A dance rehabilitation training system based on electroencephalography and electrodermal interaction

Abstract

This invention discloses a dance rehabilitation training system integrating EEG and EEG sensors. The system includes an EEG headband and EEG sensors, each with built-in EEG and EEG acquisition modules. The dance rehabilitation training process includes: device wearing and initialization; personalized physiological baseline acquisition; simultaneous acquisition and preprocessing of dual-modal signals; multimodal feature fusion and emotion recognition, obtaining emotional state through dual-model fusion and smoothing; personalized safety threshold decision-making and intervention level determination; multi-channel dance intervention execution; emergency stress handling mechanism; closed-loop iteration and real-time control; and finally, generation of an evaluation report. This invention solves the problems of static intervention parameters, low feedback accuracy, and easy triggering of stress responses in traditional dance training by using dual-modal signal fusion, personalized safety threshold modeling, and closed-loop real-time control, achieving real-time, precise, and safe personalized rehabilitation intervention.

Description

Technical Field

[0001] This invention belongs to the field of biofeedback and dance rehabilitation training technology, specifically a dance rehabilitation training system that links brain electrical activity with skin electrical activity. Background Technology

[0002] Dance-based rehabilitation training, as a holistic intervention approach, plays a positive role in improving an individual's emotional regulation, sensorimotor coordination, and social participation abilities. It is particularly suitable for individuals with difficulties in social communication, repetitive and stereotyped behaviors, and sensorimotor dysfunction (such as those with autism spectrum disorder). Traditional dance-based intervention methods typically rely on the therapist's subjective observation and experience, facing the following technical bottlenecks during implementation: Firstly, intervention parameters often use fixed movement sequences and rhythms, making it difficult to dynamically adjust according to the trainee's real-time physiological and emotional state. Secondly, due to reliance on single physiological signals (such as using only EEG or TE), it is difficult to comprehensively and accurately reflect the trainee's emotional arousal level and physiological response characteristics. Furthermore, traditional methods lack personalized safety threshold settings, easily triggering discomfort or stress responses in trainees during the intervention process. Moreover, the data generated during the intervention is often isolated and unsystematic, failing to form a closed-loop recording, evaluation, and optimization mechanism.

[0003] While some biofeedback-based rehabilitation systems have emerged, existing technologies still have significant limitations. First, most methods utilize single-modal physiological signals (such as EEG or TE alone), which are susceptible to noise interference in dynamic motion scenarios, significantly reducing signal stability and recognition accuracy. Second, due to the complexity of data processing, these systems struggle to achieve millisecond-level real-time response and control, leading to intervention delays. Existing methods often employ uniform assessment and intervention thresholds, failing to fully consider the individual differences and heterogeneity of children with autism, and lacking truly personalized adaptation. Most of these systems fail to construct a complete "collection-decision-execution-optimization" closed loop; intervention parameters cannot be dynamically adjusted and continuously optimized based on the trainee's real-time state, limiting the improvement of intervention effectiveness and the system's own evolutionary capabilities. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a dance rehabilitation training system that integrates EEG and EMSE. By fusing and analyzing EEG and EMSE signals, modeling personalized safety thresholds, and implementing a millisecond-level closed-loop control mechanism, this system solves the technical problems of static intervention parameters, low feedback accuracy, and the tendency to trigger stress responses in trainees in traditional dance training, thereby achieving real-time, precise, and safe personalized rehabilitation intervention.

[0005] To achieve the above objectives, the present invention provides the following technical solutions.

[0006] A dance rehabilitation training system integrating EEG and ESC is disclosed. The system includes an EEG headband and an ESC sensor. The EEG headband and ESC sensor each have a built-in EEG acquisition module and an ESC acquisition module, respectively. The EEG headband is made of flexible silicone material, has no fewer than 8 electrodes, a sampling frequency of ≥250Hz, and supports Bluetooth 5.0 wireless transmission. The built-in EEG acquisition module is equipped with an anti-motion artifact algorithm. The ESC sensor has an IP65 waterproof rating and a sampling frequency of ≥10Hz. The dance rehabilitation training process of the system includes the following steps: Step S1: Device wearing and initialization; The EEG headband was worn on the trainee's head, covering the forehead, parietal lobe, and occipital lobe areas. The electrodermal sensor was worn on the trainee's non-dominant wrist. The EEG headband and electrodermal sensor settings were initialized, and EEG and electrodermal signals were collected separately.

[0007] Step S2: Personalized physiological baseline acquisition; The trainee was placed in a resting state, and EEG and ESC signals were collected for 3-5 minutes using an EEG headband and a skin conductance sensor, respectively, as a personalized physiological baseline. The EEG signal is filtered by a 50Hz power frequency notch filter to eliminate power interference, and then filtered by a 0.5-45Hz bandpass filter to extract the effective frequency band. The EEG signal is smoothed to eliminate high-frequency noise. The denoised EEG and EEG signals are used as a personalized physiological baseline, and personalized physiological baseline feature values ​​are calculated. These personalized physiological baseline feature values ​​include the power of each frequency band of the EEG signal. and mean skin conductance The calculation formula is as follows: (1); (2); In the above formula, The total duration of baseline acquisition is expressed in seconds. It is a function of α-wave power as a function of time; It is a function of skin conductance as a function of time; It is a time variable.

