Quantitative evaluation method for interference effect of trombone playing training on suppression control capability

Abstract

The invention discloses a quantitative evaluation method for an interference effect of trombone playing training on suppression control ability, which comprises the following steps of: acquiring pulse wave signals in various states when a subject performs trombone playing training, denoising, and extracting physiological features from the pulse wave signals; the method comprises the following steps: collecting behavior data when a subject executes playing training, including a playing airflow signal and position control data, and extracting behavior characteristics from the behavior data; when the correct rate and the response of executing the inhibition control test task by the subject are recorded, calculating the score of the test task through probability distribution fitting; and constructing a data set by taking the physiological features and behavior features of the subject as explanatory variables and taking the test task score as a target variable, training a regression model and screening an optimal feature subset to construct a quantitative formula, and obtaining a quantitative evaluation result of the intervention effect. According to the invention, a quantitative analysis method is provided for research on the intervention effect of the music therapy on the suppression control ability by relying on trombone playing training.

Description

Technical Field

[0001] The present invention belongs to the technical field of information technology and intelligent health, and particularly relates to a quantitative evaluation method for the intervention effect of trombone playing training on inhibitory control ability. Background Art

[0002] Inhibitory Control ability refers to the ability of an individual to inhibit the dominant or automatic response in the face of interfering stimuli or conflicting information and focus on executing the goal of the current cognitive activity. It is an important part of the brain's executive function and plays a key role in higher cognitive activities such as individual cognitive decision-making and behavior planning.

[0003] Currently, the intervention research on inhibitory control ability mainly focuses on three aspects: cognitive training, exercise intervention, and music therapy. In these intervention studies, the evaluation of the intervention effect mostly adopts three evaluation paradigms: subjective scale measurement, "pre-intervention - post-intervention test", and "intervention group - control group". The subjective scale measurement method requires the subjects to express their evaluation feelings on various dimensions of the intervention effect, and this method is highly subjective. The "pre-intervention - post-intervention test" method requires the subjects to perform the inhibitory control test task paradigm multiple times, and evaluates the intervention effect by comparing the performance of the test tasks before and after the intervention of the subjects. It lacks the dynamic analysis of the intervention process and the data dimension is single. The "intervention group - control group" paradigm, on the basis of the "pre-intervention - post-intervention test", further judges the effectiveness of the intervention by comparing the performance of the test tasks of different experimental groups, but also lacks the data analysis of the intervention process. Most of the above three evaluation paradigms evaluate the effectiveness of the intervention through significance tests, with a single evaluation dimension and lack of quantitative indicators, making it difficult to construct a quantitative evaluation model of "intervention input - performance output" for the intervention effect and unable to realize the dynamic monitoring of the intervention process. Summary of the Invention

[0004] To overcome the defects and deficiencies of the intervention effect evaluation method in the current intervention research on inhibitory control ability, the present invention provides a quantitative evaluation method for the intervention effect of trombone playing training on inhibitory control ability. This method relies on the music therapy of trombone playing training, and conducts modeling analysis on the scores of inhibitory control test tasks based on the physiological characteristics and behavioral characteristics of each stage of the playing training, realizing the multi-modal quantitative evaluation of the intervention effect, which has practical significance for the intervention research on inhibitory control ability.

[0005] To achieve the above object, the present invention adopts the following technical solutions: A quantitative evaluation method for the intervention effect of trombone playing training on inhibitory control ability, the quantitative evaluation method comprising the following steps: S1. Use photoplethysmography (PPG) to collect the baseline pulse wave signal, task-state pulse wave signal, and recovery-state pulse wave signal of the subject during the performance of the AI trombone playing training. Collect the blowing airflow signal of the subject during the performance of the playing training through a pickup module, record the level signal output by the ultrasonic ranging module, which is the original fingering control data, and record the accuracy rate and reaction time of the inhibitory control test task.

[0006] Among them, the AI trombone is an intelligent instrument obtained by transforming a traditional trombone in the laboratory and is mainly used for data collection in the present invention. The baseline pulse wave signal refers to the pulse wave signal collected in the resting state of the subject 5 minutes before the performance of the playing training; the state of performing the playing training is called the task state, and the pulse wave signal collected in the task state is called the task-state pulse wave signal; the state 5 minutes after the performance of the playing training is called the recovery state, and the pulse wave signal collected in the recovery state is called the task-state pulse wave signal. The inhibitory control test task includes the Stroop test task and the variant Flanker test task.

[0007] S2. For the level signal obtained in step S1, use the variational mode decomposition and wavelet threshold joint denoising method to filter out high-frequency noise and reduce the influence of motion artifacts. At the same time, further improve the quality of the pulse wave signal through filtering by a Butterworth band-pass filter. Use the inverse traversal method based on the first-order forward difference to extract the starting point of each pulse wave cycle in the pulse wave signal, and use the cubic spline interpolation method to remove the baseline drift of the pulse wave signal. After the above processing, the denoised baseline-state pulse wave signal, denoised task-state pulse wave signal, and denoised recovery-state pulse wave signal are obtained. For the blowing airflow signal, use notch filter filtering and median filtering for denoising processing to obtain the denoised blowing airflow signal. Perform data conversion on the original fingering control data to obtain the fingering control data represented by a distance sequence.

[0008] S3. Identify the main wave peak points of the denoised baseline-state pulse wave signal, denoised task-state pulse wave signal, and denoised recovery-state pulse wave signal obtained in step S2, and extract the heart rate variability characteristics on this basis. For the denoised blowing airflow signal, segment it based on the playing training sound score, and extract the blowing breath stability characteristics. For the fingering control data, segment it in the same way and extract the fingering control accuracy characteristics based on statistical principles.

[0009] This step extracts the physiological characteristics and behavioral characteristics of the subject during the performance of the playing training, and these characteristics will be used as the explanatory variables of the intervention effect quantitative evaluation method.

[0010] S4. Perform probability distribution fitting on the reaction time of the inhibitory control test task based on histogram analysis and goodness-of-fit test, and calculate the test task score by calculating the cumulative probability and combining the test task accuracy rate.

[0011] This step takes into account that the performance of the inhibitory control test task indirectly reflects the subject's inhibitory control ability. Therefore, the test task score is calculated using the correct rate and reaction time of the inhibitory control test task, and the inhibitory control ability is characterized by the test task score.

