Attention and execution function evaluation and training system and method based on virtual reality

Abstract

The invention discloses an attention and execution function evaluation and training system based on virtual reality. The system comprises an evaluation model and a cognitive ability training model. The evaluation model comprises a user registration module, a parameter adjustment module and a personalized evaluation module; the cognitive ability training model comprises a self-adaptive ability training module, a data management module and a pass-through and pass-through reward module. A pass-through and pass-through reward module is arranged in the cognitive ability training model, so that the user can be stimulated to participate in training for a long time; the evaluation report is combined with the adaptive algorithm in the adaptive ability training module, and according to the real-time performance of the user in the virtual reality technology, the task training difficulty and the task content can be dynamically adjusted according to the individual difference and the real-time performance, and the effectiveness and challenge of the training effect can be ensured.

Description

Technical Field

[0001] The present invention relates to an evaluation and training method, and in particular to an attention and executive function evaluation and training system and method based on virtual reality. Background Art

[0002] Attention and executive function are key components of the human cognitive process and have a vital impact on learning, work and life. Among them, attention refers to an individual's ability to focus on and concentrate on specific things, and executive function involves high-level cognitive processes such as planning, organization, abstract thinking, and working memory. In real life, attention and executive function will decline over time and other factors, so targeted intervention is needed to promote the improvement of cognitive function and provide support for individual learning and life.

[0003] However, traditional evaluation and training methods have the following shortcomings:

[0004] First, the ecological validity is low, that is, the existing assessment methods are single-scenario and cannot fully and truly reflect the cognitive performance of individuals in real life;

[0005] Second, the assessment and training process is boring, lacks interactivity and situational authenticity, lacks effective incentive mechanisms, and is difficult to maintain long-term participation;

[0006] 3. Most of the existing assessment tasks are fixed, using fixed difficulty levels or simple rules to adjust task difficulty. The task training difficulty and task content are not dynamically adjusted according to individual differences and real-time performance, making it difficult to ensure the effectiveness and challenge of the training content.

[0007] 4. The lack of real-time data support makes it difficult to quantify the effects and adjust the training plan based on performance. Summary of the invention

[0008] In order to solve the deficiencies of the above technologies, the present invention provides a virtual reality-based attention and executive function assessment and training system and method.

[0009] In order to solve the above technical problems, the technical solution adopted by the present invention is: the attention and executive function evaluation and training system based on virtual reality includes an evaluation model and a cognitive ability training model;

[0010] The evaluation model includes a user registration module, which is used for users to register personal accounts and establish user profiles; a parameter adjustment module, which is used to provide an interface tool to allow users to adjust the difficulty parameters of the tasks used for evaluation before the formal evaluation to achieve a personalized testing experience; a personalized evaluation module, which is used to evaluate the performance of users' attention and executive abilities according to their cognitive ability differences;

[0011] The cognitive ability training model includes an adaptive ability training module, which is used to dynamically adjust the difficulty parameters of the task based on the Q-learning algorithm to achieve continuous intervention in user capabilities; a data management module, which is used to record and manage user behavior, objective data and data of running scripts, upload the data to the data platform for standardized processing, and integrate the data to generate a training result report; a level-breaking and level-clearing reward module, which is used to enhance the user's sense of accomplishment and motivate users to participate in training.

[0012] The virtual reality-based attention and executive function assessment and training method specifically includes the following steps:

[0013] Step S1: Based on the C# programming language and the Unity3D game engine, combined with 3D modeling software, several virtual scenes are constructed, including daily life scenes and game scenes;

[0014] Step S2: construct a dynamic adjustment model to dynamically adjust the difficulty parameters of the task based on the real-time monitored behavioral data and the user's age and cognitive ability level; the difficulty parameters of the task include stimulus presentation time T_s, stimulus number N_s and reaction time limit T_r;

[0015] Step S3: Execute the task, evaluate the execution of the user's task, and generate a visual evaluation report;

[0016] Step S4: The evaluation report is combined with an adaptive algorithm to automatically adjust the current training difficulty and content of a personalized learning method based on the Q-learning algorithm according to the user's real-time performance in the virtual reality technology, and to conduct training;

[0017] Step S5: Build a full-process closed-loop system by integrating virtual reality technology head-mounted devices, handles, tablet devices and website background.

[0018] Furthermore, the formula for dynamically adjusting the model in step S2 is:

[0019] P = f(C, A, R);

[0020] Among them, P is the difficulty parameter vector of the current task, which includes T_s, N_s and T_r; C is the user's age or cognitive ability level; A is the accuracy of the current task completion; R is the reaction time;

[0021] The rule for adjusting the stimulus presentation time T_s is:

[0022] T_s=T s0 -α*A+β*max(0,T_r-R);

[0023] Among them, T s0is the initial stimulus presentation time; α is the weight of accuracy on stimulus presentation time; β is the weight of reaction time on stimulus presentation time; R is reaction time;

[0024] The rule for adjusting the number of stimuli N_s is:

[0025] N_s=N s0 +γ*A;

[0026] Among them, N s0 is the initial number of stimuli; γ is the growth factor of accuracy to the number of stimuli; A is the accuracy of the current task;

[0027] The rule for adjusting the reaction time limit T_r is:

[0028] T_r=T s0 -δ*A;

[0029] Among them, T s0 is the initial reaction time limit; δ is the influence coefficient of accuracy on the reaction time limit; A is the accuracy of the current task.

[0030] Furthermore, dynamically adjusting the difficulty parameter of the task in step S2 specifically includes the following steps:

[0031] Step S21: Initialize the difficulty parameters of the task and initialize the weights of each difficulty parameter;

[0032] Step S22: The user performs the task. After the task is completed, the accuracy A and reaction time R of the user performing the task are recorded;

[0033] Step S23: updating the difficulty parameter of the task based on the dynamic adjustment model;

[0034] Step S24: Generate a new task according to the difficulty parameter value of the updated task;

[0035] Step S25: execute the new task and determine whether the accuracy of the new task is greater than 65% and less than 85%; if so, output the user's final task performance data; if the accuracy of the new task is not greater than 65%, reduce the weight α of the accuracy to the stimulus presentation time and the growth factor γ of the accuracy to the number of stimuli; if the accuracy of the new task is not less than 85%, increase the weight α of the accuracy to the presentation time and the growth factor γ of the accuracy to the number of stimuli.