[0008] Step S3: Synchronous acquisition and preprocessing of dual-mode signals; Step S31: After the intervention begins, EEG signals are collected synchronously according to the control cycle. and skin conductance signals : (3); (4); In the above formula, for The amplitude of EEG signals collected at any time; For length is The amplitude of the EEG signal collected at the end of the time window; The length of the EEG signal sampling window (number of samples); The sampling frequency of the EEG signal, measured in Hz; for The amplitude of the skin conductance signal collected at any time; For a length of The amplitude of the skin conductance signal collected at the end of the time window; The length of the skin conductance signal sampling window (number of samples); The sampling frequency of the electrodermal signal, in Hz; Step S32: Preprocess the EEG signals; First, the power spectral density is calculated using the Welch method. : (5); In the above formula, Window length (number of sampling points); As the normalization factor, ; For the Hanning window function, For window function index; The value of the EEG signal at the m-th sampling point; is the base of the natural logarithm; The imaginary unit; For frequency; Then calculate the power of each frequency band. : (6); In the above formula, This is the upper limit frequency of the frequency band; This is the lower limit frequency of the frequency band; This is the index corresponding to the upper limit of the frequency. This is the index corresponding to the lower frequency limit; Let K be the power spectral density at the Kth frequency point; Frequency resolution; extract , , , , The absolute and relative power of the bands are measured, and artifact detection is performed simultaneously. Based on the detected amplitude, gradient, and energy threshold, blink and electromyographic interference signals in the EEG signal are identified. The , , , , The wavebands are the international standard wavebands for electroencephalography (EEG), with waveband intervals of 0.5–4Hz, 4–8Hz, 8–13Hz, 13–30Hz, and 30–45Hz. It represents deep sleep, unconscious state, and the dominant wave in infancy; It represents light sleep, meditation, creative thinking, and mood swings; It represents a relaxed and alert state, resting with eyes closed, and being calm and focused. It represents a state of alertness and activity, thinking, anxiety, and focused attention; It represents higher cognitive functions, sensory integration, and learning and memory; Step S33: Process the skin conductance signal; First, a smoothing filter is applied. Then, the Tonic / Phasic decomposition method is used to separate the low-frequency trend component Tonic and the high-frequency fluctuation component Phasic. The morphological characteristics of skin conductance response (SCR) events are detected in the Phasic component. (7); (8); In the above formula, Represents a low-pass filter function; This represents the original skin conductance signal; The cutoff frequency; The detection conditions for the skin conductance response (SCR) event are set as follows: rise time 0.5-5.0s, amplitude threshold > 0.05μs, and minimum interval > 2s.

[0009] Step S4: Multimodal feature fusion and emotion recognition; A 20-dimensional fusion feature vector was extracted from the preprocessed EEG and SC signals. : (9); In the above formula, Represents 6-dimensional brainwaves Features, including power, power, power, ratio, Ratio and signal entropy; The six-dimensional skin conductance (SC) characteristics include SC mean, SC standard deviation, SC slope, response rate, tonic component, and pulsed component. Representing 2D interactive features, including Power × SC mean sum Power × SC slope; It represents 3-dimensional statistical characteristics, including EEG mean, EEG standard deviation, and SC entropy; It represents 3D historical features, including the trend of feature changes based on the most recent 5 time windows; The extracted 20-dimensional fusion features are standardized based on baseline data: (10); In the above formula, The value of the i-th feature after standardization; This represents the original value of the i-th feature; Let be the baseline mean of the i-th feature; Let be the baseline standard deviation of the i-th feature; A dual-model fusion architecture of SVM and random forest is used for emotion recognition: (11); (12); In the above formula, This represents the fused emotion category probability vector; This is the probability vector of the sentiment category output by the SVM model; This is the probability vector of the sentiment category output by the random forest model; Rate your mood; Indicates the first The basic values ​​of various emotional states ; Specifically, the SVM model performs stably in high-dimensional, small-sample scenarios and is sensitive to EEG spectral features, so it is given a higher weight (0.6); while the random forest model can effectively handle nonlinear interaction features and provides feature importance analysis, so it is given a weight (0.4) as a supplementary model.

[0010] Furthermore, to avoid abrupt changes in sentiment values, an inertia factor is introduced for smoothing: (13); In the above formula, The emotion score after smoothing at time t; The current emotion score at time t; The emotion score after smoothing from the previous moment; Referring to the commonly used settings of clinical biofeedback systems, a weighting ratio of 0.7:0.3 can both suppress noise-induced mutations and reflect real emotional changes in autism data in a timely manner. Emotional state is determined based on the smoothed sentiment score: (14); In the above formula, Indicates emotional state; For deep relaxation; For relaxation; Neutral; To be vigilant; For stress; This is due to severe stress.

[0011] Step S5: Personalized safety threshold decision and intervention level determination; Based on established personalized physiological baseline characteristics and historical data, a dynamic safety threshold model is used to determine a state of stress (emotional score ≥ 7) when the following conditions are met: (15); Based on the emotional state and the current intervention level, an adaptive strategy is used to determine the updated intervention level: (16); In the above formula, Indicates the level of intervention after the update; Indicates the current intervention level; This represents the smoothed sentiment score at time t; A state transition probability matrix is ​​used for progressive adjustment of the intervention level. This matrix defines the transition probability from the current intervention level. and emotional state Move to the next intervention level The probability distribution of is expressed as: (17); In the above formula, This is a state transition probability matrix, where rows correspond to the current intervention level (0-3) and columns correspond to the next intervention level (0-3). Based on the current intervention level CurrentLevel, the corresponding row vector is selected as the probability distribution, and NextLevel is determined by random sampling or the maximum probability method. For example, if CurrentLevel=1, the probabilities of the next level being 0, 1, 2, and 3 are 0.2, 0.5, 0.2, and 0.1, respectively. If the maximum probability method is used, the level with the highest probability, 1, is selected as NextLevel.

[0012] To prevent trainees from becoming uncomfortable due to sudden changes in intervention strategies, a safety constraint is set: maximum stress duration. seconds, minimum relaxation time Seconds, frequency of intervention level changes times per minute.

[0013] Step S6: Execute multi-channel dance intervention; According to the determined intervention level, the corresponding dance intervention is carried out. 3-5 movement sequences suitable for the trainee's ability level and sensory preferences are selected from the preset movement library, and the music rhythm is adjusted in sync. The beats per minute (BPM) is linearly related to the intervention level. (18); In the above formula, Intervention level; A random integer uniformly distributed in the range [-5, 5] is used to add natural variations to the musical rhythm; The correspondence between intervention levels and BPM is as follows: Intervention level 0 corresponds to 50-60 BPM, with breathing guidance and static movements performed; Intervention level 1 corresponds to 65-75 BPM, with gentle intervention and simple stretching movements performed; Intervention level 2 corresponds to 80-90 BPM, with moderate intervention and rhythmic movements performed; Intervention level 3 corresponds to 95-105 BPM, with strong intervention and coordinated movements performed. The preset motion library includes movements such as stretching, swaying, clapping, stepping, and rotating. Specifically, during the dance intervention, multi-channel real-time feedback is provided: visual feedback is achieved through colored LED lights, with green indicating a relaxed state, blue indicating an alert state, and red indicating a stressed state; tactile feedback is achieved through a vibrating wristband, with vibration intensity... Auditory feedback plays prompts during key state transitions.