[0012] S5. Construct a dataset based on the physiological and behavioral characteristics of the subject, train a random forest regression model with the test task score obtained in step S4 as the target variable, evaluate the performance of different feature subsets through recursive feature elimination and cross-validation, and select the feature subset with the highest performance, that is, the optimal feature subset. Among them, the physiological characteristics are the pulse rate variability characteristics extracted from the baseline state pulse wave signal, task state pulse wave signal, and recovery state pulse wave signal. The behavioral characteristics are the characteristics of the stability of the blowing breath and the accuracy of the finger position control extracted from the blowing air flow signal and finger position control data.

[0013] S6. Conduct a correlation analysis on the optimal feature subset obtained in step S5, dimensionlessize the features according to the analysis results, transform the direction of all features to be positively correlated with the test task score, that is, positively correlated with the inhibitory control ability, retrain the random forest regression model, construct a quantitative evaluation formula based on the feature importance, and obtain the "trombone playing training - inhibitory control ability" quantitative score. This score explains the influence of physiological and behavioral characteristics on the inhibitory control ability during the playing training process. By analyzing the change of the quantitative score, the quantitative evaluation of the intervention effect of trombone playing training on the inhibitory control ability is realized.

[0014] Furthermore, the process of step S1 is as follows: S101. The subject sits in front of a screen, and the UI interface of the trombone playing training module is displayed on the display screen. There is a PPG sensor on the desktop where the display screen is located. First, a 5-minute countdown is displayed on the screen, and at the same time, the subject is prompted to adjust the position of the sensor and place the middle finger of the left hand on the PPG sensor, sit quietly for 5 minutes, and collect the baseline pulse wave signal of the subject before training. After the countdown ends, the screen prompts to enter the playing training.

[0015] S102. The playing training fingering chart is displayed on the screen. During the training process, the subject is required to keep the upper body upright, wear headphones, hold the slide tube of the AI trombone with the right hand, hold the grip of the AI trombone with the left hand, put on the PPG sensor finger sleeve for the middle finger of the left hand, tilt the bell of the AI trombone downward by 15 degrees, and align the lips with the mouthpiece of the AI trombone naturally; among them, the PPG sensor finger sleeve is used to collect the task state pulse wave signal.

[0016] This step is to guide the subjects to conduct the experiment in the correct posture. Using headphones to receive the playing feedback is to prevent the pickup from collecting the trombone sound. Using the subject's fixed hand as the hand for detecting the pulse wave signal is to reduce the motion artifacts caused by arm swinging.

[0017] S103. The sound fingering chart is played in the form of an animation on the screen. The subject needs to play the trombone with a steady breath and move the slide to the correct position according to the notes and position tips shown in the sound fingering chart. The playing training consists of several rounds. In each round, a sound fingering chart is randomly played. The pickup module installed on the AI trombone is used to collect the playing airflow signal of the subject during the playing training, and the level signal output by the ultrasonic ranging module is recorded, which is the original position control data.

[0018] S104. After the playing training, the subject is prompted on the screen to remove the finger cot and place the middle finger of the left hand on the desktop PPG sensor. A 5-minute countdown is displayed, and the subject is prompted to sit still for 5 minutes to collect the subject's resting-state pulse wave signal.

[0019] S105. The subject sits in front of a display screen, and the UI interface of the inhibitory control test task module is displayed on the screen. The inhibitory control test task consists of the Stroop test task and the variant Flanker test task. The screen first prompts the subject to perform the Stroop test task, that is, the font color recognition task. A word representing a color is displayed in the center of the screen. When the font color of the word is the same as the color represented by the word, it is called a congruent task. When the font color of the word is different from the color represented by the word, it is called an incongruent task. The subject needs to select from the color options below the word. After the subject makes a selection, the screen will display the answering accuracy rate and reaction time of this question and automatically enter the next question after 1 s until all Stroop test tasks are completed. The accuracy rate and reaction time of each test task are recorded.

[0020] S106. After the subject has completed all Stroop test tasks, the display screen prompts the subject to perform the variant Flanker test task. A 7x7 matrix composed of directional arrows is displayed in the center of the display screen. There are at most two directions among the directional arrows in the matrix. If there are two directions, these two directions must be opposite directions. All the directional arrows except those in the middle column of the matrix point in the same direction, and this direction is called the row interference direction; all the directional arrows except those in the middle row in the middle column of the matrix point in the same direction, and this direction is called the column interference direction; the directional arrow at the center of the diagonal of the matrix is called the target directional arrow. The subject needs to judge the directionality of the row interference direction and the column interference direction, and press the keyboard arrow keys according to the direction of the target directional arrow. When the row interference direction and the column interference direction are the same, the subject needs to press the keyboard arrow key in the same direction as the target directional arrow, and at this time the test task is called a consistent task; when the row interference direction and the column interference direction are opposite, the subject needs to press the keyboard arrow key in the opposite direction to the target directional arrow, and at this time the test task is called an inconsistent task. Among them, the opposite direction of the direction " " is " ", and the opposite direction of the direction " " is " ". After the subject presses the keyboard arrow keys, the response accuracy rate and reaction time of this question will be displayed on the screen, and it will automatically enter the next question after 1 s until all variant Flanker test tasks are completed. Record the accuracy rate and reaction time of each test task; Further, the process of step S2 is as follows: S201. Preprocess the pulse wave signals collected in step S1. The specific steps are as follows: S201.1. Use variational mode decomposition on the baseline-state pulse wave signal, task-state pulse wave signal, and recovery-state pulse wave signal to obtain multiple intrinsic mode function (IMF) components. Perform a fast Fourier transform on each IMF component and calculate the frequency corresponding to the maximum peak in the spectrogram .

[0021] S201.2. For the high-frequency IMF components of , use the wavelet threshold method for denoising to filter out high-frequency noise and reduce the influence of motion artifacts; among them, the is the maximum heart rate frequency of the normal population.

[0022] S201.3. Superimpose the high-frequency IMF components processed in step S201.2 with the remaining IMF components to reconstruct the pulse wave signal.

[0023] S201.4. Filter the reconstructed pulse wave signal using a Butterworth band-pass filter to further improve the quality and signal-to-noise ratio of the pulse wave signal.

[0024] S201.5. Use the first-order forward difference reverse traversal method to extract the starting point of each pulse wave cycle in the pulse wave signal, and obtain the pulse wave cycle starting point sequence.

