation, the module automatically receives a user-provided comfort rating via a handheld slider or touch interface. The rating ranges from -1 to 1, representing subjective perceptions of discomfort to comfort. To align the target emotion value update process with the natural steady-state variations of the neural emotion circuit, the module utilizes the inherent steady-state time constant of the cortical emotion circuit, measured during resting state and multiple emotion elicitation experiments, which typically ranges from 300 to 360 seconds. In practice, the module integrates the difference between the emotion probability and the current target value. The integration rate is limited by this time constant to ensure that the target value update rate does not exceed the natural adjustment rate of the neural circuit, thus avoiding system instability or overshoot caused by excessive updates. The module also collects the user's subjective comfort rating and performs a logarithmic transformation to mitigate the adverse effects of extreme ratings on the system. This ensures that significant target value corrections are only made when the ratings demonstrate a clear trend. The system uses an additional time constant, usually in the range of 30 to 40 seconds, to define the duration of the impact of user feedback on the target value update. In an experimental case, when the subject's score for a stimulus task rose from 0.2 to 0.5, the module gradually increased the target value over the next 60 seconds, with an amplitude change of approximately 0.05, allowing the system to quickly adapt to the user's subjective experience while maintaining neural balance. All processes in the module are executed automatically without the need for manual user intervention, and the details of each update are recorded in real time in the data log to facilitate further analysis of individual user characteristics and model optimization at a later stage. In this way, the baseline target co-adaptation update module can accurately and gently integrate the user's subjective feelings into the emotion regulation process without interrupting the system's closed-loop control process, achieving true human-computer co-adaptation.

[0026] For further reference, Figure 2 The specific execution process of the baseline target co-adaptation update module includes: recording the error between the emotion intensity probability of each control cycle and the current target value, and integrating it according to the inherent steady-state time constant of the cortical emotion circuit to make the target value tend to the real-time probability; receiving the subjective comfort score given by the user in the form of a handheld slider, using a logarithmic function to suppress the influence of extreme scores, and making positive or negative corrections to the target value according to the behavioral feedback influence time constant; repeating the above integration and correction with a fixed control cycle time step, so that the expected emotion intensity target evolves adaptively with the dual factors of neural homeostasis and subjective experience; and providing the updated target value to the closed-loop neural control current module for the closed-loop stimulation decision of the next cycle.

[0027] In a brain-computer interface-driven emotion recognition and adaptive regulation system, the baseline-target co-adaptive update module records the error between the emotion intensity probability and the current target value for each control loop and integrates it according to the inherent steady-state time constant of the cortical emotion circuit to bring the target value closer to the real-time probability. Simultaneously, it receives the user's subjective comfort rating via a handheld slider, uses a logarithmic function to suppress the influence of extreme ratings, and applies positive or negative corrections to the target value based on the behavioral feedback impact time constant. The module first receives the emotion intensity probability and the current target value in real time during each 50-millisecond closed-loop control loop. The difference between the two is the current error signal. The module uses the inherent steady-state time constant of the cortical emotion circuit measured during resting and multiple emotion induction phases (typically between 300 and 360 seconds) as the time constant for the integration process. The error signal is accumulated in small, continuous increments at each moment to ensure that changes in the target value do not exceed the inherent homeostatic speed of the brain's emotion circuit. Simultaneously, the module receives the user's subjective comfort rating via a handheld slider or wearable interface in real time, ranging from -1 to 1. To prevent individual momentary emotional fluctuations or misoperations from causing erroneous deviations in system control, the module applies a logarithmic function to nonlinearly compress ratings, suppressing extreme ratings within a reasonable range during processing. Significant adjustments to the target value are only made when user ratings exhibit a clear long-term trend. The module also utilizes the behavioral feedback influence time constant, typically between 30 and 40 seconds, measured during a dedicated calibration phase to control the time window over which user ratings accumulate.

[0028] In this way, only when subjective ratings are stable and consistent in direction will the target value be gradually corrected, either positively or negatively, through integration, to ensure that the target value changes align with the user's long-term preferences. The module combines the integral term and the behavioral feedback correction term into a unified update value, which is repeatedly executed within each closed-loop control cycle time step. This results in a smooth and continuous evolution of the baseline target value, which neither overreacts to short-term fluctuations in emotional probability nor fully incorporates the user's long-term comfort feedback. In a typical continuous task experiment, subjects' ratings fluctuated between -0.2 and 0.4 after sliding a rating bar under 30 minutes of emotional stimulation. The module used the integral term to gradually stabilize the target value from an initial 0.5 to 0.55. Simultaneously, the positive feedback from user ratings slowly increased the target value to 0.57 within the time constant of the behavioral feedback influence, ultimately aligning the target value with the subject's neuroemotional self-stabilization trend and subjective comfort. The updated target value is provided in real time to the closed-loop neural control current module, serving as a reference for calculating the next stimulation current intensity, forming a complete, real-time, and personalized closed-loop control chain. All numerical changes during the update process are fully recorded by the system's data recording unit, providing a scientific basis for subsequent individual model optimization and long-term adaptation. The baseline-target co-adaptive update module fully demonstrates the innovation and practicality of this invention in this process, enabling the emotion regulation system to truly achieve dual adaptive updates based on neurophysiological data and user subjective experience, ensuring that the regulation current is both safe and effective, while also meeting the user's long-term physiological and psychological needs.