[0014] Step S7, Emergency Stress Management Mechanism; When a trainee's emotional state of ≥7 and duration >10s is detected as a stress state, an emergency treatment mechanism is immediately triggered: the intervention level is forcibly adjusted to level 0, a breathing guidance mode is implemented, the music is adjusted to 60±5BPM, the visual feedback is switched to the purple breathing mode, the vibration intensity is increased, and stress event information is recorded and saved for subsequent analysis and method optimization.

[0015] Step S8: Closed-loop iteration and real-time control; Repeat steps S3-S7 for closed-loop iteration in each control cycle. Within the system, all key parameters are recorded in real time, including emotion value, intervention level, executed actions, beats per minute (BPM) value, EEG characteristics, and SC characteristics of skin conductance, forming a complete data stream for real-time visualization and subsequent analysis. (19); In the above formula, For signal acquisition time; For signal processing and feature extraction time; Timeframes for emotion recognition and decision-making; To intervene in the execution time.

[0016] Step S9: Intervention termination and comprehensive assessment; After the intervention, a comprehensive evaluation report is generated, which includes an emotion change curve, physiological indicator trends, intervention effect analysis, and personalized recommendations. The emotion change curve is based on the smoothed emotion score at time t. Corresponding emotional state get; The physiological indicators include Wave power, skin conductance mean Ratio and intervention level; The intervention effectiveness analysis was performed by calculating participation scores. and improve scores get: (20); (twenty one); In the above formula, This represents the participation score; The proportion of individuals whose emotional value E falls between 3 and 6; The standard deviation of the sentiment score E; The number of times the intervention level changes; This represents the total number of records (number of data points). This represents the average sentiment score during the first half of the conversation. This represents the average sentiment score for the second half of the conversation. Based on the obtained emotion change curves, physiological indicator trends, and intervention effect analysis results, corresponding personalized suggestions are given. The comprehensive assessment report and all data are integrated to create personalized profiles and a safety threshold database, which are then uploaded to the cloud for storage, providing a basis for optimization in subsequent interventions.

[0017] Compared with the prior art, the present invention has the following beneficial effects: 1. Compared with traditional single-modal methods, this invention adopts EEG and TESC dual-modal signal fusion analysis technology. By extracting 20-dimensional fusion feature vectors and using a weighted fusion architecture of SVM model and random forest dual model, the accuracy of emotion recognition is improved from 78% of the single-modal method to 92%, and the false alarm rate is reduced from 15% to below 5%. At the same time, a millisecond-level closed-loop control mechanism is designed. By optimizing the control cycle allocation and parallel processing of hardware interface, the 1-2 second delay of the traditional biofeedback system is greatly shortened to less than 500ms, achieving a 200ms-level real-time response.

[0018] 2. This invention establishes a personalized dynamic safety threshold model, which, based on individual physiological baseline data and multi-indicator joint judgment, and combined with a historical data adaptive adjustment mechanism, reduces the stress generation rate of trainees from 25% in the traditional fixed threshold method to below 10%. At the same time, a progressive adjustment mechanism based on the state transition probability matrix is ​​designed, which, through the participation score calculation formula and continuous learning optimization, increases the average participation score from 45% in the traditional method to over 65%.

[0019] 3. This invention forms a complete closed-loop system of "collection-decision-execution-evaluation", which completes a complete control cycle every 200ms. All data is automatically recorded and a comprehensive evaluation report is generated, supporting data-based evidence-based medicine research and continuous scientific optimization of rehabilitation programs. Attached Figure Description

[0020] To provide a more intuitive understanding of the technical implementation of this invention, the accompanying drawings involved in the embodiments of this invention are briefly described below. These drawings are used to assist in illustrating the implementation methods and are not intended to limit the invention. Those skilled in the art can make derivative designs based on the drawings without creative effort.

[0021] Figure 1 This is a flowchart of the dance rehabilitation training process of a brainwave-cortical electroresonance system according to the present invention. Figure 2 This is a schematic diagram of real-time monitoring of the autism children's dance intervention and regulation system in an embodiment of the present invention. Detailed Implementation

[0022] To facilitate understanding and implementation of the present invention by those skilled in the art, the various steps of the method proposed in this invention are described in detail below. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various modifications or alterations to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

[0023] Example 1 like Figure 1 As shown, this embodiment discloses a dance rehabilitation training system that integrates EEG and EEG sensors. The system includes an EEG headband and an EEG sensor. The EEG headband and the EEG sensor each have a built-in EEG acquisition module and an EEG acquisition module, respectively. The EEG headband is made of flexible silicone material and has no fewer than 8 electrodes. It has a sampling frequency of ≥250Hz and supports Bluetooth 5.0 wireless transmission. The built-in EEG acquisition module is equipped with an anti-motion artifact algorithm. The EEG sensor has an IP65 waterproof rating and a sampling frequency of ≥10Hz. The dance rehabilitation training process of the system includes the following steps: Step S1: Device wearing and initialization; The EEG headband was worn on the trainee's head, covering the forehead, parietal lobe, and occipital lobe areas. The electrodermal sensor was worn on the trainee's non-dominant wrist. The EEG headband and electrodermal sensor settings were initialized, and EEG and electrodermal signals were collected separately.