[0025] S201.6. Perform cubic spline interpolation on the pulse wave cycle starting point sequence, fit the pulse wave baseline, remove the baseline drift of the pulse wave signal, and obtain the denoised baseline state pulse wave signal, denoised task state pulse wave signal, and denoised recovery state pulse wave signal.

[0026] S202. For the blowing airflow signal collected in step S1, use a notch filter and median filter for denoising.

[0027] The design principle of the notch filter is as follows: The transfer function of the notch filter under the Laplace transform can be expressed as:

[0028]

[0029] Among them, is the notch center frequency, , is the notch factor; The notch filter is mainly designed depending on 3 indicators: (1) : The notch center angular frequency, ; (2) : The notch depth, that is, the attenuation depth at the center frequency; (3) : The notch bandwidth, , and are the cut-off angular frequencies.

[0030] The notch factor can be obtained by solving through the Z-transform:

[0031] Among them, , substitute , into formula (7) to complete the design of the notch filter.

[0032] This step takes into account that the blowing airflow signal is essentially a non-stationary broadband signal, and there is no need to extract specific frequency components, but the signal may be interfered by power frequency noise and pulse interference. Therefore, a notch filter and median filter are selected for denoising.

[0033] S203. For the original finger position control data obtained in S1, according to the speed of sound and the high-level duration Convert the level signal into distance to obtain the position control data represented by a distance sequence.

[0034] The distance conversion formula is as follows:

[0035] wherein, is the distance between the ultrasonic transmitter and the receiver.

[0036] Further, the process of step S3 is as follows: S301. Extract the pulse rate variability features from the denoised baseline state pulse wave signal, denoised task state pulse wave signal, and denoised recovery state pulse wave signal obtained in step S2.

[0037] S301.1. Use the first-order forward difference traversal method to identify the main wave peak points of the pulse wave signal.

[0038] S301.2. Perform a difference operation on the main wave peak point sequence to obtain a time interval sequence between adjacent two pulse wave peaks, that is, the PP interval sequence.

[0039] S301.3. Estimate the frequency domain power spectrum of the PP interval sequence through an autoregressive model (AutoRegressive Model, AR model).

[0040] The expression of the p - order AR model is as follows:

[0041] wherein, is the sequence value at time , is the constant term, is the autoregressive coefficient, indicating the influence of the th lag term on the current sequence value, is white noise, .

[0042] The AR model calculates the autoregressive coefficient through a parameter estimation method based on the least squares method, and obtains the transfer function of the AR model as follows:

[0043] Further, there is the -order AR model spectrum estimation result of the PP interval sequence:

[0044] wherein, the noise variance can be calculated by the following formula:

[0045] Among them, is the length of the PP interval sequence, indicating the predicted value of the AR model for the sequence at the moment.

[0046] This step takes into account that extracting the frequency-domain features of pulse rate variability requires a high frequency resolution. Using cubic spline interpolation to reconstruct the PP interval sequence is easily limited by the interpolation frequency and has a low frequency resolution. Therefore, an autoregressive model of order is selected, and the power spectral density of the PP interval sequence is estimated by calculating the autoregressive coefficients.

[0047] S301.4. Extract the time-domain, frequency-domain, and non-linear features of pulse rate variability (PRV) from the PP interval sequence. Extract the PRV features from the baseline pulse wave signal to obtain the baseline PRV features, and form the baseline PRV feature vector ; extract the PRV features from the task-state pulse wave signal to obtain the task-state PRV features, subtract the baseline PRV features from the task-state PRV features to obtain the baseline-removed task-state PRV features, and form the baseline-removed task-state PRV feature vector ; extract the PRV features from the recovery-state pulse wave signal to obtain the recovery-state PRV features, subtract the baseline PRV features from the recovery-state PRV features to obtain the baseline-removed recovery-state PRV features, and form the baseline-removed recovery-state PRV feature vector .

[0048] The pulse rate variability features extracted in this step include 6 time-domain features: mean pulse interval (PRV_MEANPP), standard deviation of pulse interval (PRV_SDPP), root mean square of difference between adjacent pulse intervals (PRV_RMSSD), proportion of the number of times the difference between adjacent pulse intervals is greater than 20 ms and 50 ms in the total number of pulse intervals (PRV_PNN20, PRV_PNN50), and triangular index (TriangleIndex, PRV_TI); 3 frequency-domain features: low-frequency band power (PRV_LF), high-frequency band power (PRV_HF), and ratio of low-frequency band power to high-frequency band power (PRV_LF / PRV_HF); and 4 non-linear features: sample entropy (PRV_SampEn), short-term change of Poincare plot (PRV_SD1), long-term change of Poincare plot (PRV_SD2), and PRV_SD1 / PRV_SD2; The specific calculation steps of the sample entropy are as follows: 1) Assume that the length of the PP interval sequence is For a given pattern dimension a set of dimensional vectors can be constructed. Usually take , then there is

[0049] 2) Set the similarity tolerance threshold at the endpoint. Assume , , the intervals respectively represent and the tolerance ranges. If the corresponding endpoints of are all within the tolerance range, then it is considered that and are similar under the similarity tolerance threshold . Define the ratio of the approximate number of vectors and to the total number of the two as . The average value of the ratio of the approximate number to the total number can be obtained:

[0050] 3) Let , repeat the above steps to obtain . Finally, the sample entropy of the sequence is obtained:

[0051] S302. For the preprocessed blowing airflow signal obtained in step S2, segment it based on the blowing training sound handwritten score measures. Each segment contains the same number of notes. Use the sliding window algorithm to frame each segment of the signal, calculate the short-time energy (Short-Time Energy, STE) of each frame of the signal to obtain the short-time energy sequence, calculate the coefficient of variation (Coefficient of Variation, CV) and sample entropy of each short-time energy sequence, and finally take the mean to obtain the average coefficient of variation (STE_MEANCV) and average sample entropy (MEANSampEn) of the short-time energy sequence. At the same time, calculate the standard deviation (STE_MEAN_SD) of the mean value of each short-time energy sequence.

[0052] This step takes into account that the blowing airflow signal is essentially a non-stationary broadband signal. Therefore, the short-time energy is calculated by framing, and the stability of the blowing breath is characterized by evaluating the fluctuation of the short-time energy. Among them, STE_MEANCV and MEANSampEn reflect the stability of the blowing breath of each segment of notes of the subject, and STE_MEAN_SD reflects the consistency of the blowing breath stability of each segment of notes of the subject, that is, the overall blowing breath stability.