[0036] Furthermore, the tasks in step S2 are cognitive tasks in the field of attention and executive function, and the cognitive tasks include sustained attention test, selective attention test, working memory test, response inhibition test, cognitive flexibility test and delayed gratification test.

[0037] Furthermore, in step S3, the execution of the user task is evaluated, which specifically includes the following steps:

[0038] Step S31: collect objective data and subjective data respectively, and transmit both objective data and subjective data to the backend server; objective data includes accuracy, reaction time and task completion; subjective data is the subjective report data of the user in the questionnaire;

[0039] Step S32: Standardize the objective data and subjective data to ensure that all data are integrated under the same dimension; normalize the subjective data, and the normalization formula is:

[0040] normalized_score=(score-min_score) / (max_score-score);

[0041] Among them, score is the initial score; min_score is the minimum value in the score; max_score is the maximum value in the score;

[0042] Step S33: construct a hybrid model for multi-dimensional fusion, which is used to combine objective data with subjective data to form a comprehensive score S; the calculation formula of the comprehensive score S is:

[0043] S=0.4×ZR+0.5×ZA+0.1×normalized_score+b;

[0044] Among them, ZR is the Z score of reaction time; ZA is the Z score of accuracy; b is the bias term, which is a constant;

[0045] Step S34: Generate a visual evaluation report based on the results of multi-dimensional data fusion; the evaluation report includes scores of various cognitive abilities, comprehensive evaluations, and improvement suggestions.

[0046] Furthermore, the formula of the Q-learning algorithm in step S4 is:

[0047] Q(S,A1)=Q(S,A1)+α[R+Υmax A1 Q(S′,A1′)-Q(S,A1)];

[0048] Among them, Q(S,A1) is the Q value of taking action A1 in state S; R is the immediate reward obtained after taking the action in the current state; max A Q(S',A1') is the maximum Q value obtained by taking the optimal action A1' in the next state S'; α is the learning rate; Υ is the discount factor.

[0049] Furthermore, the calculation formula for the instant reward is:

[0050] R′=c 1 *(1-reaction time / maximum time limit)+c 2 *Accuracy;

[0051] Among them, c 1 is the weight coefficient of reaction time; c 2 is the weight coefficient of accuracy.

[0052] Furthermore, the personalized learning method for automatically adjusting the current training difficulty and content based on the Q-learning algorithm in step S4 specifically includes the following steps:

[0053] Step S41: setting the difficulty parameter of the initial task according to the evaluation result and determining the initial training level;

[0054] Step S42: generating an initial task based on the initial training level and feeding back to the user;

[0055] Step S43: collect objective data of the user in the task through the virtual reality technology head mounted device and the handle, and normalize the objective data into a state vector as the current state S;

[0056] Step S44: Create a Q value table, and select an action A1 from the Q value table according to the current state S;

[0057] Step S45: updating the Q value table based on the Q-learning algorithm;

[0058] Step S46: Setting the difficulty adjustment step length δ to avoid excessive adjustment of the task difficulty, which may cause discontinuous changes in the task difficulty;

[0059] Step S47: Collect objective data and subjective data in real time through the virtual reality technology head-mounted device and handle, and update the Q value table until the Q value table converges.

[0060] Furthermore, the formula for setting the difficulty adjustment step length δ in step S46 is:

[0061] δ=max(0.1,1 / sqrt);

[0062] Where sqrt is the number of completed tasks.

[0063] The present invention discloses a virtual reality-based attention and executive function assessment and training system and method, which has the following beneficial effects:

[0064] 5. By setting up several virtual scenes, the scenarios of the evaluation method are diversified, and virtual reality technology is used to more realistically reflect the cognitive performance of individuals in real life;

[0065] 6. By setting up a level-breaking and level-clearing reward module in the cognitive ability training model, users can be motivated to participate in training for a long time;

[0066] 7. By combining the evaluation report with an adaptive algorithm and based on the user's real-time performance in virtual reality technology, the difficulty and content of task training can be dynamically adjusted according to individual differences and real-time performance, ensuring the effectiveness and challenge of the training effect;

[0067] 8. By collecting objective data in real time through virtual reality headsets and controllers, training plans can be adjusted more accurately based on user performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 The flowchart of constructing a virtual scene in an embodiment of the present invention.

[0069] Figure 2 The following is an operational flow chart of the evaluation model in the present invention.

[0070] Figure 3 It is a block diagram of the evaluation and training system in the present invention.

[0071] Figure 4 This is the operating procedure of the cognitive training model in the present invention. DETAILED DESCRIPTION

[0072] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0073] like Figures 1 to 4 The virtual reality-based attention and executive function assessment and training system shown includes an assessment model and a cognitive ability training model;

[0074] The evaluation model includes a user registration module, which is used for users to register personal accounts and establish user profiles; a parameter adjustment module, which is used to provide an interface tool to allow users to adjust the difficulty parameters of the tasks used for evaluation before the formal evaluation to achieve a personalized testing experience; a personalized evaluation module, which is used to evaluate the performance of users' attention and executive abilities according to their cognitive ability differences;

[0075] The cognitive ability training model includes an adaptive ability training module, which is used to dynamically adjust the difficulty parameters of the task based on the Q-learning algorithm to achieve continuous intervention in user capabilities; a data management module, which is used to record and manage user behavior, objective data and data of running scripts, upload the data to the data platform for standardized processing, and integrate the data to generate a training result report; a level-breaking and level-clearing reward module, which is used to enhance the user's sense of accomplishment and motivate users to participate in training.