[0024] Step S2: Personalized physiological baseline acquisition; The trainee was placed in a resting state, and EEG and ESC signals were collected for 3-5 minutes using an EEG headband and a skin conductance sensor, respectively, as a personalized physiological baseline. The EEG signal is filtered by a 50Hz power frequency notch filter to eliminate power interference, and then filtered by a 0.5-45Hz bandpass filter to extract the effective frequency band. The EEG signal is smoothed to eliminate high-frequency noise. The denoised EEG and EEG signals are used as a personalized physiological baseline, and personalized physiological baseline feature values ​​are calculated. These personalized physiological baseline feature values ​​include the power of each frequency band of the EEG signal. and mean skin conductance The calculation formula is as follows: (1); (2); In the above formula, The total duration of baseline acquisition is expressed in seconds. It is a function of α-wave power as a function of time; It is a function of skin conductance as a function of time; It is a time variable.

[0025] Step S3: Synchronous acquisition and preprocessing of dual-mode signals; Step S31: After the intervention begins, EEG signals are collected synchronously according to the control cycle. and skin conductance signals : (3); (4); In the above formula, for The amplitude of EEG signals collected at any time; For length is The amplitude of the EEG signal collected at the end of the time window; The length of the EEG signal sampling window (number of samples); The sampling frequency of the EEG signal, measured in Hz; for The amplitude of the skin conductance signal collected at any time; For a length of The amplitude of the skin conductance signal collected at the end of the time window; The length of the skin conductance signal sampling window (number of samples); The sampling frequency of the electrodermal signal, in Hz; Step S32: Preprocess the EEG signals; First, the power spectral density is calculated using the Welch method. : (5); In the above formula, Window length (number of sampling points); As the normalization factor, ; For the Hanning window function, For window function index; The value of the EEG signal at the m-th sampling point; is the base of the natural logarithm; The imaginary unit; For frequency; Then calculate the power of each frequency band. : (6); In the above formula, This is the upper limit frequency of the frequency band; This is the lower limit frequency of the frequency band; This is the index corresponding to the upper limit of the frequency. This is the index corresponding to the lower frequency limit; Let K be the power spectral density at the Kth frequency point; Frequency resolution; extract , , , , The absolute and relative power of the bands are measured, and artifact detection is performed simultaneously. Based on the detected amplitude, gradient, and energy threshold, blink and electromyographic interference signals in the EEG signal are identified. The , , , , The wavebands are the international standard wavebands for electroencephalography (EEG), with waveband intervals of 0.5–4Hz, 4–8Hz, 8–13Hz, 13–30Hz, and 30–45Hz. It represents deep sleep, unconscious state, and the dominant wave in infancy; It represents light sleep, meditation, creative thinking, and mood swings; It represents a relaxed and alert state, resting with eyes closed, and being calm and focused. It represents a state of alertness and activity, thinking, anxiety, and focused attention; It represents higher cognitive functions, sensory integration, and learning and memory; Step S33: Process the skin conductance signal; First, smoothing filtering is performed to eliminate high-frequency noise. Then, the Tonic / Phasic decomposition method is used to separate the low-frequency trend component Tonic and the high-frequency fluctuation component Phasic. The morphological characteristics of skin conductance response (SCR) events are detected in the Phasic component. (7); (8); In the above formula, Represents a low-pass filter function; This represents the original skin conductance signal; The cutoff frequency; The detection conditions for the skin conductance response (SCR) event are set as follows: rise time 0.5-5.0s, amplitude threshold > 0.05μs, and minimum interval > 2s.

[0026] Step S4: Multimodal feature fusion and emotion recognition; A 20-dimensional fusion feature vector was extracted from the preprocessed EEG and SC signals. : (9); In the above formula, Represents 6-dimensional brainwaves Features, including power, power, power, ratio, Ratio and signal entropy; The six-dimensional skin conductance (SC) characteristics include SC mean, SC standard deviation, SC slope, response rate, tonic component, and pulsed component. Representing 2D interactive features, including Power × SC mean sum Power × SC slope; It represents 3-dimensional statistical characteristics, including EEG mean, EEG standard deviation, and SC entropy; It represents 3D historical features, including the trend of feature changes based on the most recent 5 time windows; The extracted 20-dimensional fusion features are standardized based on baseline data: (10); In the above formula, The value of the i-th feature after standardization; This represents the original value of the i-th feature; Let be the baseline mean of the i-th feature; Let be the baseline standard deviation of the i-th feature; A dual-model fusion architecture of SVM and random forest is used for emotion recognition: (11); (12); In the above formula, This represents the fused emotion category probability vector; This is the probability vector of the sentiment category output by the SVM model; This is the probability vector of the sentiment category output by the random forest model; Rate your mood; Indicates the first The basic values ​​of various emotional states ; Specifically, the SVM model performs stably in high-dimensional, small-sample scenarios and is sensitive to EEG spectral features, so it is given a higher weight (0.6); while the random forest model can effectively handle nonlinear interaction features and provides feature importance analysis, so it is given a weight (0.4) as a supplementary model.

[0027] Furthermore, to avoid abrupt changes in sentiment values, an inertia factor is introduced for smoothing: (13); In the above formula, The emotion score after smoothing at time t; The current emotion score at time t; The emotion score after smoothing from the previous moment; Referring to the commonly used settings of clinical biofeedback systems, a weighting ratio of 0.7:0.3 can both suppress noise-induced mutations and reflect real emotional changes in autism data in a timely manner. Emotional state is determined based on the smoothed sentiment score: (14); In the above formula, Indicates emotional state; For deep relaxation; For relaxation; Neutral; To be vigilant; For stress; This is due to severe stress.

[0028] Step S5: Personalized safety threshold decision and intervention level determination; Based on established personalized physiological baseline characteristics and historical data, a dynamic safety threshold model is used to determine a state of stress (emotional score ≥ 7) when the following conditions are met: (15); In the above formula, Pressure threshold; Based on the emotional state and the current intervention level, an adaptive strategy is used to determine the updated intervention level: (16); In the above formula, Indicates the level of intervention after the update; Indicates the current intervention level; This represents the smoothed sentiment score at time t; A state transition probability matrix is ​​used for progressive adjustment of the intervention level. This matrix defines the transition probability from the current intervention level. and emotional state Move to the next intervention level The probability distribution of is expressed as: (17); In the above formula, This is a state transition probability matrix, where rows correspond to the current intervention level (0-3) and columns correspond to the next intervention level (0-3). Based on the current intervention level CurrentLevel, the corresponding row vector is selected as the probability distribution, and NextLevel is determined by random sampling or the maximum probability method. For example, if CurrentLevel=1, the probabilities of the next level being 0, 1, 2, and 3 are 0.2, 0.5, 0.2, and 0.1, respectively. If the maximum probability method is used, the level with the highest probability, 1, is selected as NextLevel.