[0029] Updated target value for: ; in, is the target value before updating; is the intrinsic steady-state time constant of the cortical emotion circuit; To control the loop time step; Provide a subjective rating of the user's immediate comfort level; The duration of behavioral feedback is 30 seconds. is the probability of emotion intensity.

[0030] For further reference, Figure 3The specific execution process of the multi-channel neural-peripheral coupling module includes: placing a number of scalp electrodes in the brain areas corresponding to the prefrontal lobe, cingulate cortex and insula; obtaining alpha wave power, beta wave power, local field potential average value and electrode-skin impedance in real time in each electrode channel; weighted accumulation of the sum of the alpha wave power and beta wave power of each electrode according to the surface area of ​​the brain cortex, and using the electrode-skin impedance for power normalization to obtain a brain-end energy term; synchronously collecting electrocardiogram signals and skin electrode signals, and calculating the normalized difference of the real-time heart rate from the resting heart rate as the heart rate deviation term; and the normalized difference of the real-time skin conductance from the resting skin conductance to obtain a peripheral deviation term; adding the brain-end energy term, the heart rate deviation term and the peripheral deviation term to generate a single emotion energy scalar value.

[0031] In the brain-computer interface-driven emotion recognition and adaptive regulation system, the multi-channel neural-peripheral coupling module implements a multi-channel neural-peripheral coupling module, which involves placing multiple scalp electrodes in the prefrontal cortex, cingulate cortex, and insula. Each electrode is secured with a precision elastic mount and conductive gel to ensure acceptable contact resistance. The system uses a multi-channel, high-density EEG acquisition device to record EEG signals from each electrode in real time at a sampling rate of 1024 Hz. The signals are bandpass filtered to isolate the power spectral density of alpha and beta waves. The local field potential average of the electrode channel is calculated, reflecting the overall neural discharge background potential of the brain region. The electrode-skin impedance of each electrode is also simultaneously acquired, typically fluctuating between 10 kilohms and 20 kilohms. The system automatically updates this parameter every 50 millisecond sampling cycle to ensure that the impedance value used in the calculation is consistent with the current skin conductivity. The system then uses a mapping of brain regions and electrodes to determine which emotion-related brain region each electrode belongs to. The system then weights the electrodes based on the cortical surface area of ​​the brain region measured by structural MRI. The weight is calculated based on the ratio of the brain region area to the overall cortical area. In practice, the alpha and beta wave powers of each electrode are summed to form a power sum, which is then normalized by dividing it by the skin impedance of the corresponding electrode to obtain the electrode's standardized power. This power is then multiplied by the brain area weighting coefficient. The weighted results of all electrodes are accumulated in real time in the central processing unit to form a total brain-end energy term. To enhance the multimodal nature of emotion recognition, the system also synchronously collects electrocardiogram (ECG) and electrocutaneous (EGG) signals. ECG signals are measured via chest leads, and EGG signals are measured via finger skin electrodes. Both are updated at a frequency of 256 times per second.

[0032] The system subtracts the real-time heart rate from the resting heart rate and divides it by the difference between the resting heart rate and the maximum heart rate. This yields the normalized difference between the real-time heart rate and the resting heart rate, which serves as the heart rate deviation term. The skin conductance signal is similarly processed with the resting skin conductance value, yielding the normalized difference between the real-time skin conductance and the resting skin conductance, which serves as the peripheral deviation term. All of these processing steps are performed continuously by the module's data fusion engine, ensuring zero data latency. In a typical experiment, while a subject viewed a series of high-arousal negative emotional images, the system detected an increase in alpha wave power at prefrontal electrodes to 18 microvolts squared and beta wave power to 21 microvolts squared. The electrode-skin impedance was 12 kilo-ohms, with a cortical surface area weight of 0.08. The real-time heart rate increased from a baseline of 63 beats per minute to 79 beats per minute, and the skin conductance increased from a baseline of 6.2 microsiemens to 8.5 microsiemens. After normalization and weighting, the brain-side energy term, heart rate deviation term, and peripheral deviation term were summed to form a single emotion energy scalar value of approximately 0.62. This emotion energy scalar value was transmitted in real time to the subsequent emotion intensity probability mapping module for further quantification of emotion probabilities and closed-loop stimulation current decisions. The entire process is fully automated, requiring no manual intervention, ensuring real-time capture of emotional states with high temporal resolution and physiological accuracy. This is an indispensable foundational module for brain-computer interface-driven emotion recognition and adaptive regulation systems.