[0076] The virtual reality-based attention and executive function assessment and training method specifically includes the following steps:

[0077] Step S1: Based on the C# programming language and the Unity3D game engine, combined with 3D modeling software, several virtual scenes are constructed, including daily life scenes and game scenes;

[0078] In this embodiment, the daily life scenes include bedroom scenes, bookshelf scenes, gymnasium scenes, museum scenes and road scenes; the game scenes include shooting range scenes, spaceship battle scenes, whack-a-mole scenes, cartoon forest scenes and sandbox scenes.

[0079] The tree-structured scene graph manages the elements in the virtual scene, including objects, materials, lighting, and interactions. Each node in the tree structure represents an element and contains various components. For example, a toy node includes a Transform position component, a Mesh model component, a Material component, a Collider component, and a Rigidbody component.

[0080] The physics engine in the Unity3D game engine simulates the physical laws of the real world, and uses rigid body components to give objects physical properties; collisions between objects are detected through collision body components; and scripts are combined to achieve interactive functions, including grabbing and throwing.

[0081] The audio system in the Unity3D game engine provides the sound of the environment in the virtual scene, and special effects are produced based on the particle system; for example, special effects such as shooting and explosions are provided in the game scene to enhance authenticity and user immersion.

[0082] Adding several virtual scenarios can not only truly reflect the users’ cognitive performance in daily life, but also increase the fun and participation of the assessment through gamification scenarios, thus ensuring the comprehensiveness and authenticity of the assessment and avoiding the limitations and biases brought by a single scenario.

[0083] Step S2: construct a dynamic adjustment model to dynamically adjust the difficulty parameters of the task based on the real-time monitored behavioral data and the user's age and cognitive ability level; the difficulty parameters of the task include stimulus presentation time T_s, stimulus number N_s and reaction time limit T_r;

[0084] The stimulus presentation time is the duration of the target stimulus on the display. Shorter stimulus presentation times are used for individuals with high cognitive abilities, and longer stimulus presentation times are used for individuals with low cognitive abilities or children.

[0085] The number of stimuli refers to the number of stimuli presented simultaneously in the adjustment task scenario, that is, the total number of stimuli that the user needs to recognize and respond to.

[0086] Reaction time limits are set to set the reaction time window allowed for users to complete a task; for example, the time limit for task completion will gradually shorten as performance improves.

[0087] The formula for the dynamic adjustment model is:

[0088] P = f(C, A, R);

[0089] Among them, P is the difficulty parameter vector of the current task, which includes T_s, N_s and T_r; C is the user's age or cognitive ability level; A is the accuracy of the current task completion; R is the reaction time.

[0090] The rule for adjusting the stimulus presentation time T_s is:

[0091] T_s=T s0 -α*A+β*max(0,T_r-R);

[0092] Among them, T s0 is the initial stimulus presentation time; α is the weight of accuracy to stimulus presentation time; β is the weight of reaction time to stimulus presentation time; R is the reaction time.

[0093] In this embodiment, the initial stimulus presentation time is set according to the age of the user; for a 5-year-old child, the initial stimulus presentation time is set to 1500ms; for a 6-year-old child, the initial stimulus presentation time is set to 1350ms; for a 7-year-old child, the initial stimulus presentation time is set to 1200ms; for an 8-year-old child, the initial stimulus presentation time is set to 1050ms; for a 9-year-old child, the initial stimulus presentation time is set to 900ms; for a 10-year-old child, the initial stimulus presentation time is set to 750ms; for a user over 11 years old, the initial stimulus presentation time is set to 500ms;

[0094] The initial value of the weight α of accuracy to stimulus presentation time is 0.1, which means that when the accuracy rate increases by 10% each time, the stimulus presentation time decreases by 10%;

[0095] The initial value of the weight β of reaction time to stimulus presentation time is 0.2; if the user's actual reaction time exceeds the reaction time limit, β is increased by 0.05 each time to ensure that the user has more reaction time; if the user's actual reaction time is improved, β is reduced by 0.05 each time.

[0096] The rule for adjusting the number of stimuli N_s is:

[0097] N_s=N s0 +γ*A;

[0098] Among them, N s0is the initial number of stimuli; γ is the growth factor of accuracy to the number of stimuli; A is the accuracy of the current task;

[0099] The initial stimulation quantity of this embodiment increases by 5% for every 1 year increase in age;

[0100] The initial value of the growth factor γ of the accuracy rate to the number of stimuli is 0.5; if the accuracy rate is lower than 65%, the γ value is reduced, that is, the γ value is reduced by 0.05 to slow down the growth rate of the number of stimuli; if the accuracy rate is higher than 85%, the γ value is increased, that is, the γ value is increased by 0.05 to increase the speed of the increase of the number of stimuli.

[0101] The rule for adjusting the reaction time limit T_r is:

[0102] T_r=T s0 -δ*A;

[0103] Among them, T s0 is the initial reaction time limit; δ is the influence coefficient of accuracy on reaction time limit; A is the accuracy of the task currently completed;

[0104] In this embodiment, the initial reaction time limit T s0 , set according to age groups; for 5-year-old children, the initial response time is set to 2000ms; for 6-year-old children, the initial response time is set to 1800ms; for 7-year-old children, the initial response time is set to 1600ms; for 8-year-old children, the initial response time is set to 1400ms; for 9-year-old children, the initial response time is set to 1200ms; for 10-year-old children, the initial response time is set to 1000ms; for users over 11 years old, the initial response time is set to 800ms;

[0105] The initial value of the coefficient δ of the influence of accuracy on the reaction time limit is 0.1. If the reaction time decreases faster, the δ value can be reduced, that is, the δ value is reduced by 0.05.