[0029] To prevent trainees from becoming uncomfortable due to sudden changes in intervention strategies, a safety constraint is set: maximum stress duration. seconds, minimum relaxation time Seconds, frequency of intervention level changes times per minute.

[0030] Step S6: Execute multi-channel dance intervention; According to the determined intervention level, the corresponding dance intervention is carried out. 3-5 movement sequences suitable for the trainee's ability level and sensory preferences are selected from the preset movement library, and the music rhythm is adjusted in sync. The beats per minute (BPM) is linearly related to the intervention level. (18); In the above formula, Intervention level; A random integer uniformly distributed in the range [-5, 5] is used to add natural variations to the musical rhythm; The correspondence between intervention levels and BPM is as follows: Intervention level 0 corresponds to 50-60 BPM, with breathing guidance and static movements performed; Intervention level 1 corresponds to 65-75 BPM, with gentle intervention and simple stretching movements performed; Intervention level 2 corresponds to 80-90 BPM, with moderate intervention and rhythmic movements performed; Intervention level 3 corresponds to 95-105 BPM, with strong intervention and coordinated movements performed. The preset motion library includes movements such as stretching, swaying, clapping, stepping, and rotating. Specifically, during the dance intervention, multi-channel real-time feedback is provided: visual feedback is achieved through colored LED lights, with green indicating a relaxed state, blue indicating an alert state, and red indicating a stressed state; tactile feedback is achieved through a vibrating wristband, with vibration intensity... Auditory feedback plays prompts during key state transitions.

[0031] Step S7, Emergency Stress Management Mechanism; When a trainee's emotional state of ≥7 and duration >10s is detected as a stress state, an emergency treatment mechanism is immediately triggered: the intervention level is forcibly adjusted to level 0, a breathing guidance mode is implemented, the music is adjusted to 60±5BPM, the visual feedback is switched to the purple breathing mode, the vibration intensity is increased, and stress event information is recorded and saved for subsequent analysis and method optimization.

[0032] Step S8: Closed-loop iteration and real-time control; Repeat steps S3-S7 for closed-loop iteration in each control cycle. Within the system, all key parameters are recorded in real time, including emotion value, intervention level, executed actions, beats per minute (BPM) value, EEG characteristics, and SC characteristics of skin conductance, forming a complete data stream for real-time visualization and subsequent analysis. (19); In the above formula, For signal acquisition time; For signal processing and feature extraction time; Timeframes for emotion recognition and decision-making; To intervene in the execution time.

[0033] Step S9: Intervention termination and comprehensive assessment; After the intervention, a comprehensive evaluation report is generated, which includes an emotion change curve, physiological indicator trends, intervention effect analysis, and personalized recommendations. The emotion change curve is based on the smoothed emotion score at time t. Corresponding emotional state get; The intervention effectiveness analysis was performed by calculating participation scores. and improve scores get: (20); (twenty one); In the above formula, This represents the participation score; The proportion of individuals whose emotional value E falls between 3 and 6; The standard deviation of the sentiment score E; The number of times the intervention level changes; This represents the total number of records (number of data points). This represents the average sentiment score during the first half of the conversation. This represents the average sentiment score for the second half of the conversation. Based on the obtained emotion change curves, physiological indicator trends, and intervention effect analysis results, corresponding personalized suggestions are given. The comprehensive assessment report and all data are integrated to create personalized profiles and a safety threshold database, which are then uploaded to the cloud for storage, providing a basis for optimization in subsequent interventions.

[0034] The technical effects of the present invention will be further illustrated by the following example.

[0035] Rehabilitation training target: A 6-year-old boy with autism, moderate sensory-motor integrated ability, visually sensitive, and child ID ASD001.

[0036] Implementation steps: Step S1: Device wearing and initialization; Wear the child-specific 8-channel EEG headband and wristband-type electrodermal sensor correctly, ensuring good contact between each electrode and the skin; the EEG headband covers the forehead, parietal lobe, and occipital lobe areas, and is made of flexible silicone material to adapt to the child's head shape; the electrodermal wristband is worn on the non-dominant wrist and adjusted to a comfortable position; load the child's personalized configuration profile (ID ASD001), including their physiological signal safety thresholds, sensory preference parameters, and basic dance movement library, and complete the hardware connection and initial setup.

[0037] Step S2: Personalized baseline acquisition; Children were guided to sit quietly and relax for 3 minutes. Resting-state EEG and ductal nerve conductance (TEF) signals were collected to establish a personalized physiological baseline. During signal acquisition, an EEG headband monitored signal quality in real time to ensure signal stability. After the personalized physiological baseline was acquired, personalized physiological characteristic values ​​were calculated: baseline α power was 3.2 μV², mean SC was 5.1 μS, and α / β ratio was 1.05. Based on these personalized physiological characteristic values, a personalized safety threshold was set. .

[0038] In the above formula, Pressure threshold; Step S3: Real-time monitoring and preprocessing of dual-mode signals; After the intervention began, EEG and TE signals were collected synchronously with a control cycle of 200ms. The following real-time data were collected in the 5-minute time window after the intervention began: EEG signals showed that the relative power of β waves increased to 42%, the mean value of the electrodermal conductance signal increased to 13.5 μS, and significant skin conductance response (SCR) events were detected. Preprocessing of the EEG signals involved: first, a 50 Hz power frequency notch filter was applied to eliminate power interference; then, a 0.5-45 Hz bandpass filter was used to extract the effective frequency band; the Welch method was used to calculate the power spectral density (PSD), and the absolute and relative power of each frequency band were calculated. Smoothing filtering and Tonic / Phasic decomposition were applied to the electrodermal conductance signal to detect the morphological characteristics of SCR events.