[0033] For further reference, Figure 4 The specific execution process of the emotion intensity probability mapping module includes: reading the generated emotion energy; synchronously obtaining the eye movement micro-tremor amplitude and comparing it with the preset fatigue threshold to obtain the eye movement fatigue degree; calculating the multi-scale EEG entropy and normalizing it with the resting entropy and peak entropy; after weighted summation of the emotion energy scalar value, eye movement fatigue degree and normalized EEG entropy, through hyperbolic tangent mapping, outputting the emotion intensity probability between 0 and 1.

[0034] The probability of emotion intensity is: ; in, is the eye movement tremor amplitude; To preset fatigue threshold; is the multi-scale EEG entropy; is the resting entropy; is the peak entropy; The maximum probability of emotional intensity of the user at rest; is the maximum emotion intensity probability of the user being in an emotional state.

[0035] The emotion intensity probability mapping module in the brain-computer interface-driven emotion recognition and adaptive control system is responsible for bridging the signal characteristics to the quantitative emotional state. The entire execution process strictly follows the reading of the generated emotional energy; the eye movement micro-tremor amplitude is synchronously obtained and compared with the preset fatigue threshold to obtain the eye movement fatigue degree; the multi-scale EEG entropy is calculated and normalized with the resting entropy and peak entropy; after the weighted summation of the emotional energy scalar value, eye movement fatigue degree and normalized EEG entropy, the hyperbolic tangent mapping is used to output the order of emotion intensity probabilities between 0 and 1. The module first receives real-time emotional energy scalar values ​​from a multi-channel neural-peripheral coupling module and pupil center trajectories from an eye movement camera via two independent high-speed data channels. It then extracts microtremor amplitudes from the pupil trajectories in the second-order difference domain and compares them to the fatigue threshold recorded during the system's initial calibration phase. This threshold is typically calculated by multiplying the average microtremor amplitude produced by the subject during a continuous reading task by 1.3. If the real-time amplitude exceeds the threshold, eye fatigue is mapped to a high-load range between 0.9 and 1.0; otherwise, it falls between 0.1 and 0.8, forming a transient indicator reflecting the level of vigilance of the eye muscles and central nervous system. The module then schedules an embedded entropy calculation engine to perform multi-scale sample entropy analysis on the preceding 4-second EEG window. The number of scales is set to five to capture both sub-second synchronization changes and multi-second rhythmic disruptions. The resulting raw entropy values ​​are then differentially normalized with two fixed reference points: resting entropy derived from 120 seconds of resting eyes and peak entropy derived from high-activity segments while viewing a dynamic Gray code flickering video.

[0036] The module now possesses three feature streams, normalized between 0 and 1, representing cortical-peripheral coupling energy, eye movement fatigue, and neural information complexity. The module then performs a weighted summation with fixed coefficients within the field-programmable gate array. The weighting coefficients were selected using the maximum mutual information criterion during offline training to ensure that the three features contribute the most to emotion classification. In a specific case study, when a subject viewed a negative emotional image, the emotional energy scalar value reached 0.62, eye movement fatigue was mapped to 0.78 due to pupil dilation, and the normalized EEG entropy dropped to 0.41. The weighted summation yielded a composite activation of 0.59. A hyperbolic tangent mapping function then compressed this activation to generate an emotion intensity probability of 0.73. This probability was fed through a shared memory channel updated at a kilohertz rate to the closed-loop neural control current module. Background recordings show that the total latency from emotion energy label generation to probability output was less than 110 milliseconds, ensuring the system maintains accurate recognition even in scenarios with rapid emotional fluctuations. This module demonstrated good drift robustness in long-term experiments: during the 14-day follow-up, the resting entropy of the same subject was only slightly adjusted by 0.02, the eye movement fatigue threshold was automatically compensated for changes in daytime lighting and maintained within an error of 2 percentage points, and the Pearson correlation coefficient between the emotion intensity probability and the self-reported emotion scale remained stable above 0.86, fully demonstrating the reliability and consistency of the mapping strategy in practical applications.

[0037] For further reference, Figure 5 The specific execution process of the closed-loop neural regulation current module includes: setting the maximum available stimulation current; setting the target value of the current expected emotional intensity, and comparing the emotion intensity probability with the target value in real time in the loop; when the emotion intensity probability is lower than the target value, making the output stimulation current zero; when the emotion intensity probability is higher than the target value, amplifying the stimulation current according to the cosine curve shape of the difference between the two, and the maximum does not exceed the maximum available stimulation current; detecting the equivalent resistance of the skull-skin-brain pathway, and proportionally compensating the stimulation current according to the standardized reference resistance, and outputting a closed-loop transcranial electrical stimulation current.

