A method for constructing a virtual kitchen environment and training a model for cognitive training.

Abstract

This invention discloses a virtual kitchen environment construction and model training method for cognitive training, belonging to the field of cognitive assessment technology. Specifically, it includes: constructing a first task region and a second task region; plotting the patient's training score changes with the number of training sessions within a set period to obtain a first change curve and a second change curve; calculating the average slope of the two change curves; if the average slope of either change curve is greater than 0, calculating the similarity value between the two change curves; if the similarity value is less than a preset threshold, selecting the second change curve as the target curve; if the similarity value is greater than or equal to the preset threshold, selecting the curve segment with the higher score in each training task and marking it as the preferred curve segment; splicing the preferred curve segments and marking them as the target curve; and adjusting the difficulty of the patient's subsequent training tasks based on the average training score of the target curve. This invention improves the accurate identification of the patient's cognitive training progress.

Description

Technical Field

[0001] This invention relates to the field of cognitive assessment technology, specifically to a method for constructing a virtual kitchen environment and training a model for cognitive training. Background Technology

[0002] Cognitive function encompasses many aspects, including memory, comprehension, calculation, language, visuospatial skills, and judgment. Impairment in one or more of these cognitive functions can be considered a cognitive disorder.

[0003] Virtual reality technology has developed rapidly in recent years. Its inherent characteristics of high immersion, multi-sensory perception, and strong interactivity provide a means of reproduction and expansion for cognitive function assessment methods. Virtual reality-based systems can flexibly control the representation and content of stimuli, the complexity of tasks, and the subject's response state. At the same time, they can directly record rich and accurate data information of subjects in tasks, providing a foundation for further processing using artificial intelligence algorithms. Thus, they offer a feasible new means for the reliability and ecological effectiveness of cognitive function assessment.

[0004] A Chinese patent with publication number CN118675707A discloses a modeling method, monitoring method, and system for monitoring the cognitive training process. In its solution, coordinate points are marked based on multiple cognitive training scores of patients with cognitive impairment, and multiple standard learning curves are fitted using a preset function to obtain multiple sets of parameter combinations. This enables the monitoring of the cognitive training process of a single patient and outputs the patient's relative achievement level and learning progress.

[0005] However, this approach has inherent limitations: as the number of training sessions increases, task proficiency may interfere with the effectiveness of cognitive training scores. Specifically, repeatedly performing the same task may lead patients to develop "task process memory," potentially shifting neural processing from a "cognitive control mode" to an "automatic processing mode." In this case, the cognitive resources required for task execution, such as working memory and attention, are significantly reduced. The improved patient performance in this situation does not stem from genuine cognitive ability improvement, but rather from increased mechanical familiarity with the task process. This causes the assessment model to misjudge "proficiency gain" as "cognitive function progress," ultimately leading to systematic errors in the training progress assessment and affecting the accuracy of judging the patient's learning progress. Summary of the Invention

[0006] The purpose of this invention is to provide a method for constructing a virtual kitchen environment and training a model for cognitive training, thereby solving the following technical problems:

[0007] Without establishing a mechanism to separate "cognitive ability contribution" from "proficiency contribution," it is impossible to distinguish the respective proportions of genuine improvement in cognitive function and familiarity with task procedures in score improvement, thus affecting the accuracy of judging the patient's learning progress.

[0008] The objective of this invention can be achieved through the following technical solutions:

[0009] A method for constructing a virtual kitchen environment and training a model for cognitive training includes the following steps:

[0010] S1. Construct a first task area and a second task area in a virtual kitchen scene; the first task area is used to provide training tasks with the same task format and difficulty; the second task area is used to provide training tasks with different task formats but the same difficulty, and the training tasks in the first task area and the second task area have the same difficulty.

[0011] S2, within a set period, plot the curves of the patient’s training score as a function of the number of training sessions in the first task region and the second task region respectively, to obtain the first curve and the second curve.

[0012] S3, calculate the average slope of the two curves. If the average slope of either curve is greater than 0, calculate the similarity value between the two curves.

[0013] S4. If the similarity value is less than the preset threshold, the second change curve is selected as the target curve; if the similarity value is greater than or equal to the preset threshold, the first change curve and the second change curve are superimposed, the curve segment with the higher score in each training task is selected and marked as the preferred curve segment, the preferred curve segments are spliced ​​together and marked as the target curve, and the difficulty of the patient's subsequent training tasks is adjusted according to the average training score of the target curve.