[0039] Step S4: Multimodal feature fusion and emotion recognition; A 20-dimensional fusion feature vector was extracted from the preprocessed EEG and SC signals. : ; Specifically, it includes: 6-dimensional EEG features: power( =2.8μV²), power( =8.2μV²), power( =4.1 μV²), ratio( =0.34), ratio( =0.5), signal entropy ( =0.82); 6-dimensional SC features: mean ( =13.5μS), standard deviation ( =2.1μS), slope ( =0.35), response rate ( =0.2), Element( =12.8μS), Element( =0.7μS); 2D interaction features: Power × SC mean ( =37.8), Power × SC slope ( =2.87); 3-dimensional statistical characteristics: EEG mean ( =0.3), EEG standard deviation ( =0.8), SC entropy ( =1.2); 3D historical trend characteristics: =0.1、 =-0.05、 =0.02; All of the above features are standardized based on the baseline data collected in step S2; A dual-model fusion architecture of SVM and random forest was used for emotion recognition, and the recognition results are as follows: ; ; In the above formula, This is the probability vector of the sentiment category output by the SVM model; This is the probability vector of the sentiment category output by the random forest model; Perform weighted fusion calculation: ; In the above formula, This represents the fused emotion category probability vector; Calculate the emotion score : ; Based on an emotion score of 7.24, the emotional state is determined to be STRESS.

[0040] Step S5: Personalized safety threshold decision and intervention level determination; Check the individualized safety threshold conditions: SC mean 13.5μS>12μS, β wave relative power 42%>35%, α / β ratio 0.34<0.8. If all conditions are met simultaneously, the stress state detection is confirmed. Based on the emotional state STRESS and the current intervention level 2, an adaptive strategy is used to adjust the intervention level.

[0041] According to the intervention level mapping rule (Formula 16): when hour, ; The current level is 2, and the new level is calculated to be 1; however, considering the duration and intensity of the stress state, the emergency response mechanism is triggered, and the intervention level is further adjusted to level 0. The decision-making process is recorded for subsequent analysis and optimization.

[0042] Step S6: Execute multi-channel dance intervention; Based on the adjusted intervention level of 0, implement the emergency intervention plan: Action Adjustment: Immediately switch to breathing guidance mode, select a static breathing action from the action library, and display a deep breathing guidance animation; Rhythm adjustment: The music BPM has been adjusted from 85 to 60, adopting a soothing meditation music style; Visual feedback: The LED light switches to a red breathing mode, and the brightness changes slowly with the breathing rhythm; Haptic feedback: The vibration wristband intensity is set to 0.8, using a slow pulse mode to provide tactile guidance; Auditory prompt: Play a gentle voice prompt: "Please take a deep breath with me, inhale...exhale...".

[0043] Step S7, Emergency Stress Management Mechanism; Continuous monitoring of children's physiological signals and emotional state changes. Two minutes after emergency intervention, the mean value of skin conductance signal decreased from 13.5 μS to 8.2 μS, the relative power of β wave decreased from 42% to 28%, and the emotion score dropped from 7.24 to 3.1; the emotional state was detected to have recovered from STRESS to RELAXED.

[0044] Based on the state transition probability matrix and the gradual adjustment mechanism: according to the current intervention level CurrentLevel = 0 and the current emotional state EmotionState = RELAXED, the probability vector of the first row (index 0) of the corresponding behavior in the state transition probability matrix (Formula 17) is selected: ; The probabilities of the next intervention level being 0, 1, 2, and 3 are 0.6, 0.3, 0.1, and 0.0, respectively. In this embodiment, the maximum probability method is used, and level 0, corresponding to the maximum probability, is selected as NextLevel. To further achieve gradual recovery, considering that the current emotional state is RELAXED and the intervention has lasted for 2 minutes, the system appropriately increases the intervention intensity in the next decision cycle, adjusting NextLevel to level 1, thereby achieving a smooth transition from level 0 to level 1 and avoiding secondary stress caused by abrupt changes in intervention intensity.

[0045] Step S8: Closed-loop iteration and real-time control; Throughout the intervention, a strict 200ms control period was maintained. ; Verification via the real-time monitoring interface showed that signal acquisition took approximately 30ms, signal processing and feature extraction took approximately 80ms, emotion recognition and decision-making took approximately 60ms, and intervention execution took approximately 20ms, with a total cycle time of 190ms, meeting the real-time requirements. Intervention parameters were updated every 200ms, forming a complete closed-loop control system.

[0046] Step S9: Intervention termination and comprehensive assessment; This intervention session lasted 10 minutes and generated 3000 records (200ms / record). Figure 2 As shown, a comprehensive evaluation report will be generated: 1. Basic data statistics: Session duration: 10 minutes; Total number of data records: 3000; Average sentiment score: 3.34; Stress incidence rate: 0%; Number of intervention level adjustments: 3 times.

[0047] 2. Participation Analysis: ; In the above formula, This represents the participation score; Improvement assessment: Initial emotional score 3.8, final emotional score 3.1. Since the emotional score was consistently within the ideal range (3-6), the improvement score was 0, indicating that the child's emotional stability was effectively maintained.

[0048] .

[0049] In the above formula, To improve scores; 3. Changes in physiological indicators: like Figure 2 As shown in the curve of the α / β ratio change, the α / β ratio changes from an initial 1.05 to a final 0.82, indicating a slight decrease in the degree of relaxation; like Figure 2 As shown in the mid-skin conductivity (SC) curve, the mean SC value changed from a baseline of 5.1 μS to a final value of 5.8 μS, maintaining a stable range. like Figure 2 As shown in the curve of relative β-wave power change, the relative β-wave power changes from the baseline value to a final 28%, which is at an appropriate level.

[0050] 4. Personalized suggestions: Recommendation 1: The child's emotional stability is good, and the method response is timely and effective under stress. It is recommended to appropriately increase challenging actions, such as coordination exercises, in subsequent interventions. Recommendation 2: The α / β ratio is low (0.82). It is recommended to add relaxation training to the intervention to strengthen α wave activity. Recommendation 3: The red breathing mode is effective in stress intervention and responds positively to visual feedback. It is recommended to strengthen the color guidance settings in the personalized configuration. Recommendation 4: Based on the data from this intervention, optimize the personalized parameters of children's ASD001 to provide a more accurate benchmark for subsequent interventions.