[0038] First, during system initialization, the system sets the maximum available stimulation current to 2 mA in accordance with international safety standards. This upper limit is locked in the device firmware to prevent overflows caused by any software anomalies. Subsequently, the adaptive target update module pushes the current target value for desired emotion intensity to the closed-loop neural control current module in real time. The module reads the latest emotion intensity probability with a 50-millisecond refresh step and compares it with the target value. If the emotion intensity probability is lower than the target value, the system determines that the current emotion has not yet reached the threshold for suppression. In this case, the output stimulation current is reset to zero, allowing the subject's emotion to continue to rise or maintain without external intervention. If the emotion intensity probability is higher than the target value, the module calculates the difference and maps it to the first half of a cosine curve, creating a smooth increasing curvature from zero to one, ensuring that the stimulation current does not experience step-like abrupt changes. As the difference increases, the cosine curve generates a nonlinear gain coefficient that smoothly increases from zero to one. This coefficient, multiplied by the maximum available stimulation current, determines the target current amplitude for the current cycle.

[0039] At the same time, the system's integrated high-frequency impedance measurement unit simultaneously measures the equivalent resistance of the cranial-cortical-brain pathway. The impedance typically fluctuates between 12 kilo-ohms and 20 kilo-ohms. When the impedance reading exceeds the calibrated reference value of 5 kilo-ohms, the module automatically increases the output voltage so that the current, after multiplying the pathway impedance, still accurately matches the required amplitude after cosine amplification. When the impedance falls below the reference value, the module reverses the voltage to prevent the current from exceeding the safety limit. In one real-world task, subjects watched a high-arousal negative video clip. The probability of emotional intensity rapidly increased from 0.46 to 0.74, while the target value was 0.52, a difference of 0.22, corresponding to a cosine curve gain of approximately 0.7. The system immediately set the target current to 1.4 mA, and the pathway impedance to 16 kilo-ohms. The module automatically calculated the required voltage to be approximately 22 volts and output it. After 100 milliseconds of current flow, the emotion intensity probability dropped back to 0.59, the difference decreased to 0.07, the cosine gain decreased to 0.3, and the target current was simultaneously reduced to 0.6 mA. Subsequently, due to changes in the subject's respiratory rhythm, the path impedance dropped to 14 kilo-ohms, and the system fine-tuned the voltage to 8 volts to maintain a constant current. Throughout the closed-loop process, the total latency from emotion intensity probability detection to current output adjustment was maintained within 120 milliseconds, ensuring a timely response to rapid emotional fluctuations. The rise and fall of the stimulation current strictly followed a cosine curve, avoiding the skin irritation and nerve discomfort that square-wave electrical stimulation can cause.

[0040] The closed-loop neural control current module also logs at the end of each cycle, recording fields such as emotion intensity probability, target value, difference, cosine gain, current amplitude, path impedance, and voltage, providing complete data for subsequent safety audits and individual model iterations. Long-term validation showed that over 14 consecutive days of 20 minutes per day in a multi-scenario interactive experiment, the closed-loop neural control current module was able to stably maintain the average comfort score of subjects between -0.05 and 0.1. The system's built-in temperature sensor detected that the temperature rise at the electrode-skin interface never exceeded 0.3 degrees Celsius, further demonstrating that the module not only ensures the effectiveness of neural control but also possesses reliable safety and engineering practical value.

[0041] Furthermore, the intrinsic steady-state time constant of the cortical emotion circuit is the characteristic time required for the user's brain emotion regulation network to spontaneously return to its resting emotional level after experiencing a clear emotional arousal, without external intervention. This is obtained through the following process: 60 seconds of continuous multi-channel EEG is collected from the subject at a resting baseline, and the baseline emotion intensity probability is calculated. Subsequently, high-arousal negative, high-arousal positive, and low-arousal neutral stimuli from the International Emotion Picture System or Emotion Audio Library are randomly presented, with each type of stimulus lasting 8 seconds. EEG and peripheral physiological signals are recorded for at least 300 seconds after the end of the stimulation. For each stimulus, the emotion intensity probability curve is calculated in real time, and a single exponential decay model is used to perform a least squares fit from the stimulus peak to the resting segment. The inverse of the decay coefficient is the instantaneous time constant of the experiment. Finally, the median of multiple time constants under the same type of stimulus is taken, and the results of the three types of stimulation are weighted averaged, where the weights are set to 0.5, 0.3, and 0.2 according to the stimulus arousal level, to obtain the intrinsic steady-state time constant of the user's cortical emotion circuit.