[0053] The calculation formula for short-time energy is as follows:

[0054] Wherein, represents the short-time energy of the th frame, is the frame length, is the th sampling value of the sampling point.

[0055] CV is the ratio of the signal standard deviation and the mean value , and the formula is as follows:

[0056] S303. For the fingering (Slide Position) control data obtained in step S2, segment it based on the musical score measures of the playing training sound. Each segment contains the same number of notes, and each note has its corresponding target fingering. Calculate the mean absolute error (SP_MAE) and root mean square error (SP_RMSE) of the distance between the actual fingering and the target fingering, and further calculate the fingering control accuracy rate (SP_ACC).

[0057] The AI trombone has a total of 7 fingerings. The distance from the th fingering to the ranging module is , then there is a fingering distance sequence . When the actual fingering meets the fingering decision criterion, the AI trombone determines the actual fingering as the target fingering. The fingering decision criterion is as follows:

[0058] Wherein, is the actual fingering, represents that the actual fingering is determined to be the th fingering, is the distance from the actual fingering to the ranging module, is the distance from the maximum length position of the slide tube to the ranging module.

[0059] Based on the fingering decision criterion, for a note segment with a duration of , if the duration for which the actual fingering is determined to be the target fingering is less than , it is considered that the note segment is played incorrectly. Therefore, the calculation method of SP_ACC is as follows:

[0060] Furthermore, the process of step S4 is as follows: S401. Plot a reaction time histogram for the reaction time of each inhibitory control test task to analyze the data distribution pattern.

[0061] S402. Through the Anderson-Darling (A-D) goodness-of-fit test, select the target probability distribution model that accepts the null hypothesis as the probability distribution of the reaction time, and obtain the cumulative probability of the reaction time. . Among them, the null hypothesis is that the reaction time follows the target probability distribution.

[0062] This step is to verify the rationality of fitting the target probability distribution for the reaction time.

[0063] S403. Regarding the correct rate of the test task, since the correct rate is already a normalized index, multiply the correct rate of each test task by the reaction time of that task to obtain the score of each test task, and add up the scores of each test task to get the score of the subject for this test task.

[0064] Furthermore, the process of step S5 is as follows: S501. Construct a data set based on the pulse rate variability features, blowing breath stability features, and fingering control accuracy features extracted in step S3.

[0065] S502. Using the test task score obtained in step S4 as the target variable and the physiological and behavioral features as the explanatory variables, train a random forest regression model.

[0066] S503. Taking the random forest regression model obtained in step S502 as the base model, use recursive feature elimination to gradually eliminate the features with the lowest importance, and calculate the RMSE and coefficient of determination (Coefficient of Determination, R 2 ) of the feature subset through cross-validation to evaluate the performance of the feature subset at each recursion. Finally, select the feature subset with the highest R 2 and the lowest RMSE. If there is a conflict between R² and RMSE, preferentially select the feature subset with the highest R². Among them, the RMSE and R 2 are the most commonly used indicators for evaluating the performance of the regression model in cross-validation, and the calculation formulas are as follows:

[0067]

[0068] Among them, is the number of samples, is the true value of the th sample, which represents the test task score here, is the The predicted value of a sample, is the average value of the true values of the samples.

[0069] Furthermore, the process of step S6 is as follows: S601. Perform Spearman correlation analysis on the optimal feature subset obtained in step S5 to obtain the positive and negative correlations between each feature in the subset and the test task score.

[0070] The principle of Spearman correlation analysis is to convert the feature values of the samples into rank data, that is, the ranking of the feature value for all samples of this feature. By calculating the correlation between the rank data, the Spearman rank correlation coefficient is obtained , and the value range is , indicates that the relationship between the two variables is positively correlated, indicates that the relationship between the two variables is negatively correlated, indicates that there is no correlation between the two variables, The calculation formula is as follows:

[0071] Among them, is the rank difference of the th observation value of the two variables, is the number of samples.

[0072] This step takes into account that the physiological and behavioral characteristics and the target variable may not be a single linear correlation. Therefore, Spearman correlation analysis is selected to measure the monotonic relationship between the feature and the target variable based on the correlation between the data ranks.

[0073] S602. Based on the positive and negative correlations obtained in step S601, perform Min-Max normalization on each feature.

[0074] For positively correlated features, the Min-Max normalization expression is as follows:

[0075] Among them, represents the th sample's original data of feature , represents the normalized data, , are the minimum and maximum values of this feature in the dataset respectively.

[0076] For negatively correlated features, the Min-Max inverse normalization expression is as follows:

[0077] This step is to unify the feature directions of all features, eliminate the differences in feature correlations, and facilitate the construction of subsequent quantitative evaluation formulas.

[0078] S603. Retrain the regression model, construct a quantitative evaluation formula based on feature importance, and obtain the quantitative score of "trombone playing training - inhibitory control ability". This score explains the influence of physiological and behavioral features on inhibitory control ability during the playing training process. By analyzing the changes in the quantitative score, the quantitative evaluation of the intervention effect of trombone playing training on inhibitory control ability is realized.

[0079] Compared with the prior art, the present invention has the following advantages and beneficial effects: 1) The present invention collects pulse wave, blowing, and finger position distance data during trombone playing training through low-cost and easy-to-use data acquisition devices, extracts physiological and behavioral features from them to establish a prediction model for the inhibitory control test score, constructs a multi-modal quantitative evaluation technical solution for the intervention effect of trombone playing training on inhibitory control ability, realizes the objective and dynamic evaluation of the intervention effect, and overcomes the problems of strong subjectivity and single evaluation index dimension in traditional methods.

[0080] 2) The present invention proposes a method for calculating the score of the inhibitory control test task based on statistics. Through histogram analysis and goodness-of-fit test, the probability distribution of the reaction time of the inhibitory control test task is fitted, the cumulative probability of each reaction time is calculated, and by weighted summing the cumulative probability and the corresponding test task correct rate, an objective and scientific score of the inhibitory control test task is obtained, establishing a reliable standard for quantitatively evaluating the intervention effect.

[0081] 3) By combining feature selection and cross-validation, the present invention calculates the importance of physiological and behavioral features, and according to the optimal number of features obtained from cross-validation, uses the recursive feature elimination algorithm to eliminate the features with the lowest importance one by one, removing redundant and irrelevant features, and improving the operation efficiency and accuracy.