[0014] As a further aspect of the present invention: S2 further includes obtaining the end timestamp of each cognitive training session, and calculating the time difference between the end timestamps of the i-th and i+1-th cognitive training sessions. If the time difference is less than or equal to a preset time threshold, the training data corresponding to the i+1-th session is removed.

[0015] As a further aspect of the present invention: in step S3, the specific calculation process for the similarity value is as follows:

[0016] S11, calculate the mean training score A' of the first change curve and the mean training score B' of the second change curve respectively;

[0017] S12, construct a deviation sequence corresponding to any task region based on the mean training score, calculate the covariance based on the deviation sequence, and calculate the similarity value between the change curves based on the covariance.

[0018] S13, the formula for calculating similarity values ​​is:

[0019] ;

[0020] Where n is the total number of training sessions for the patient in any task region, Ai is the patient's training score in the i-th training session in the first task region, and Bi is the patient's training score in the i-th training session in the second task region.

[0021] A further aspect of this invention includes obtaining the training score sequence A=[A1,A2,...,Az] corresponding to the first task region if the total number of training iterations in the first task region is not equal to the total number of training iterations in the second task region; similarly, obtaining the training score sequence B=[b1,b2,...,bm] corresponding to the second task region, constructing a cumulative distance matrix Cost based on the training score sequences A and B, calculating the minimum distance value Cost[z,m] of the cumulative distance matrix Cost, and then applying the calculation formula... The similarity value is calculated.

[0022] As a further aspect of the present invention: if the patient's training behavior data within a set period is less than a preset training behavior data threshold, the set period [G, G+T] is adjusted to [G, G+T+t'], where t' is a preset time length and T is the time length of the set period; if the patient's training behavior data within the adjusted set period continues to be less than the preset training behavior data threshold, the above process is repeated until the patient's training behavior data within the adjusted set period is greater than or equal to the preset training behavior data threshold.

[0023] As a further aspect of the present invention: the process of obtaining the target curve is as follows:

[0024] The first and second change curves are spatially superimposed, and an aligned coordinate system is constructed based on the training number sequence dimension. For each training number K, the training scores of the two curves under that number are extracted, and the maximum value is selected as the optimal data point corresponding to number K by numerical comparison. All optimal data points are arranged in the time series order of the training number sequence and connected to obtain the target curve.

[0025] As a further aspect of the present invention: S4 further includes obtaining the training method for the task region corresponding to the target curve, and determining the training method as the optimal training method for the patient.

[0026] As a further aspect of the present invention, S3 also includes issuing an early warning message and reducing the difficulty of subsequent training tasks for patients if the average slope of all the changing curves is less than 0.

[0027] The beneficial effects of this invention are:

[0028] 1) This invention constructs two types of task areas within a virtual kitchen. The first type provides training tasks with fixed format and difficulty, while the second type provides tasks with varying formats but constant difficulty. This differentiated design of task formats effectively separates the contributions of "cognitive ability improvement" and "task proficiency growth" to training scores. When patients develop procedural memory through repetitive training in the first type of area, the second type of area, with its variations in operational modality and information presentation, reduces the interference of proficiency on scores. By comparing the training curves of the two types of areas, the proportion of genuine improvement in cognitive function within the score improvement can be accurately identified, avoiding the misjudgment of performance improvement resulting from rote memorization as cognitive progress.

[0029] 2) This invention filters out abnormal training data using timestamps, dynamically adjusts the set period to ensure sufficient data volume, and combines dynamic time warping algorithms with covariance calculations to determine curve similarity, thus constructing a rigorous data processing and analysis system. This system can eliminate invalid data caused by excessively short training intervals, avoid statistical biases caused by insufficient data, and quantify the similarity of training curves in different task regions through normalization and path distance calculation. This process can accurately capture the influence of proficiency on changes in training scores, providing a reliable data foundation for subsequent evaluation and improving the scientific rigor and credibility of cognitive training process assessment.

[0030] 3) This invention establishes a hierarchical evaluation strategy based on the comparison of curve similarity with a preset similarity threshold: when the similarity is low, the second type of region curve, which is resistant to proficiency interference, is used as the evaluation basis; when the similarity is high, the first and second change curves are superimposed, and the curve segment with the higher score in each training task is selected and marked as the preferred curve segment. The preferred curve segment is then spliced ​​together and marked as the target curve. When the similarity is higher than the preset threshold, the target curve is constructed by superimposing the two curves and splicing the high-scoring segments, thereby achieving dynamic integration of the advantages of the two types of tasks. This hierarchical strategy constructs an evaluation system that integrates anti-interference and dynamic adaptability through a progressive process of "threshold comparison - curve screening - segment splicing". On the one hand, the second region curve is used as the "anti-proficiency baseline" to ensure the reliability of the lower limit of evaluation; on the other hand, the "upper limit of cognitive ability" is formed by splicing high-scoring segments to comprehensively cover the patient's best performance under different task formats. This strategy can quickly locate the true cognitive level when proficiency interference is significant, and can also optimize the evaluation dimensions by comprehensively analyzing multi-task data during the ability improvement stage, providing a decision-making basis that is both accurate and flexible for subsequent training difficulty adjustment and personalized plan formulation. Attached Figure Description

[0031] The invention will now be further described with reference to the accompanying drawings.