[0042] First, the subject was placed in a completely resting seated position with their eyes slightly closed but awake. The system continuously collected multi-channel EEG and peripheral physiological signals for 60 seconds. The emotion intensity probability algorithm calculated this data frame by frame and took the average value, which served as the reference point for the resting emotion intensity probability. Then, during the stimulation phase, the experimental software randomly selected three types of stimuli: high-arousal negative, high-arousal positive, and low-arousal neutral, from the International Emotional Image System or the Emotional Audio Library. Each stimulus lasted for eight seconds. During the stimulation, a countdown was displayed in the center of the screen to prevent attention drift. The system maintained a constant EEG sampling rate of 124 Hz. At the end of the stimulation, the system recorded the peak of the emotion intensity probability and immediately switched to a pure gray background to eliminate residual visual input. From this point on, EEG and peripheral physiological signals were collected for at least 300 seconds to observe the natural decay of the emotion signal. During this decay period, the system applied a single exponential decay model to the emotion intensity probability curve using a least-squares fit. The inverse of the decay coefficient returned by the model was considered the instantaneous time constant of the stimulus. To reduce the impact of random noise, each stimulation category is repeated ten times, resulting in ten instantaneous time constants. The median of these ten values ​​is then calculated to determine the representative time constant for that category. After completing the three stimulation categories, the system then takes a weighted average of the three representative time constants for high-arousal negative, high-arousal positive, and low-arousal neutral. The weights are set to 0.5, 0.3, and 0.2, respectively, based on the stimulus's arousal level. High-arousal negative, due to its greatest emotional amplitude, receives the highest weight. The resulting value is the intrinsic steady-state time constant of the user's cortical emotion circuit. If this value falls between 180 and 360 seconds, the self-stabilization speed of the user's cortical emotion circuit is within the typical range for healthy adults. If the value deviates significantly from this range, the system prompts the operator to review electrode contact quality or rule out physiological abnormalities for the subject that day. After measurement, the time constant is written to the user's unique configuration file and used as an integral factor in each subsequent control loop. This ensures smooth and responsive target value updates, preventing overshoot or delay in emotion regulation. Through this standardized process, the system can establish regulatory rhythms that strictly match the neural characteristics of different individuals, provide a solid timing foundation for the closed-loop neural regulatory current module, and significantly improve the user's comfort and emotional stability during long-term use.

[0043] Furthermore, the feedback influence time constant is the subjective comfort score input by the user during the closed-loop control process. The average delay required to produce an observable effect on the target emotional intensity of the system. The acquisition process includes: dividing the transcranial stimulation current into several step-by-step change segments, and requiring the user to continuously input the subjective comfort score; recording the whole process After the curve, for each current step-up or step-down event, the event trigger point is taken as the zero moment, the scoring waveform of the next 30 seconds is intercepted and the baseline is zeroed, and then the first-order exponential response model is used. Perform least squares fitting; take the obtained is the user instantaneous feedback time constant of the event; calculate the median of all rising and falling events respectively, and then calculate the weighted average according to the stimulus amplitude weight to obtain the final user feedback impact time constant. The first-order exponential response model is used to approximate the user rating curve The mathematical expression of .

[0044] First, the system divides the maximum available stimulation current into several percentage steps, such as 0, 25, 50, 75, and 100. Each level lasts for six seconds. The entire sequence first increases and then decreases, ensuring that both low-to-high excitation and high-to-low inhibition are covered. At the beginning of the experiment, the user holds the calibration slider and outputs a subjective comfort score in real time. The system records the score curve at a frequency of 256 Hz and simultaneously records the triggering time of each current step increase or decrease. To eliminate user bias in the slider's zero position, the system first zeroes the entire score curve using the individual resting mean as a baseline. Then, a 30-second segment of the score is captured backward from each trigger point to form a set of response samples. The system then uses its built-in exponential model fitting engine, treating each segment as a first-order exponential process. Starting from zero delay, the simulated curve is fitted point by point using least squares. During the fitting process, the engine automatically corrects for the upper limit of the amplitude, ultimately outputting a time constant that best describes the rate of increase or decrease in the score. For more than ten time constants generated in the same gradient direction, the system calculates their median to weaken the influence of outliers; then, using the current step amplitude as the weight, the system calculates the weighted average of all rising events and all falling events respectively, thereby obtaining two representative feedback time constants of the user under positive and negative stimulation.

[0045] Considering that people often experience more significant changes in comfort levels under high-amplitude stimulation, the system typically assigns greater weights to higher-amplitude steps. For example, the highest level in the five steps might have a weight of 0.45, with the remaining weights decreasing in descending order. This ensures that the final result truly reflects the user's sensitivity to intense stimulation. After calculating the positive and negative weights, the system averages them again, outputting a single feedback impact time constant, which is then stored in the user's unique parameter configuration file. If this time constant is between eight and twenty seconds, the user's subjective feedback speed in response to electrical stimulation is consistent with that of a normal adult. If it deviates significantly from this range, the system prompts you to review the slider sensitivity or recalibrate it. Subsequently, the baseline-target co-adaptive update module, using a fixed 50-millisecond time step for the control loop, logarithmically compresses the user's ratings and accumulates them into an integral pool of integer seconds. The integral decay rate is determined by the feedback impact time constant. This ensures that brief fluctuations in the user's mood do not immediately lead to significant target adjustments, while sustained comfort or discomfort can significantly affect the target value after several cycles. Throughout the entire process, all scoring curves, time constant calculation results, and weight distribution schemes are fully archived by the system's data recording unit, providing a traceable basis for long-term tracking of user experience changes, improving model weights, or adjusting slider sensitivity. Through this refined process, the brain-computer interface-driven emotion recognition and adaptive control system ensures that the role of subjective comfort in the electrical stimulation strategy is neither overly exaggerated nor ignored, achieving a truly human-centric, tightly coupled human-computer co-adaptive closed loop.