[0082] 4) The present invention eliminates the dimension of physiological and behavioral features through dimensionless processing, and obtains the quantitative score result of the intervention effect by weighted summing with the importance weights of each feature. The quantitative score can help people establish an accurate and intuitive understanding of the intervention effect and help people formulate a more scientific training plan. In addition, by adapting to the specific behavioral features of different instrument performances, the evaluation technical solution proposed by the present invention can be applied to the intervention research of different instrument performances. Description of the Drawings

[0083] To more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0084] Figure 1 is a flowchart of a quantitative evaluation method for the intervention effect of trombone playing training on inhibitory control ability disclosed in Embodiment 1 of the present invention; Figure 2 is a structural diagram of the experimental system module in Embodiment 1 of the present invention; Figure 3 is a flowchart of the preprocessing of physiological and behavioral data in Embodiment 1 of the present invention; Figure 4 is a flowchart of the extraction of physiological and behavioral characteristics in Embodiment 1 of the present invention; Figure 5 is a flowchart of establishing a quantitative evaluation model for the intervention effect based on physiological and behavioral characteristics in Embodiment 1 of the present invention; Figure 6 is the original pulse wave signal waveform diagram in Embodiment 2 of the present invention; Figure 7 is the waveform diagram of the blowing air flow signal in Embodiment 2 of the present invention; Figure 8 is the schematic diagram of the fingering control data sequence in Embodiment 2 of the present invention; Figure 9 is the recognition result diagram of the characteristic points of the denoised pulse wave signal in Embodiment 2 of the present invention; Figure 10 is the distribution curve diagram of the reaction time of inconsistent tasks in Embodiment 2 of the present invention; Figure 11 is the RMSE and R of different feature subsets during recursive feature elimination in Embodiment 2 of the present invention 2 Schematic diagram. Detailed implementation manners

[0085] To enable those skilled in the art of this technology to better understand the solution of the present application, the following will clearly and completely describe the technical solutions in the embodiments of the present application in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0086] References to "embodiments" in this application mean that the specific features, structures, or characteristics described in connection with the embodiments can be included in at least one embodiment of this application. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described in this application can be combined with other embodiments.

[0087] Embodiment 1 Figure 1 is a flowchart of a quantitative evaluation method for the intervention effect of trombone playing training on inhibitory control ability provided by an embodiment of the present invention. The specific steps are as follows: S1. Collect the baseline pulse wave signal, task-state pulse wave signal, and recovery-state pulse wave signal of the subject during the trombone playing training using the AI trombone. Collect the blowing airflow signal of the subject during the trombone playing training through the pickup module, record the level signal output by the ultrasonic ranging module, that is, the original position control data, and record the correct rate and reaction time of the inhibitory control test task.

[0088] Among them, the AI trombone is an intelligent instrument obtained by transforming a traditional trombone in the laboratory and is mainly used for data collection in the present invention. The state 5 minutes before the start of the trombone playing training is called the baseline state, and the pulse wave signal collected in this state is called the baseline pulse wave signal; the state of performing the trombone playing training is called the task state, and the pulse wave signal collected in the task state is called the task-state pulse wave signal; the state 5 minutes after the trombone playing training is called the recovery state, and the pulse wave signal collected in the recovery state is called the task-state pulse wave signal.

[0089] The place where the subject performs the trombone playing training and the inhibitory control test task is a stable, quiet, and comfortable indoor place. Complete the trombone playing training and test tasks on the experimental system according to the operator's instructions. The module structure diagram of the experimental system is as Figure 2 shown.

[0090] S2. Preprocess the collected physiological data and behavioral data.

[0091] In this embodiment, the collected pulse wave signal is subject to noise interference such as baseline drift, electromyographic noise, and power frequency noise, and the blowing airflow signal is subject to noise interference such as power frequency noise and pulse noise. The flowchart of the preprocessing of physiological and behavioral signals is as Figure 2 shown, and the specific steps are as follows: S201. Preprocess the collected pulse wave signal. The specific steps are as follows: S201.1. Use variational mode decomposition on the baseline pulse wave signal, task-state pulse wave signal, and recovery-state pulse wave signal to obtain 10 IMF components. Perform a fast Fourier transform on each IMF component and calculate the frequency corresponding to the maximum peak in the spectrogram. .

[0092] S201.2. Considering that the human heart rate range is 45 - 180 bpm, corresponding to frequencies of 0.75 - 3 Hz, for the IMF components, use the wavelet threshold method for denoising to filter out high-frequency noise and reduce the influence of motion artifacts.

[0093] The parameters of the wavelet threshold method used in this embodiment are the "db9" wavelet basis function, the decomposition level is 5, the global threshold, and the hard threshold function. The global threshold formula is as follows: , where is the noise standard deviation, is the signal length.

[0094] S201.3. Superimpose the denoised high-frequency IMF components with the remaining IMF components to reconstruct the pulse wave signal.

[0095] S201.4. For the reconstructed pulse wave signal, use a 4th-order Butterworth band-pass filter with a passband frequency range of 0.5 - 3.25 Hz to further improve the quality and signal-to-noise ratio of the pulse wave signal.

[0096] S201.5. Use the first-order forward difference reverse traversal method to extract the starting point of each pulse wave cycle in the pulse wave signal.

[0097] The execution steps of the first-order forward difference reverse traversal method are as follows: 1) Perform a first-order forward difference operation on the pulse wave signal to obtain the first-order difference sequence of the pulse wave signal. The formula is as follows:

[0098] 2) Extract the maximum points in the first-order difference sequence where the interval is greater than , that is, the rising peak points of the pulse wave. Among them, is the sampling rate of the pulse wave signal, is the estimated frequency obtained by performing a fast Fourier transform on the pulse wave signal. In this embodiment, .

[0099] 3) Based on the rising peak points, forward search for the difference zeros in the difference sequence to obtain the starting points of the pulse wave cycles.

[0100] S201.6. Perform cubic spline interpolation on the pulse wave cycle start point sequence to fit the pulse wave baseline, remove the baseline drift of the baseline state pulse wave signal, task state pulse wave signal, and recovery state pulse wave signal, and obtain the denoised baseline state pulse wave signal, denoised task state pulse wave signal, and denoised recovery state pulse wave signal.

[0101] S202. For the collected blowing air flow signal, use , , a notch filter with a bandwidth of 5 Hz to filter out power frequency noise interference, and use median filtering with a window size of 21 to filter out pulse noise interference.