[0051] All data is automatically uploaded to cloud storage, updating the personalized profile of children with ASD001 and generating a complete PDF assessment report, including mood change curves, physiological indicator trend charts, intervention effect analysis, and detailed recommendations to support therapists in making evidence-based decisions and optimizing rehabilitation programs.

[0052] This embodiment verifies the effectiveness and practicality of the present invention. The system successfully detected the stress state and triggered emergency intervention, and completed the complete closed loop from signal acquisition to intervention execution within 200ms. Multimodal fusion analysis accurately identifies changes in emotional state and can provide timely and effective intervention guidance. Experimental data show that the present invention can significantly improve the safety, accuracy and effectiveness of dance intervention for children with autism, and has good clinical application prospects.

[0053] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A dance rehabilitation training system based on electroencephalography (EEG) and electrodermal oscillation (EDS), characterized in that, The system includes an EEG headband and a skin conductance sensor, wherein the EEG headband and the skin conductance sensor respectively have built-in EEG acquisition modules and skin conductance acquisition modules; The dance rehabilitation training process of the system includes the following steps: Step S1: Device wearing and initialization; The EEG headband was worn on the trainee's head, covering the forehead, parietal lobe, and occipital lobe areas. The electrodermal sensor was worn on the trainee's non-dominant wrist. The EEG headband and electrodermal sensor settings were initialized, and EEG and electrodermal signals were collected respectively. Step S2, Personalized physiological baseline acquisition: The trainee is placed in a resting state, and EEG signals and EEG signals are collected for 3-5 minutes using an EEG headband and a skin conductance sensor, respectively, as a personalized physiological baseline; Step S3, Simultaneous Acquisition and Preprocessing of Dual-Modal Signals: After the intervention begins, EEG and TE signals are acquired synchronously according to the control cycle, and preprocessed. Step S4, Multimodal Feature Fusion and Emotion Recognition: Extract a 20-dimensional fusion feature vector from the preprocessed EEG and EEG signals. Standardize the extracted 20-dimensional fusion features based on baseline data. Use a dual-model fusion architecture of SVM and random forest for emotion recognition. Inertia factor is introduced for smoothing to obtain trainee emotion scores, thereby determining the trainee's emotional state. Step S5, Personalized safety threshold decision and intervention level determination: Based on the established personalized physiological baseline features and historical data, a dynamic safety threshold model is used to determine the current intervention level according to the trainee's emotional state, and an adaptive strategy is used to update the intervention level, with gradual adjustments based on the state transition probability matrix; Step S6, Multi-channel dance intervention execution: Execute the corresponding dance intervention according to the updated intervention level, select 3-5 movement sequences from the preset movement library that are suitable for the trainee's ability level and sensory preferences, and adjust the music rhythm in sync; Step S7, Emergency Stress Handling Mechanism: When a trainee is detected to be in a state of stress, the emergency handling mechanism is immediately triggered, and stress event information is recorded and saved at the same time. Step S8, Closed-loop Iteration and Real-time Control: Repeat steps S3-S7 in each control cycle. Internally, it records all key parameters in real time, including emotion value, intervention level, executed actions, beats per minute (BPM) value, and EEG and SC characteristics of the brain. Step S9: Intervention ends, generate comprehensive assessment report.

2. The brainwave-cortical electroreceptor linkage dance rehabilitation training system according to claim 1, characterized in that, The EEG signal described in step S2 is filtered by a 50Hz power frequency notch filter to eliminate power interference, and then filtered by a 0.5-45Hz bandpass filter to extract the effective frequency band. The electroretinogram signal is smoothed and filtered to eliminate high-frequency noise; The denoised EEG and TE signals were used as a personalized physiological baseline, and personalized physiological baseline feature values ​​were calculated. These personalized physiological baseline feature values ​​included the power of each frequency band of the EEG. and mean skin conductance The calculation formula is as follows: （1）； （2）； In the above formula, The total duration of baseline acquisition is expressed in seconds. It is a function of α-wave power as a function of time; It is a function of skin conductance as a function of time; It is a time variable.

3. The brainwave-cortical electroreceptor linkage dance rehabilitation training system according to claim 2, characterized in that, The preprocessing of EEG and ductal nerve signals in step S3 is as follows: Step S31: After the intervention begins, EEG signals are collected synchronously according to the control cycle. and skin conductance signals : （3）； （4）； In the above formula, for The amplitude of EEG signals collected at any time; For length is The amplitude of the EEG signal collected at the end of the time window; This refers to the length of the EEG signal sampling window. The sampling frequency of the EEG signal, measured in Hz; for The amplitude of the skin conductance signal collected at any time; For a length of The amplitude of the skin conductance signal collected at the end of the time window; This is the length of the skin conductance signal sampling window; The sampling frequency of the electrodermal signal, in Hz; Step S32: Preprocess the EEG signals; First, the power spectral density is calculated using the Welch method. : （5）； In the above formula, The length of the window; As the normalization factor, , For the Hanning window function, For window function index; The value of the EEG signal at the m-th sampling point; is the base of the natural logarithm; The imaginary unit; For frequency; Then calculate the power of each frequency band. : （6）； In the above formula, This is the upper limit frequency of the frequency band; This is the lower limit frequency of the frequency band; This is the index corresponding to the upper limit of the frequency. This is the index corresponding to the lower frequency limit; Let K be the power spectral density at the Kth frequency point; Frequency resolution; extract , , , , The absolute and relative power of the bands are measured, and artifact detection is performed simultaneously. Based on the detected amplitude, gradient, and energy threshold, blink and electromyographic interference signals in the EEG signal are identified. Step S33: Process the skin conductance signal; First, smoothing filtering is performed to eliminate high-frequency noise. Then, the Tonic / Phasic decomposition method is used to separate the low-frequency trend component Tonic and the high-frequency fluctuation component Phasic. The morphological characteristics of skin conductance response (SCR) events are detected in the Phasic component. （7）； （8）； In the above formula, Represents a low-pass filter function; This represents the original skin conductance signal; The cutoff frequency; The detection conditions for the skin conductance response (SCR) event are set as follows: rise time 0.5-5.0s, amplitude threshold > 0.05μs, and minimum interval > 2s.