[0046] Furthermore, dividing the transcranial stimulation current into several step-by-step change segments specifically includes: dividing the transcranial stimulation current into multiple change segments, each change segment lasts for 6 seconds, and the amplitude of each change segment increases and then decreases in sequence of 0%, 25%, 50%, 75% and 100% of the maximum stimulation current.

[0047] The maximum available stimulation current is first written to the reference register of the digital-to-analog converter. Then, a time-amplitude lookup table is generated in the firmware based on preset percentages. Row 1 corresponds to 0%, row 2 to 25%, row 3 to 50%, row 4 to 75%, and row 5 to 100%. Rows 6 through 9 then decrease symmetrically until they reach 0% again. The system's main loop, clocked by a 1 kHz timer, automatically switches the lookup table row at the end of every 6000 timer cycles, ensuring that each amplitude change lasts exactly 6 seconds and maintains a constant amplitude. At the moment of the amplitude change event, a hardware trigger simultaneously writes a synchronization marker to the data acquisition link, ensuring that the time axis of the rating curve is precisely aligned with the point at which the current step occurs. During the amplitude ramp-up process, participants typically experience a gradual sensation of scalp tingling or mild flashes, with the slider score often shifting from near 0 to positive or negative values. During the amplitude ramp-down process, these sensations decrease as the current decreases, resulting in a mirror-symmetrical downward segment of the rating curve. Because each step maintains a constant amplitude for up to 6 seconds, the subjects have ample time to form a stable subjective evaluation of the current stimulation level, thereby reducing operational errors and improving the signal-to-noise ratio when fitting the first-order exponential response model. The system monitors the product of the output voltage and output current in real time in the background. If it finds that the actual current deviates from the target amplitude by more than 5% due to fluctuations in skin impedance, it immediately pauses the timing, readjusts the voltage, and then resumes counting to ensure that the effective duration of each change segment does not deviate from the nominal duration by more than 40 milliseconds. After the entire step sequence is completed, the system obtains score response samples for up to 10 rising events and 10 falling events, providing balanced forward and reverse data support for the subsequent calculation of the transient feedback time constant, and also laying the foundation for distinguishing the nonlinear characteristics of the impact of different amplitudes on subjective scores when allocating weights.

[0048] The subject's resting heart rate was 63 beats per minute, skin conductance was 6.2 microsiemens, and multi-channel EEG measured average alpha and beta wave powers of 11 and 13 microvolts squared in the prefrontal, cingulate, and insular regions, respectively. The impedance of each electrode was approximately 14 kilo-ohms. After the formal experiment began, the system presented a high-arousal negative image at the 120th second mark. Within the current sampling window, the four high-weight electrodes measured alpha wave powers of 14, 15, 13, and 14 microvolts squared, and beta wave powers of 17, 18, 16, and 17 microvolts squared, corresponding to impedances between 10 and 12 kilo-ohms. Normalized using brain area weights and impedance, the calculated brain-end energy term was 0.48. At the same time, the heart rate rose to 80 beats per minute, with a normalized heart rate deviation of |80-63| / (180-63)≈0.145; the skin conductance rose to 8.4 microsiemens, with a normalized skin conductance deviation of (8.4-6.2) / (15-6.2)≈0.25. Adding these three items yielded a single emotional energy scalar value of 0.875.

[0049] The subject's resting heart rate was 63 beats per minute, skin conductance was 6.2 microsiemens, and multi-channel EEG measured average alpha and beta wave powers of 11 and 13 microvolts squared in the prefrontal, cingulate, and insular regions, respectively. The impedance of each electrode was approximately 14 kilo-ohms. After the formal experiment began, the system presented a high-arousal negative image at the 120th second mark. Within the current sampling window, the four high-weight electrodes measured alpha wave powers of 14, 15, 13, and 14 microvolts squared, and beta wave powers of 17, 18, 16, and 17 microvolts squared, corresponding to impedances between 10 and 12 kilo-ohms. Normalized using brain area weights and impedance, the calculated brain-end energy term was 0.48. At the same time, the heart rate rose to 80 beats per minute, with a normalized heart rate deviation of |80-63| / (180-63)≈0.145; the skin conductance rose to 8.4 microsiemens, with a normalized skin conductance deviation of (8.4-6.2) / (15-6.2)≈0.25. Adding these three items yielded a single emotional energy scalar value of 0.875.