[0102] S203. For the collected original finger position control data, according to the sound speed and the high-level duration convert the level signal into distance to obtain the finger position control data represented by a distance sequence.

[0103] S3. Identify the main wave peak points of the denoised baseline state pulse wave signal, denoised task state pulse wave signal, and denoised recovery state pulse wave signal. On this basis, extract the pulse rate variability characteristics. For the denoised blowing air flow signal, segment it based on the blowing training sound score and extract the blowing breath stability characteristics. For the finger position control data, similarly segment it and extract the finger position control accuracy characteristics based on statistical principles.

[0104] In this embodiment, the time domain, frequency domain, and non-linear characteristics of the pulse rate variability are extracted, and at the same time, referring to common statistical indicators, the statistical characteristics of the blowing air flow signal and finger position control data are extracted. The flow chart of physiological and behavioral feature extraction is as Figure 4 shown, and the feature extraction process is as follows: S301. Extract the pulse rate variability characteristics from the denoised baseline state pulse wave signal, denoised task state pulse wave signal, and denoised recovery state pulse wave signal.

[0105] S301.1. Use the first-order forward difference traversal method to identify the main wave peak points of the pulse wave signal.

[0106] S301.2. Perform a difference operation on the main wave peak point sequence to obtain the time interval sequence between adjacent two pulse wave peaks, that is, the PP interval sequence.

[0107] S301.3. Use a 21-order AR model to estimate the frequency domain power spectrum of the PP interval sequence.

[0108] S301.4. Extract the time domain, frequency domain, and non-linear characteristics of the pulse rate variability (PRV) from the sequence to obtain the baseline pulse rate variability feature vector and the de-baselined task state pulse rate variability feature vector and the eigenvectors of pulse rate variability in the de - baseline - restored state .

[0109] The features of pulse rate variability include 6 time - domain features: PRV_MEANPP, PRV_SDPP, PRV_RMSSD, PRV_PNN20, PRV_PNN50, PRV_TI; 3 frequency - domain features: PRV_LF, PRV_HF, PRV_LF / PRV_HF; and 4 non - linear features: PRV_SampEn, PRV_SD1, PRV_SD2, PRV_SD1 / PRV_SD2.

[0110] Therefore, a total of , and a total of 3 groups of pulse rate variability eigenvectors are extracted from the pulse wave signal. The 3 groups of vectors are distinguished by different prefixes. Taking PRV_MEANPP as an example, BL_PRV_MEANPP is the vector feature, TB_PRV_MEANPP is the vector feature, and RB_PRV_MEANPP is the vector feature.

[0111] S302. For the denoised blowing airflow signal, segment it according to the handwritten score notes of the blowing training sound, calculate the short - time energy for each frame of the signal, and based on statistical principles, extract 3 blowing breath stability features: STE_MEANCV, MEANSampEn, and STE_MEAN_SD from the short - time energy sequence.

[0112] S303. For the fingering control data, segment it according to the handwritten score notes of the blowing training sound, and based on statistical principles, extract 3 fingering control accuracy features: SP_MAE, SP_RMSE, and SP_ACC.

[0113] S4. Based on histogram analysis and goodness - of - fit test, perform probability distribution fitting on the reaction time of the inhibitory control test task, and obtain the test task score by calculating the cumulative probability and combining it with the test task accuracy rate. The specific steps are as follows: S401. For the reaction time of each inhibitory control test task, plot a histogram of the reaction time to analyze the data distribution pattern. Combining the current research on the distribution model of the reaction time, in this embodiment, a log - normal distribution is selected to perform fitting analysis on the reaction - time histogram.

[0114] The probability density function of the log - normal distribution is as follows:

[0115] where is the expectation, is the standard deviation.

[0116] The cumulative probability function of the lognormal distribution is as follows:

[0117] where is the cumulative distribution function of the standard normal distribution.

[0118] S402. Through the A-D goodness-of-fit test, accept the null hypothesis, that is, it is considered that the reaction time of the test task follows the lognormal distribution, and calculate the cumulative probability of the reaction time .

[0119] S403. For the accuracy rate of the test task, since the accuracy rate is already a normalized index, multiply the accuracy rate of each test task by the reaction time of this task to obtain the score of each test task, and add up the scores of each test task to obtain the score of the subject for this test task.

[0120] The calculation expression of the test task score is as follows:

[0121] where , respectively represent the accuracy rate and reaction time of the subject's th test task, and the value range of is .

[0122] S5. Construct a data set based on the physiological and behavioral characteristics of the subject, train a random forest regression model with the test task score obtained in step S4 as the target variable, evaluate the performance of different feature subsets through recursive feature elimination and cross-validation, and select the feature subset with the highest performance.

[0123] The flowchart of establishing a quantitative evaluation model for intervention effect based on physiological and behavioral characteristics is as shown in Figure 5 .

[0124] S501. Construct a data set based on the 42 physiological characteristics of the 3 groups of pulse rate variability feature vectors extracted in step S3 and the 6 behavioral characteristics of the blowing breath stability and the finger position control accuracy.

[0125] S502. Use the test task score obtained in step S4 as the target variable and the physiological and behavioral characteristics as the explanatory variables to train a random forest regression model.

[0126] In this step, considering that there may be coexisting linear and non-linear associations between the physiological and behavioral characteristics and the target variable, directly using a linear regression model such as ridge regression may lead to loss of feature information. Therefore, in this embodiment, a random forest regression model is selected to obtain the predicted importance of each feature for the target variable, providing a reliable basis for subsequent feature screening.

[0127] S503. Using the random forest regression model obtained in step S502 as the base model, recursively eliminate the features with the lowest importance using recursive feature elimination, and calculate the RMSE and R of the feature subset through 10-fold cross-validation 2 , evaluate the performance of the feature subset at each recursion, and finally select the feature subset with the highest R 2 and the lowest RMSE. If there is a conflict between R² and RMSE, preferentially select the feature subset with the highest R².

[0128] S6. Perform Spearman correlation analysis on the optimal feature subset obtained in S5. According to the analysis results, perform Min-Max normalization on the features, transform the directions of all features to be positively correlated with the inhibitory test task score, that is, positively correlated with the inhibitory control ability, retrain the regression model, construct a quantitative evaluation formula based on feature importance, and obtain the quantitative score of "trombone playing training - inhibitory control ability".

[0129] The specific steps are as follows: S601. Perform Spearman correlation analysis on the optimal feature subset obtained in step S5.