[0106] The step S2 dynamically adjusts the difficulty parameter of the task, specifically including the following steps:

[0107] Step S21: Initialize the difficulty parameter of the task; set the initial value P0 of the difficulty parameter of the task, and initialize the weight of each difficulty parameter; in this embodiment, each weight parameter is initially set, wherein the initial value of the weight α of the accuracy rate to the stimulus presentation time is 0.1, the initial value of the weight β of the reaction time to the stimulus presentation time is 0.2, the initial value of the growth factor γ of the accuracy rate to the number of stimuli is 0.5, and the initial value of the influence coefficient δ of the accuracy rate on the reaction time limit is 0.1;

[0108] Step S22: The user performs the task. After the task is completed, the accuracy A and reaction time R of the user performing the task are recorded;

[0109] Step S23: updating the difficulty parameter of the task based on the dynamic adjustment model;

[0110] Step S24: generating a new task according to the difficulty parameter value of the updated task;

[0111] Step S25: execute the new task and determine whether the accuracy of the new task is greater than 65% and less than 85%; if so, output the user's final task performance data; if the accuracy A is not greater than 65% after the task is completed, reduce the weight α of the accuracy to the stimulus presentation time and the value of the growth factor γ of the accuracy to the number of stimuli, for example, reduce the value of α to 0.05, and reduce the value of γ to 0.4; if the accuracy is not less than 85% after the task is completed, increase the weight α of the accuracy to the stimulus presentation time and the value of the growth factor γ of the accuracy to the number of stimuli, for example, increase the value of α to 0.2, and increase the value of γ to 0.6.

[0112] By real-time monitoring of task performance, the difficulty parameters of the task are adjusted in real time, and then the task content is dynamically adjusted to adapt to users of different ages and cognitive ability levels, avoiding evaluation errors caused by fixed tasks; the difficulty parameters of the task are adjusted based on feedback from user behavior data to ensure that the evaluation challenge is moderate and avoid excessive or insufficient adjustment.

[0113] The task in step S2 of this embodiment is to set cognitive tasks in the field of attention and executive function, and the cognitive tasks include sustained attention test, selective attention test, working memory test, response inhibition test, cognitive flexibility test and delayed gratification test;

[0114] In this embodiment, a sustained attention test is set in the bedroom scene and the shooting range scene. The sustained attention test means that the user is placed in a structured bedroom environment, surrounded by various household items and background sounds; within a fixed time, the user needs to quickly identify and select a specific target, that is, hear or see numbers related to the task, and respond by pressing keys or clicking the handle. The time intervals between the numbers appearing are irregular to test the user's ability to maintain concentration for a long time; during this process, the user operates through the handle, and the system records the user's accuracy and reaction time.

[0115] In this embodiment, a selective attention test is set in the bookshelf scene and the spaceship battle scene. The selective attention test means that the user enters a quiet study environment where the surrounding bookshelves are filled with books of different colors and sizes. Among a pile of books, the user needs to distinguish and respond to a specific stimulus, that is, only respond to books of a specific color, or a specific size, or a specific content. The user needs to quickly find the books that match the target color and respond by clicking or selecting. This environment encourages the user to filter out irrelevant information in a complex background and make quick responses.

[0116] In this embodiment, a working memory test is set in the gymnasium scene and the whack-a-mole scene. The working memory test means that the user is in a bright indoor sports room, facing two large baskets and a table, and balls of different colors or shapes constantly appear on the table. The user needs to remember the color or shape of the current ball. If the current ball has the same color or shape as the previous ball, the user should put the ball in the left basket; if the current ball has a different color or shape from the previous ball, the user should put the ball in the right basket. This test exercises the user's ability to continuously update and improve memory.

[0117] In this embodiment, a cognitive flexibility test is set in the road scene and the sand table scene. The cognitive flexibility test means that the user is in the production workshop of a currency factory, facing a complex banknote printing machine, and there are switching prompts with different rules posted on the wall. The user needs to select and obtain currency according to different rules. For example, the user needs to select according to the overall size and shape, or according to the local small shape. The user needs to quickly adapt and make the correct choice. This can test the user's ability to flexibly switch thinking between different rules.

[0118] In this embodiment, a response inhibition test is set in the museum scene and the cartoon forest scene. The response inhibition test means that the user strolls in a large museum, and in front of the queue there are various animal exhibits, animal models of different sizes. The user needs to distinguish animal models of different sizes and make responses according to the task requirements. The user needs to quickly inhibit irrelevant responses in this environment to ensure the accuracy of the task.

[0119] In this embodiment, a delay of gratification test is also set in the cartoon forest scene. The delay of gratification test means that in the cloning factory environment, the user needs to make a choice between an immediate reward in front and a delayed reward behind, so as to test the user's choice ability between immediate gratification and delayed gratification.

[0120] All cognitive tasks are a combination of daily life scenarios and game scenarios. In daily life scenarios, the life scenarios that users usually encounter in daily life are restored as much as possible to make users feel as if they are there; these scenarios not only truly reflect the situations in daily life, but are also combined with specific task paradigms, so that they can more accurately evaluate the user's ability performance; in game scenarios, they are mainly designed as virtual games or cartoon scenes, adding more fresh and exciting and more interesting interactive methods, and introducing reward and punishment mechanisms, including point rewards and penalties, to encourage users to better participate in the evaluation. Specific tasks are combined with specific scenario logic to enable users to experience and perform evaluation tasks in real life or gamified environments.

[0121] Step S3: Execute the task, evaluate the execution of the user's task, and generate a visual evaluation report;

[0122] In step S3, the execution of the user task is evaluated, which specifically includes the following steps:

[0123] Step S31: collect objective data and subjective data respectively; transmit both objective data and subjective data to a backend server;

[0124] The user's task performance data, i.e., objective data, is collected in real time through a virtual reality headset and a controller. The objective data includes accuracy, reaction time, and task completion. The user's subjective report data, i.e., subjective data, is collected through a questionnaire scale. In this embodiment, the subjective data includes the questionnaire results filled out by the user, i.e., the child's parent or teacher.