[0050] The closed-loop neural control current module currently has a target emotional intensity of 0.55, so the difference Δ = 0.819 - 0.55 = 0.269. The maximum available stimulation current is set to 2 mA. The module uses a cosine curve gain of 1-cos(πΔ) ≈ 1-cos(0.843) ≈ 0.334, resulting in a target stimulation current of 2 × 0.334 ≈ 0.668 mA. The real-time measured cranial-cortical-brain impedance is 14 kΩ, the reference impedance is 5 kΩ, and the proportional compensation factor is 5 / 14 ≈ 0.357. The output closed-loop transcranial electrical stimulation current I ≈ 0.668 × 0.357 ≈ 0.238 mA, corresponding to a drive voltage of approximately 0.238 mA × 14 kΩ ≈ 3.33 V.

[0051] The baseline target co-adaptation update module records the error Δ and integrates it according to the intrinsic steady-state time constant of the cortical emotion circuit of 300 seconds ≈ 0.05 / 300 = 1.67×10 -4 , the integral increment is about 1.67×10 -4 ×0.269≈4.48×10 -5 , the target value increases from 0.55 to 0.550045. At this time, the user's subjective comfort score of the slider is -0.4, and the system takes the logarithmic compression ln(1+0.4≈0.336). The behavioral feedback influence time constant is 15 seconds, and the feedback increment is ≈-0.05 / 15×0.336≈-1.12×10 -3 After the two items are merged, the target value is updated to 0.550045-0.00112≈0.549, which is written to the cache and called in the next 0.05 second loop.

[0052] The baseline target co-adaptation update module records the error Δ and integrates it according to the intrinsic steady-state time constant of the cortical emotion circuit of 300 seconds ≈ 0.05 / 300 = 1.67×10-4 , the integral increment is about 1.67×10 -4 ×0.269≈4.48×10 -5 , the target value increases from 0.55 to 0.550045. At this time, the user's subjective comfort score of the slider is -0.4, and the system takes the logarithmic compression ln(1+0.4≈0.336). The behavioral feedback influence time constant is 15 seconds, and the feedback increment is -0.05 / 15×0.336≈-1.12×10 -3 After the two items are merged, the target value is updated to 0.550045-0.00112≈0.549, which is written to the cache and called in the next 0.05 second loop.

[0053] Although specific embodiments of the present invention have been described above, those skilled in the art will appreciate that these specific embodiments are merely illustrative, and that those skilled in the art may omit, substitute, and modify the details of the methods and systems described above without departing from the principles and spirit of the present invention. For example, combining the above method steps to perform substantially the same functions in substantially the same manner to achieve substantially the same results falls within the scope of the present invention. Accordingly, the scope of the present invention is limited solely by the appended claims.

Claims

1. A brain-computer interface-driven emotion recognition and adaptive regulation system, characterized in that: The system includes: a multi-channel neural-peripheral coupling module, an emotion intensity probability mapping module and a closed-loop neural control current module; the multi-channel neural-peripheral coupling module is used to synchronously collect EEG, heart rate and skin electricity from multiple brain regions, and use area weighting and impedance normalization to fuse alpha wave power, beta wave power, local field potential and peripheral deviation into a single emotion energy scalar, thereby achieving overall quantification of central and peripheral emotional physiology; the emotion intensity probability mapping module is used to utilize emotion energy, combined with eye movement fatigue and brain electroencephalogram entropy, to generate a continuous emotion probability ranging from 0 to 1 through normalization and nonlinear mapping, provide a quantitative estimate of the current arousal-valence state, and complete real-time recognition; the closed-loop neural control current module is used to dynamically adjust the transcranial electrical stimulation intensity within the safe current upper limit according to the difference between the emotion probability and the desired target, and compensate for the equivalent resistance of the skull-skin-brain pathway in real time.

2. The brain-computer interface-driven emotion recognition and adaptive control system according to claim 1, characterized in that: The system further comprises: a baseline target co-adaptation update module for performing time constant integration and behavior feedback correction on the emotional target value according to the subjective comfort score input by the user.

3. The brain-computer interface-driven emotion recognition and adaptive control system according to claim 2, characterized in that: The specific execution process of the baseline target co-adaptation update module includes: recording the error between the emotion intensity probability of each control cycle and the current target value, and integrating it according to the inherent steady-state time constant of the cortical emotion circuit to make the target value tend to the real-time probability; receiving the subjective comfort score given by the user in the form of a handheld slider, using a logarithmic function to suppress the influence of extreme scores, and making positive or negative corrections to the target value according to the behavioral feedback influence time constant; repeating the above integration and correction with a fixed control cycle time step, so that the expected emotion intensity target evolves adaptively with the dual factors of neural homeostasis and subjective experience; and providing the updated target value to the closed-loop neural control current module for closed-loop stimulation decision-making in the next cycle.