[0130] S602. According to the correlation between each feature of the optimal feature subset and the test task score, perform Min-Max normalization on each feature.

[0131] S603. Retrain the regression model, construct a quantitative evaluation formula based on feature importance, and obtain the quantitative score of "trombone playing training - inhibitory control ability".

[0132] Example 2 This example further discloses a quantitative evaluation method for the intervention effect of trombone playing training on inhibitory control ability, including the following steps: S1. Referring to the corresponding step S1 in Example 1, collect the physiological and behavioral data of a subject during the execution of playing training, that is, the pulse wave signals in the baseline, task state, and recovery state, the playing air flow signal, and the fingering control data, and record the correct rate and reaction time of the subject performing the inhibitory control test task at the same time. The original pulse wave signal waveform diagram is as Figure 6 shown, and the playing air flow signal waveform diagram is as Figure 7 shown.

[0133] S2. Referring to the corresponding step S2 in Example 1, obtain the preprocessed physiological and behavioral data. The schematic diagram of the fingering control data sequence in the behavioral data is as Figure 8 shown.

[0134] S3. Referring to the corresponding step S3 in Example 1, extract the physiological and behavioral characteristics of the subject. The recognition result diagram of the feature points of the denoised pulse wave signal is asFigure 9 as shown

[0135] S4. Refer to the corresponding step S4 in Example 1, fit the probability distribution of the reaction time during the fitting reaction, and calculate the inhibitory control test task score in combination with the correct rate . The distribution curve graph of the reaction time of the inconsistent task is as Figure 10 shown

[0136] S5. Refer to the corresponding step S5 in Example 1, use recursive feature elimination with cross-validation to obtain the optimal feature subset with 4 features. The features of the optimal feature subset include PRV_SDPP, PRV_HF, STE_MEAN_SD, and SP_MAE. The RMSE and R of different feature subsets during recursive feature elimination 2 The schematic diagram is as Figure 11 shown

[0137] S6. On the basis of Example 1, the dimensionless methods for each feature are obtained as shown in Table 1: Table 1. Dimensionless method table

[0138] After re-training the random forest regression model with each feature of the optimal feature subset dimensionless, the importance weights of each feature are obtained. The feature importance weights are shown in Table 2: Table 2. Feature importance weight table

[0139] In summary, through weighted summation of the feature importance weights, the quantitative score of "trombone playing training - inhibitory control ability" is obtained. This score explains the influence of physiological and behavioral characteristics on inhibitory control ability during the playing training process. By analyzing the changes in the quantitative score of each playing training, the quantitative evaluation of the intervention effect of trombone playing training on inhibitory control ability is realized

[0140] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification

[0141] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications made without departing from the spirit and principle of the present invention shall be equivalent replacement methods and are all included in the protection scope of the present invention

Claims

1. A quantitative evaluation method for the intervention effect of trombone playing training on inhibitory control ability, characterized in that: The quantitative evaluation method comprises the following steps: S1. Use photoplethysmography to collect the baseline pulse wave signal, task pulse wave signal and recovery pulse wave signal of the subject during the playing training of the AI ​​trombone, collect the blowing airflow signal of the subject during the playing training through the pickup module, record the level signal output by the ultrasonic ranging module, the level signal is the original position control data, and record the accuracy and reaction time of the inhibition control test task; wherein, the AI ​​trombone is an intelligent musical instrument obtained by laboratory transformation of the traditional trombone, which is used for data collection; the state 5 minutes before the start of the playing training is called the baseline state, and the pulse wave signal collected in this state is called the baseline pulse wave signal; the state of performing the playing training is called the task state, and the pulse wave signal collected in the task state is called the task state pulse wave signal; the state 5 minutes after the end of the playing training is called the recovery state, and the pulse wave signal collected in the recovery state is called the recovery state pulse wave signal; S2. For the baseline pulse wave signal, the task pulse wave signal and the recovery pulse wave signal, the variational mode decomposition and wavelet threshold joint denoising method are used to filter out high-frequency noise and reduce the influence of motion artifacts, and the baseline drift of the pulse wave signal is removed by the cubic spline interpolation method to obtain the denoised baseline pulse wave signal, the denoised task pulse wave signal and the denoised recovery pulse wave signal; for the blowing airflow signal, the notch filter and the median filter are used to perform denoising processing to obtain the denoised blowing airflow signal; the original position control data is converted to obtain the position control data represented by the distance sequence; S3, identify the main wave peak points of the denoised baseline pulse wave signal, the denoised task pulse wave signal and the denoised recovery pulse wave signal, and extract the pulse rate variability characteristics on this basis; segment the denoised blowing airflow signal based on the blowing training sound score, extract the blowing breath stability characteristics, and segment the position control data in the same way, and extract the position control accuracy characteristics based on statistical principles; S4. Based on histogram analysis and goodness of fit test, probability distribution fitting is performed on the reaction time of the inhibitory control test task, and the test task score is obtained by calculating the cumulative probability and combining it with the test task accuracy rate; S5. Construct a data set based on the physiological characteristics and behavioral characteristics of the subjects, train a random forest regression model with the test task score obtained in step S4 as the target variable, evaluate the performance of different feature subsets through recursive feature elimination and cross-validation, and screen out the feature subset with the highest performance, that is, the optimal feature subset; wherein the physiological characteristics are the pulse rate variability characteristics extracted from the denoised baseline pulse wave signal, the denoised task pulse wave signal, and the denoised recovery pulse wave signal; the behavioral characteristics are the blowing breath stability and position control accuracy characteristics extracted from the blowing airflow signal and the position control data; S6. Perform correlation analysis on the optimal feature subset obtained in S5, make the features dimensionless based on the analysis results, retrain the random forest regression model, and construct a quantitative evaluation formula based on feature importance to obtain a quantitative score of "trombone playing training-inhibitory control ability". By analyzing the quantitative score, a quantitative evaluation of the intervention effect of trombone playing training on inhibitory control ability can be achieved.