[0125] Objective data is recorded through the script of Unity3D game engine and transmitted to the back-end server; subjective data is uploaded to the back-end server by manually entering it through the electronic questionnaire system. Unity3D game engine includes scene design, scene graph management, physics engine simulation and sound effect production, providing a complete set of software solutions for creating, operating and realizing any real-time interactive content.

[0126] Step S32: Standardize the objective data and subjective data respectively to ensure that all data are integrated under the same dimension; convert the objective data into Z scores, and normalize the subjective data, i.e., the scores of the questionnaire scale; the specific formula for normalizing the subjective data is:

[0127] normalized_score=(score-min_score) / (max_score-score);

[0128] Among them, score is the initial score; min_score is the minimum value in the score; max_score is the maximum value in the score;

[0129] Step S33: construct a hybrid model for multi-dimensional data fusion, which is used to combine objective data with subjective data to form a unified comprehensive score S; the calculation formula of the comprehensive score S is:

[0130] S=0.4×ZR+0.5×ZA+0.1×normalized_score+b;

[0131] Among them, ZR is the Z score of reaction time; ZA is the Z score of accuracy; b is the bias term, which is a constant;

[0132] Step S34: Based on the results of multi-dimensional data fusion, the evaluation and training system generates a visual evaluation report; the evaluation report includes scores of various cognitive abilities, comprehensive evaluations, and improvement suggestions.

[0133] By combining objective and subjective data, the evaluation results are more comprehensive and accurate, and can effectively avoid the bias of a single data source. Intuitive visual reports can help quickly understand the user's cognitive status and provide a scientific basis for subsequent training interventions.

[0134] Step S4: The evaluation report is combined with an adaptive algorithm to automatically adjust the current training difficulty and content of the personalized learning method based on the Q-learning algorithm according to the user's real-time performance in the virtual reality technology, so as to ensure that the training content is both appropriately challenging and has a maximized learning effect;

[0135] The formula for the Q-learning algorithm is:

[0136] Q(S,A1)=Q(S,A1)+α[R′+Υmax A1 Q(S′,A1′)-Q(S,A1)];

[0137] Among them, Q(S,A1) is the Q value of taking action A1 in state S; R′ is the immediate reward obtained after taking the action in the current state; max A Q(S',A1') is the maximum Q value obtained by taking the optimal action A1' in the next state S'; α is the learning rate; Υ is the discount factor.

[0138] The discount factor is used in reinforcement learning to measure the relative importance of current rewards and future rewards. The value range is 0 to 1. The closer the discount factor is to 1, the more the agent values ​​future rewards. The closer the discount factor is to 0, the more the agent values ​​current rewards. The discount factor balances the relationship between immediate returns and possible future returns.

[0139] This embodiment creates a Q value table, with an initial value set to 0 or a random value; the dimension of the Q value table depends on the number of states S and actions A1; the system observes the user's current state S, which includes the average reaction time, the average accuracy, and the current task difficulty coefficient, and represents the current state S as a vector, namely S[reaction time, accuracy, task difficulty];

[0140] The system selects an action A1 from the Q-value table according to the current state S. Action A1 includes increasing the task difficulty, reducing the task difficulty, or keeping the task difficulty unchanged. The state S is defined as [+1, -1, 0].

[0141] Among them, the ∈-greedy strategy is used to balance the difficulty of exploring the optimal training; the formula of the ∈-greedy strategy is:

[0142]

[0143] Among them, ∈ is the exploration probability, and its initial value is 0.9.

[0144] During the training process, the exploration probability ∈ gradually decreases. In this embodiment, it decreases by 0.02 every 10 trainings until the exploration probability ∈ reaches 0.1.

[0145] A personalized learning method based on the Q-learning algorithm that automatically adjusts the current training difficulty and content, specifically including the following steps:

[0146] Step S41: setting the difficulty parameter of the initial task according to the evaluation result and determining the initial training level;

[0147] Step S42: generating an initial task based on the initial training level and feeding back to the user;

[0148] Step S43: collecting the user's performance data in the task, i.e., objective data, through the head-mounted device and handle of the virtual reality technology; and normalizing the objective data into a state vector as the current state S;

[0149] Step S44: Select an action A1 from the Q value table according to the current state;

[0150] Calculate the instant reward based on the user's performance in the current task; the calculation formula for the instant reward is:

[0151] R′=c 1 *(1-reaction time / maximum time limit)+c 2 *Accuracy;

[0152] Among them, c 1 is the weight coefficient of reaction time; c 2 is the weight coefficient of accuracy;

[0153] In this embodiment, when the number of training times is less than or equal to 20 times, c 1 Set to 0.3; c 2 Set to 0.7; when the number of training times is greater than 20, c 1 Increase to 0.5, c 2 Reduced to 0.5 to balance speed and accuracy;

[0154] Step S45: updating the Q value table based on the Q-learning algorithm;

[0155] The formula for the Q-learning algorithm is:

[0156] Q(S,A1)←Q(S,A1)+α[R′+Υmax A1 Q(S′,A1′)-Q(S,A1)];

[0157] Among them, α is the learning rate; Υ is the discount factor;

[0158] In this embodiment, the initial value of the learning rate α is 0.5, and the learning rate α is reduced by 0.01 every 10 training times; the initial value of the discount factor Υ is 0.95. If the user fails in a task twice in a row, the value of Υ can be reduced to 0.9 to increase the emphasis on immediate rewards.

[0159] Step S46: setting the difficulty adjustment step length δ to avoid excessive adjustment of the task difficulty, which would cause discontinuous changes in the task difficulty. That is, each time the task difficulty is adjusted, the change in the task difficulty parameter must not exceed δ.

[0160] Among them, δ can change dynamically according to the number of completed tasks or the number of times the Q value table is updated. In this embodiment, δ=max(0.1,1 / sqrt); sqrt is the number of completed tasks; in the early stage of training, the difficulty adjustment range can be large to quickly find the appropriate difficulty level; in the later stage of training, the adjustment range is gradually reduced to maintain the relative stability of the difficulty.