4. The brain-computer interface-driven emotion recognition and adaptive control system according to claim 3, characterized in that: The specific execution process of the multi-channel neural-peripheral coupling module includes: placing a number of scalp electrodes in the brain areas corresponding to the prefrontal lobe, cingulate cortex and insula; obtaining alpha wave power, beta wave power, local field potential average value and electrode-skin impedance in real time in each electrode channel; weighted accumulation of the sum of the alpha wave power and beta wave power of each electrode according to the surface area of ​​the brain cortex, and using the electrode-skin impedance for power normalization to obtain a brain-end energy term; synchronously collecting electrocardiogram signals and skin electrode signals, and calculating the normalized difference between the real-time heart rate and the resting heart rate as the heart rate deviation term; and the normalized difference between the real-time skin conductance and the resting skin conductance to obtain a peripheral deviation term; adding the brain-end energy term, the heart rate deviation term and the peripheral deviation term to generate a single emotion energy scalar value.

5. The brain-computer interface driven emotion recognition and adaptive control system according to claim 4, characterized in that: The specific execution process of the emotion intensity probability mapping module includes: reading the generated emotion energy; synchronously obtaining the eye movement micro-tremor amplitude and comparing it with a preset fatigue threshold to obtain the eye movement fatigue degree; calculating the multi-scale EEG entropy and normalizing it with the resting entropy and peak entropy; after taking the weighted sum of the emotion energy scalar value, eye movement fatigue degree and normalized EEG entropy, using hyperbolic tangent mapping to output the emotion intensity probability between 0 and 1.

6. The brain-computer interface-driven emotion recognition and adaptive control system according to claim 5, characterized in that: The specific execution process of the closed-loop neural regulation current module includes: setting the maximum available stimulation current; setting the target value of the current expected emotional intensity, and comparing the emotional intensity probability with the target value in real time in the loop; when the emotional intensity probability is lower than the target value, setting the output stimulation current to zero; when the emotional intensity probability is higher than the target value, amplifying the stimulation current according to the cosine curve shape of the difference between the two, and the maximum does not exceed the maximum available stimulation current; detecting the equivalent resistance of the skull-skin-brain pathway, and proportionally compensating the stimulation current according to the standardized reference resistance, and outputting a closed-loop transcranial electrical stimulation current.

7. The brain-computer interface driven emotion recognition and adaptive control system according to claim 6, characterized in that: The intrinsic steady-state time constant of the cortical emotion circuit is the characteristic time required for the emotion regulation network in the user's brain to spontaneously return to its resting emotional level after experiencing a clear emotional arousal, without external intervention. This constant is obtained through the following process: 60 seconds of continuous multi-channel EEG is collected from the subject at a resting baseline, and the baseline emotion intensity probability is calculated. High-arousal negative, high-arousal positive, and low-arousal neutral stimuli from the International Emotion Picture System or the Emotion Audio Library are then randomly presented, each lasting 8 seconds. EEG and peripheral physiological signals are then recorded for at least 300 seconds after the stimulation ends. For each stimulus, an emotion intensity probability curve is calculated in real time, and a single exponential decay model is used to perform a least squares fit of the peak stimulus to rest interval. The inverse of the decay coefficient is the instantaneous time constant for that trial. Finally, the median of multiple time constants for the same stimulus type is taken, and the weighted average of the results for the three stimulus types is calculated, with weights set to 0.5, 0.3, and 0.2 based on the stimulus arousal level, to obtain the intrinsic steady-state time constant of the user's cortical emotion circuit.

8. The brain-computer interface driven emotion recognition and adaptive control system according to claim 7, characterized in that: The feedback influence time constant is the subjective comfort score input by the user during the closed-loop control process. The average delay required to produce an observable effect on the target emotional intensity of the system. The acquisition process includes: dividing the transcranial stimulation current into several step-by-step change segments, and requiring the user to continuously input the subjective comfort score; recording the whole process After the curve, for each current step-up or step-down event, the event trigger point is taken as the zero moment, the scoring waveform of the next 30 seconds is intercepted and the baseline is zeroed, and then the first-order exponential response model is used. Perform least squares fitting; take the obtained is the user instantaneous feedback time constant of the event; calculate the median of all rising and falling events respectively, and then calculate the weighted average according to the stimulus amplitude weight to obtain the final user feedback impact time constant. The first-order exponential response model is used to approximate the user rating curve The mathematical expression of .

9. The brain-computer interface driven emotion recognition and adaptive control system according to claim 7, characterized in that: Dividing the transcranial stimulation current into several step-by-step change segments specifically includes: dividing the transcranial stimulation current into multiple change segments, each change segment lasts for 6 seconds, and the amplitude of each change segment increases and then decreases in sequence of 0%, 25%, 50%, 75% and 100% of the maximum stimulation current.

Images