2. The quantitative evaluation method for the intervention effect of trombone playing training on inhibitory control ability according to claim 1, characterized in that: The process of step S1 is as follows: S101, the subject sits in front of a screen, and the screen displays the UI interface of the trombone playing training module. A PPG sensor is placed on the desktop where the screen is located. First, the screen displays a 5-minute countdown, and at the same time prompts the subject to adjust the sensor position and place the middle finger of the left hand on the PPG sensor, sit quietly for 5 minutes, collect the baseline pulse wave signal of the subject before training, and after the countdown ends, the screen prompts to enter the playing training; S102, the screen displays the playing training sound score. During the training, the subject is required to keep the upper body upright, wear headphones, hold the slide of the AI ​​trombone with the right hand, hold the handle of the AI ​​trombone with the left hand, wear the PPG sensor finger sleeve on the middle finger of the left hand, tilt the AI ​​trombone bell downward by 15 degrees, and naturally align the lips with the AI ​​trombone mouthpiece; wherein the PPG sensor finger sleeve is used to collect task-state pulse wave signals; S103, the screen plays the vocal score in the form of animation, and the subject needs to play with a stable breath, and move the slide to the correct position according to the notes and position prompts displayed in the vocal score. The playing training is 10 rounds in total, and a random section of the vocal score is played in each round. The pickup module on the AI ​​trombone is used to collect the blowing airflow signal of the subject when performing the blowing training, and the level signal output by the ultrasonic ranging module, that is, the original position control data, is recorded; S104, after the playing training is finished, the screen prompts the subject to take off the finger sleeve and place the left middle finger on the desktop PPG sensor, displays a 5-minute countdown, and prompts the subject to sit quietly for 5 minutes to collect the subject's recovery state pulse wave signal; S105, the subject sits in front of a display screen, which displays a UI interface of an inhibitory control test task module. The inhibitory control test task consists of a Stroop test task and a variant Flanker test task. The screen first prompts the subject to perform the Stroop test task, and records the accuracy and reaction time of each test task during the execution; S106. After the subject has completed all Stroop test tasks, the display screen prompts the subject to perform a variant Flanker test task, and records the accuracy and reaction time of each test task during the execution.

3. The quantitative evaluation method for the intervention effect of trombone playing training on inhibitory control ability according to claim 1, characterized in that: The process of step S2 is as follows: S201, pre-processing the pulse wave signal collected in step S1, the specific steps are as follows: S201.

1. Decompose the baseline pulse wave signal, the task pulse wave signal and the recovery pulse wave signal using variational mode decomposition to obtain multiple intrinsic mode function components, hereinafter referred to as intrinsic mode function IMF, perform fast Fourier transform on each IMF component and calculate the frequency corresponding to the maximum peak in the spectrum diagram ; S201.2, after The high-frequency IMF components are denoised using the wavelet threshold method to filter out high-frequency noise and reduce the impact of motion artifacts; S201.3, superimposing the high-frequency IMF component processed in step S201.2 with the remaining IMF component to reconstruct the pulse wave signal; S201.4, filtering the reconstructed pulse wave signal using a Butterworth bandpass filter to further improve the pulse wave signal quality and signal-to-noise ratio; S201.

5. Extract the starting point of each pulse wave cycle in the pulse wave signal to obtain a pulse wave cycle starting point sequence; S201.

6. Performing cubic spline interpolation on the pulse wave cycle starting point sequence to fit the pulse wave baseline, remove the baseline drift of the pulse wave signal, and obtain a denoised baseline state pulse wave signal, a denoised task state pulse wave signal, and a denoised recovery state pulse wave signal; S202, using a notch filter and a median filter to perform denoising on the blowing airflow signal collected in step S1; S203 , for the original position control data obtained in step S1 , convert the level signal into distance according to the sound speed and the duration of the high level, and obtain the position control data represented by the distance sequence.

4. The quantitative evaluation method for the intervention effect of trombone playing training on inhibitory control ability according to claim 1, characterized in that: The process of step S3 is as follows: S301, for the denoised baseline pulse wave signal, the denoised task pulse wave signal and the denoised recovery pulse wave signal, use the first-order forward difference traversal method to identify the main wave peak point of the pulse wave signal, obtain the time interval sequence between two adjacent pulse wave peaks based on the main wave peak point sequence, and extract the pulse rate variability feature from the time interval sequence; S302, the denoised playing airflow signal obtained in step S2 is segmented based on the playing training sound score, each signal segment is framed using a sliding window algorithm, a short-time energy sequence is calculated, and a playing breath stability feature is extracted from the sequence; S303, segmenting the position control data obtained in step S2 based on the playing training sound score measures, and extracting position control accuracy features based on statistical principles.

5. The quantitative evaluation method for the intervention effect of trombone playing training on inhibitory control ability according to claim 4 is characterized in that: The process of step S4 is as follows: S401, for each inhibitory control test task, draw a reaction time histogram to analyze the data distribution; S402. Through the Anderson-Darling goodness-of-fit test, the target probability distribution model that accepts the null hypothesis is selected as the probability distribution of the reaction time, and the cumulative probability of the reaction time is obtained. ; Wherein, the null hypothesis is that the reaction time obeys the target probability distribution; S403: For the accuracy of the test task, since the accuracy is already a normalized indicator, the accuracy of each test task and the reaction time of the task are calculated. Multiply them together to get the score of each test task, and add up the scores of each test task to get the subject's score for this test task.

6. The quantitative evaluation method for the intervention effect of trombone playing training on inhibitory control ability according to claim 5 is characterized in that: The process of step S5 is as follows: S501, constructing a data set based on the pulse rate variability feature, blowing breath stability feature and position control accuracy feature extracted in step S3; S502, training a random forest regression model using the test task score obtained in step S4 as the target variable and the physiological and behavioral characteristics as the explanatory variables; S503: Using the random forest regression model obtained in step S502 as the base model, recursive feature elimination is used to gradually eliminate the features with the lowest importance, and the root mean square error RMSE and determination coefficient R of the feature subset are calculated through cross-validation. 2 , evaluate the performance of the feature subset at each recursion, and finally select R 2 The feature subset with the highest R² and the lowest RMSE. If there is a conflict between R² and RMSE, the feature subset with the highest R² is preferred.

7. The quantitative evaluation method for the intervention effect of trombone playing training on inhibitory control ability according to claim 6, characterized in that: The process of step S6 is as follows: S601, performing Spearman correlation analysis on the optimal feature subset obtained in step S5 to obtain the positive and negative correlation between each feature in the subset and the test task score; S602, based on the positive and negative correlations obtained in step S601, performing Min-Max standardization on each feature; S603. Retrain the random forest regression model, construct a quantitative evaluation formula based on feature importance, and obtain a quantitative score for "trombone playing training-inhibition control ability".

Images