[0161] Step S47: collect objective data and subjective data in real time, and update the Q value table until the Q value table converges or reaches the set training time and target value.

[0162] The Q-learning algorithm ensures that the training content is always in the "zone of proximal development" of the subject's ability, ensuring that the training will neither make the user feel overly difficult nor be too simple to reduce the training effect; if the user's accuracy rate continues to be lower than the target value, the system will gradually reduce the number of stimuli or extend the reaction time limit; if the user's accuracy rate continues to be higher than the target value, the system will increase the difficulty of the task. The Q-learning algorithm is adjusted based on the user's real-time data, and the system can find the optimal task difficulty setting in a relatively short period of time to help users make the greatest progress in attention and executive function.

[0163] Step S5: By integrating the virtual reality technology head-mounted device, handle, tablet device and website background, a full-process closed-loop system is constructed to generate and manage the entire process from evaluation to training;

[0164] Build a full-process closed-loop system, which includes a user registration module for users to register personal accounts, generate personal accounts, establish a user profile system based on account records and manage the test subject's evaluation data;

[0165] The parameter adjustment module is used to provide an interface tool that allows users to select or adjust assessment parameters before the formal assessment. Users can customize the task difficulty parameters in the task to achieve a personalized testing experience, including adjusting the stimulus presentation time, the number of stimuli, and the reaction time limit.

[0166] Personalized assessment module, used to adjust parameters according to user ability differences to assess the user's attention and executive function performance ability; executive function performance ability includes sustained attention ability, selective attention ability, working memory ability, response inhibition ability, cognitive flexibility ability, delayed gratification ability and emotional decision-making ability;

[0167] Adaptive ability training module, which is used to dynamically adjust task difficulty parameters based on Q-learning algorithm, i.e. reinforcement learning algorithm, to achieve continuous intervention on user ability;

[0168] The data management module is used to record and manage user behavior, cognitive performance data, and system operation script data. After uploading the data to the data platform, it is standardized, the data format is uniformly converted, the data is integrated and the result report is generated, that is, errors and redundancies in the data are eliminated, and formatted behavior data is generated; the website backend can export task reports and support data export and sharing.

[0169] Virtual reality head-mounted devices are user-side virtual reality head-mounted displays that use high-resolution displays and built-in motion tracking systems, with high-precision eye tracking and head tracking capabilities. VR APP is an internal virtual reality application developed in the Unity3D game engine, responsible for generating and rendering realistic virtual scenes and recording user operations and behavior data.

[0170] The handle is equipped with multiple sensors and buttons, and supports multiple gesture recognition functions, such as click, drag, and grab. The handle is provided with a left handle and a right handle.

[0171] The tablet device is the control terminal, equipped with a high-performance processor and a high-definition display screen, supports touch operation, and has WiFi or Bluetooth communication capabilities. The real-time screen of the content of the virtual reality head-mounted display is projected onto the tablet device, and then the tablet device is used to create files, monitor or adjust parameters for the virtual reality head-mounted display. The internal Table APP is a tablet application developed on a mobile operating system that supports functions such as user registration, parameter adjustment, and task operation. The tablet device also has a built-in storage medium for storing virtual scenes, evaluation tasks, user data, and system codes to ensure data security and integrity.

[0172] Website backend: Developed based on Web technology, it can support remote data processing and analysis, and provide secure data storage and backup. The data sources of the website backend are the user end and the control end.

[0173] Through the integration of multiple devices, the system has realized a closed loop of the entire process from evaluation, reporting, training, and management, which has improved the convenience and objectivity of evaluation and training. The combination of the user registration module, parameter adjustment module, personalized evaluation module, adaptive ability training module, and data management module can facilitate the management of user evaluation and training data, generate evaluation reports, adjust training plans, and thus better support the development of users' cognitive abilities.

[0174] Step S6: Setting the level-breaking mode and reward mechanism;

[0175] In order to enhance the fun of the training process and the user's enthusiasm for participation, this embodiment sets a level-breaking mode and a reward mechanism; the training task includes several levels, and the user will obtain virtual points or rewards for each level passed, which motivates the user to continue to participate in the training.

[0176] Each level has a clear goal and difficulty; the first level requires the user to complete a simple task, and the second level will increase the complexity and challenge of the task. The higher the level, the greater the difficulty; when the user completes the current level, the system will automatically unlock the next level, and train the user's cognitive ability by gradually increasing the difficulty of the task.

[0177] After successfully completing each level, users can obtain a certain amount of virtual currency. When users perform well in a task, they can obtain medals or trigger specific animation effects to motivate users and enhance their sense of achievement. Animation effects include visual feedback, auditory feedback, and text feedback. Visual feedback includes virtual fireworks effects, auditory feedback includes encouraging sound effects, and text feedback includes "Excellent performance!"

[0178] In the attention and executive function assessment and training system based on virtual reality technology, the specific operations are as follows:

[0179] 1. The user wears a virtual technology headset to enter the system;

[0180] 2. Open the tablet APP on the tablet device, create personal information, select the assessment task, and select and adjust the difficulty parameters of the task; the difficulty parameters include stimulus presentation time, that is, the duration of the target stimulus on the display; the number of stimuli, that is, the number of stimuli presented simultaneously in the virtual scene, and the reaction time limit, that is, the reaction time window for completing a task.

[0181] 3. Use the handle to select and click on this assessment task to enter this assessment task;

[0182] 4. Enter the practice part, and after passing the formal assessment, the game ends after the assessment.