4. The brainwave-cortical electroreceptor linkage dance rehabilitation training system according to claim 3, characterized in that, The process of multimodal feature fusion and emotion recognition in step S4 is as follows: A 20-dimensional fusion feature vector was extracted from the preprocessed EEG and SC signals. : （9）； In the above formula, Represents 6-dimensional brainwaves Features, including power, power, power, ratio, Ratio and signal entropy; The six-dimensional skin conductance (SC) characteristics include SC mean, SC standard deviation, SC slope, response rate, tonic component, and pulsed component. Representing 2D interactive features, including Power × SC mean sum Power × SC slope; It represents 3-dimensional statistical characteristics, including EEG mean, EEG standard deviation, and SC entropy; It represents 3D historical features, including feature change trends based on the most recent 5 time windows; The extracted 20-dimensional fusion features are standardized based on baseline data: （10）； In the above formula, The value of the i-th feature after standardization; This represents the original value of the i-th feature; Let be the baseline mean of the i-th feature; Let be the baseline standard deviation of the i-th feature; A dual-model fusion architecture of SVM and random forest is used for emotion recognition: （11）； （12）； In the above formula, This represents the fused emotion category probability vector; This is the probability vector of the sentiment category output by the SVM model; This is the probability vector of the sentiment category output by the random forest model; Rate your mood; Indicates the first The basic values ​​for various emotional states; Introducing an inertia factor for smoothing: （13）； In the above formula, The emotion score after smoothing at time t; The current emotion score at time t; The emotion score after smoothing from the previous moment; Emotional state is determined based on the smoothed sentiment score: （14）； In the above formula, Indicates emotional state; For deep relaxation; For relaxation; Neutral; To be vigilant; For stress; This is due to severe stress.

5. The brainwave-cortical electroreceptor linkage dance rehabilitation training system according to claim 4, characterized in that, The process of personalized safety threshold decision-making and intervention level determination in step S5 is as follows: Based on established personalized physiological baseline characteristics and historical data, a dynamic safety threshold model is used to determine a state of stress when the following conditions are met: （15）； In the above formula, Pressure threshold; Based on the emotional state and the current intervention level, an adaptive strategy is used to determine the updated intervention level: （16）； In the above formula, Indicates the level of intervention after the update; Indicates the current intervention level; This represents the smoothed sentiment score at time t; A state transition probability matrix is ​​used for progressive adjustment of the intervention level. This matrix defines the transition probability from the current intervention level. and emotional state Move to the next intervention level The probability distribution of is expressed as: （17）； In the above formula, This is a state transition probability matrix, where rows correspond to the current intervention level (0-3) and columns correspond to the next intervention level (0-3). Based on the current intervention level (CurrentLevel), select the corresponding row vector as the probability distribution, and determine the NextLevel through random sampling or the maximum probability method. Set safety constraints: maximum stress duration seconds, minimum relaxation time Seconds, frequency of intervention level changes times per minute.

6. The brainwave-cortical electroreceptor linkage dance rehabilitation training system according to claim 5, characterized in that, The process of multi-channel dance intervention in step S6 is as follows: According to the determined intervention level, the corresponding dance intervention is carried out. 3-5 movement sequences suitable for the trainee's ability level and sensory preferences are selected from the preset movement library, and the music rhythm is adjusted in sync. The beats per minute (BPM) is linearly related to the intervention level. （18）； In the above formula, Intervention level; A random integer uniformly distributed in the range [-5, 5] is used to add natural variations to the musical rhythm; The correspondence between intervention levels and BPM is as follows: Intervention level 0 corresponds to 50-60 BPM, with breathing guidance and static movements performed; Intervention level 1 corresponds to 65-75 BPM, with gentle intervention and simple stretching movements performed; Intervention level 2 corresponds to 80-90 BPM, with moderate intervention and rhythmic movements performed; Intervention level 3 corresponds to 95-105 BPM, with strong intervention and coordinated movements performed. The preset motion library includes movements such as stretching, swaying, clapping, stepping, and rotating.

7. The brainwave-cortical electroreceptor linkage dance rehabilitation training system according to claim 6, characterized in that, In step S7, when a trainee's emotional value ≥7 and the duration >10s is detected as a stress state, an emergency treatment mechanism is immediately triggered: the intervention level is forcibly adjusted to level 0, the breathing guidance mode is executed, the music is adjusted to 60±5BPM, the visual feedback is switched to the purple breathing mode, the vibration intensity is increased, and the stress event information is recorded and saved.

8. The brainwave-cortical electroreceptor linkage dance rehabilitation training system according to claim 7, characterized in that, Control cycle in step S8 The calculation formula is as follows: （19）； In the above formula, For signal acquisition time; For signal processing and feature extraction time; Timeframes for emotion recognition and decision-making; To intervene in the execution time.

9. The brainwave-cortical electroreceptor linkage dance rehabilitation training system according to claim 8, characterized in that, The comprehensive assessment report mentioned in step S9 includes an emotion change curve, physiological indicator trends, intervention effect analysis, and personalized recommendations; The emotion change curve is based on the smoothed emotion score at time t. Corresponding emotional state get; The physiological indicators include Wave power, skin conductance mean Ratio and intervention level; The intervention effectiveness analysis was performed by calculating participation scores. and improve scores get: （20）； （21）； In the above formula, This represents the participation score; The proportion of smoothed emotion scores Esmooth(t) falling in the range of 3 to 6; The standard deviation of the sentiment score E; The number of times the intervention level changes; This represents the total number of records. This represents the average sentiment score during the first half of the conversation. This represents the average sentiment score for the second half of the conversation. Based on the obtained emotional change curves, physiological indicator trends, and intervention effect analysis results, corresponding personalized suggestions are given.

Images