[0183] 5. Determine whether the accuracy is between 65% and 85%; if not, adjust the difficulty parameters of the task and evaluate again; if so, upload the task performance data, i.e., objective data and subjective data in the questionnaire scale, standardize the data, perform data fusion, obtain the evaluation results, and generate a visual evaluation report;

[0184] 6. Enter the cognitive ability training model and manually enter the threshold;

[0185] 7. According to the personalized user evaluation results, set the nth level and enter the evaluation model to analyze the current task performance and determine the cognitive ability level;

[0186] 8. Make adaptive adjustments, implement reinforcement learning models, and adjust the difficulty of tasks to increase, decrease, or maintain the difficulty of tasks;

[0187] 9. Update the task and proceed to level n+1;

[0188] 10. After passing the n+1 level, determine whether the comprehensive score exceeds the preset threshold, that is, whether various indicators are met, and record the performance; if it exceeds the preset threshold, the training ends; if it does not exceed the preset threshold, enter the evaluation model, determine the level of cognitive ability, and make adaptive adjustments again.

[0189] In summary, the present invention uses the Unity3D game engine combined with 3D modeling technology to construct several virtual scenes. Users can interact immersively through virtual reality head-mounted devices and handles, thereby enhancing the authenticity and fun of evaluation and training. The virtual scene that combines daily life scenes and game scenes is a diversified evaluation environment, which not only improves the authenticity and practical significance of the evaluation results, reflects the performance of users in real situations, but also stimulates the interest of users and increases their participation.

[0190] The difficulty of the task used for assessment is dynamically adjusted based on the user's age, cognitive ability level and real-time performance data. First, the initial value of the task difficulty parameter is set according to the user's age and initial cognitive ability level. When a task cycle is completed, the user's task performance data, i.e. objective data, is recorded, and the task difficulty coefficient is updated using a formula to regenerate a new task. Based on personalized cognitive assessment, the current training content and difficulty are dynamically adjusted in combination with the Q-learning algorithm. By adaptively adjusting the task difficulty, the optimal task setting can be found in a relatively short period of time, ensuring that the training content is both appropriately challenging and can maximize learning effects, helping users make the greatest progress in attention and executive functions. The Q-learning algorithm is used, and the training content and difficulty are adaptively adjusted based on the user's real-time performance to ensure that the training task is in the user's "zone of proximal development", thus achieving truly personalized cognitive training and avoiding excessive challenges or overly simple problems caused by fixed task settings, so that each user can achieve the best cognitive improvement effect.

[0191] Through virtual reality technology head-mounted devices and handles, the user's task performance data, i.e. objective data, is collected in real time. Through questionnaire scales, the user's subjective report data, i.e. subjective data, is collected. The objective data and subjective main sentences are standardized, and a weighted regression model, i.e. a hybrid model, is used to combine the objective data with the subjective data to generate a visual evaluation report. This overcomes the limitations of the single data source in traditional methods, significantly improves the comprehensiveness and scientificity of the evaluation results, and provides a more solid foundation for accurate judgment and intervention.

[0192] By integrating virtual reality technology headsets, handles, tablet devices and website backends, a complete full-process closed-loop system has been built. The hardware equipment ensures that users have an immersive experience in the virtual environment, and the software can realize the automated process of data collection, data processing, evaluation and training; it significantly improves the system's operational convenience and management efficiency, making it easier for professionals to efficiently manage user data, quickly generate evaluation reports, and dynamically adjust training plans.

[0193] By adding a level-breaking mode and reward mechanism to the virtual scene, a point reward and punishment mechanism is implemented to encourage user participation, thereby increasing the fun and interactivity of assessment and training; it solves the problems of user inattention and low participation that are easily caused in traditional cognitive assessment and training processes, and is suitable for long-term training and intervention of users.

[0194] VR virtual reality technology has the advantages of immersion, interactivity and controllability. Virtual reality technology can create highly realistic virtual scenes, such as complex traffic environments, work and learning scenes, etc. Through the interactivity of virtual reality technology, individuals can perform a variety of interactive tasks in a virtual environment, which can increase the fun and participation of the evaluation process, reduce the psychological burden of the subjects, and make the evaluation results more credible. Virtual reality technology can accurately record the objective data of individuals during execution, including reaction time, accuracy, etc., providing reliable data support for the quantitative evaluation of cognitive abilities and the formulation of personalized intervention plans. In addition, virtual reality technology combined with adaptive algorithms can dynamically adjust the difficulty of tasks according to the user's performance in the task, making the evaluation and training more personalized, ensuring both appropriate challenges and maintaining interest and enthusiasm.

[0195] The above implementation modes are not limitations of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the scope of the technical solution of the present invention also belong to the protection scope of the present invention.

Claims

1. A virtual reality-based attention and executive function assessment and training system, including an assessment model and a cognitive ability training model, characterized by: The evaluation model includes a user registration module, which is used for users to register personal accounts and establish user profiles; a parameter adjustment module, which is used to provide an interface tool to allow users to adjust the difficulty parameters of the tasks used for evaluation before the formal evaluation to achieve a personalized testing experience; a personalized evaluation module, which is used to evaluate the performance of users' attention and executive abilities according to their cognitive ability differences; The cognitive ability training model includes an adaptive ability training module, which is used to dynamically adjust the difficulty parameters of the task based on the Q-learning algorithm to achieve continuous intervention in user ability; The data management module is used to record and manage user behavior, objective data, and script running data, upload the data to the data platform for standardization, and integrate the data to generate a training result report; The level-breaking and level-clearing reward module is used to enhance the user's sense of achievement and motivate users to participate in training.

2. The method for evaluating and training attention and executive function based on virtual reality according to claim 1, characterized in that: The specific steps include: Step S1: Based on the C# programming language and the Unity3D game engine, combined with 3D modeling software, several virtual scenes are constructed, including daily life scenes and game scenes; Step S2: construct a dynamic adjustment model to dynamically adjust the difficulty parameters of the task based on the real-time monitored behavioral data and the user's age and cognitive ability level; the difficulty parameters of the task include stimulus presentation time T_s, stimulus number N_s and reaction time limit T_r; Step S3: Execute the task, evaluate the execution of the user's task, and generate a visual evaluation report; Step S4: The evaluation report is combined with an adaptive algorithm to automatically adjust the current training difficulty and content of a personalized learning method based on the Q-learning algorithm according to the user's real-time performance in the virtual reality technology, and to conduct training; Step S5: Build a full-process closed-loop system by integrating virtual reality technology head-mounted devices, handles, tablet devices and website background.

3. The method for assessing and training attention and executive functions based on virtual reality according to claim 2, characterized in that: The formula for dynamically adjusting the model in step S2 is: P = f(C, A, R); Among them, P is the difficulty parameter vector of the current task, which includes T_s, N_s and T_r; C is the user's age or cognitive ability level; A is the accuracy of the current task completion; R is the reaction time; The rule for adjusting the stimulus presentation time T_s is: T_s=T s0 -α*A+β*max(0,T_r-R); Among them, T s0 is the initial stimulus presentation time; α is the weight of accuracy on stimulus presentation time; β is the weight of reaction time on stimulus presentation time; R is reaction time; The rule for adjusting the number of stimuli N_s is: N_s=N s0 +γ*A; Among them, N s0 is the initial number of stimuli; γ is the growth factor of accuracy to the number of stimuli; A is the accuracy of the current task; The rule for adjusting the reaction time limit T_r is: T_r=T s0 -δ*A; Among them, T s0 is the initial reaction time limit; δ is the influence coefficient of accuracy on reaction time limit; A is the accuracy of the current task.

4. The method for assessing and training attention and executive functions based on virtual reality according to claim 3, characterized in that: The step S2 of dynamically adjusting the difficulty parameter of the task specifically includes the following steps: Step S21: Initialize the difficulty parameters of the task and initialize the weights of each difficulty parameter; Step S22: The user performs the task. After the task is completed, the accuracy A and reaction time R of the user performing the task are recorded; Step S23: updating the difficulty parameter of the task based on the dynamic adjustment model; Step S24: Generate a new task according to the difficulty parameter value of the updated task; Step S25: execute the new task and determine whether the accuracy of the new task is greater than 65% and less than 85%; if so, output the user's final task performance data; if the accuracy of the new task is not greater than 65%, reduce the weight α of the accuracy to the stimulus presentation time and the growth factor γ of the accuracy to the number of stimuli; if the accuracy of the new task is not less than 85%, increase the weight α of the accuracy to the presentation time and the growth factor γ of the accuracy to the number of stimuli.

5. The method for evaluating and training attention and executive function based on virtual reality according to claim 2, characterized in that: The tasks in step S2 are cognitive tasks in the field of attention and executive function, and the cognitive tasks include sustained attention test, selective attention test, working memory test, response inhibition test, cognitive flexibility test and delayed gratification test.

6. The method for assessing and training attention and executive functions based on virtual reality according to claim 2, characterized in that: The step S3 evaluates the execution of the user task, which specifically includes the following steps: Step S31: collect objective data and subjective data respectively, and transmit both objective data and subjective data to the backend server; objective data includes accuracy, reaction time and task completion; subjective data is the subjective report data of the user in the questionnaire; Step S32: Standardize the objective data and subjective data to ensure that all data are integrated under the same dimension; normalize the subjective data, and the normalization formula is: normalized_score=(score-min_score) / (max_score-score); Among them, score is the initial score; min_score is the minimum value in the score; max_score is the maximum value in the score; Step S33: construct a hybrid model for multi-dimensional fusion, which is used to combine objective data with subjective data to form a comprehensive score S; the calculation formula of the comprehensive score S is: S=0.4×ZR+0.5×ZA+0.1×normalized_score+b; Among them, ZR is the Z score of reaction time; ZA is the Z score of accuracy; b is the bias term, which is a constant; Step S34: Generate a visual evaluation report based on the results of multi-dimensional data fusion; the evaluation report includes scores of various cognitive abilities, comprehensive evaluations, and improvement suggestions.

7. The method for assessing and training attention and executive functions based on virtual reality according to claim 2, characterized in that: The formula of the Q-learning algorithm in step S4 is: Q(S,A1)=Q(S,A1)+α[R+Υmax A1 Q(S',A1′)-Q(S,A1)]; Among them, Q(S,A1) is the Q value of taking action A1 in state S; R is the immediate reward obtained after taking the action in the current state; max A Q(S',A1') is the maximum Q value obtained by taking the optimal action A1' in the next state S'; α is the learning rate; Υ is the discount factor.

8. The method for assessing and training attention and executive functions based on virtual reality according to claim 7, characterized in that: The calculation formula of the instant reward is: R′=c1*(1-reaction time / maximum time limit)+c2*accuracy; Among them, c1 is the weight coefficient of reaction time; c2 is the weight coefficient of accuracy.

9. The method for assessing and training attention and executive functions based on virtual reality according to claim 7, characterized in that: The personalized learning method for automatically adjusting the current training difficulty and content based on the Q-learning algorithm in step S4 specifically includes the following steps: Step S41: setting the difficulty parameter of the initial task according to the evaluation result and determining the initial training level; Step S42: generating an initial task based on the initial training level and feeding back to the user; Step S43: collect objective data of the user in the task through the virtual reality technology head mounted device and the handle, and normalize the objective data into a state vector as the current state S; Step S44: Create a Q value table, and select an action A1 from the Q value table according to the current state S; Step S45: updating the Q value table based on the Q-learning algorithm; Step S46: Setting the difficulty adjustment step length δ to avoid excessive adjustment of the task difficulty, which may cause discontinuous changes in the task difficulty; Step S47: Collect objective data and subjective data in real time through the virtual reality technology head-mounted device and handle, and update the Q value table until the Q value table converges.

10. The method for assessing and training attention and executive functions based on virtual reality according to claim 9, characterized in that: The formula for setting the difficulty adjustment step length δ in step S46 is: δ=max(0.1,1 / sqrt); Among them, sqrt is the number of completed tasks.

Images