[0032] Figure 1This is a schematic diagram of a method for constructing a virtual kitchen environment and training a model for cognitive training according to the present invention. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] Please see Figure 1 As shown, this invention provides a method for constructing a virtual kitchen environment and training a model for cognitive training, comprising the following steps:

[0035] S1. Construct a first task area and a second task area in a virtual kitchen scene; the first task area is used to provide training tasks with the same task format and difficulty; the second task area is used to provide training tasks with different task formats but the same difficulty, and the training tasks in the first task area and the second task area have the same difficulty.

[0036] S2, within a set period, plot the curves of the patient’s training score as a function of the number of training sessions in the first task region and the second task region respectively, to obtain the first curve and the second curve.

[0037] S3, calculate the average slope of the two curves. If the average slope of either curve is greater than 0, calculate the similarity value between the two curves.

[0038] S4. If the similarity value is less than the preset threshold, the second change curve is selected as the target curve; if the similarity value is greater than or equal to the preset threshold, the first change curve and the second change curve are superimposed, the curve segment with the higher score in each training task is selected and marked as the preferred curve segment, the preferred curve segments are spliced ​​together and marked as the target curve, and the difficulty of the patient's subsequent training tasks is adjusted according to the average training score of the target curve.

[0039] 1) This invention constructs two types of task areas within a virtual kitchen. The first type provides training tasks with fixed format and difficulty, while the second type provides tasks with varying formats but constant difficulty. This differentiated design of task formats effectively separates the contributions of "cognitive ability improvement" and "task proficiency growth" to training scores. When patients develop procedural memory through repetitive training in the first type of area, the second type of area, with its variations in operational modality and information presentation, reduces the interference of proficiency on scores. By comparing the training curves of the two types of areas, the proportion of genuine improvement in cognitive function within the score improvement can be accurately identified, avoiding the misjudgment of performance improvement resulting from rote memorization as cognitive progress.

[0040] 2) This invention filters out abnormal training data using timestamps, dynamically adjusts the set period to ensure sufficient data volume, and combines dynamic time warping algorithms with covariance calculations to determine curve similarity, thus constructing a rigorous data processing and analysis system. This system can eliminate invalid data caused by excessively short training intervals, avoid statistical biases caused by insufficient data, and quantify the similarity of training curves in different task regions through normalization and path distance calculation. This process can accurately capture the influence of proficiency on changes in training scores, providing a reliable data foundation for subsequent evaluation and improving the scientific rigor and credibility of cognitive training process assessment.

[0041] 3) This invention establishes a hierarchical evaluation strategy based on the comparison of curve similarity with a preset similarity threshold: when the similarity is low, the second type of region curve, which is resistant to proficiency interference, is used as the evaluation basis; when the similarity is high, the first and second change curves are superimposed, and the curve segment with the higher score in each training task is selected and marked as the preferred curve segment. The preferred curve segment is then spliced ​​together and marked as the target curve. When the similarity is higher than the preset threshold, the target curve is constructed by superimposing the two curves and splicing the high-scoring segments, thereby achieving dynamic integration of the advantages of the two types of tasks. This hierarchical strategy constructs an evaluation system that integrates anti-interference and dynamic adaptability through a progressive process of "threshold comparison - curve screening - segment splicing". On the one hand, the second region curve is used as the "anti-proficiency baseline" to ensure the reliability of the lower limit of evaluation; on the other hand, the "upper limit of cognitive ability" is formed by splicing high-scoring segments to comprehensively cover the patient's best performance under different task formats. This strategy can quickly locate the true cognitive level when proficiency interference is significant, and can also optimize the evaluation dimensions by comprehensively analyzing multi-task data during the ability improvement stage, providing a decision-making basis that is both accurate and flexible for subsequent training difficulty adjustment and personalized plan formulation.

[0042] In a preferred embodiment of the present invention, step S2 further includes obtaining the end timestamp of each cognitive training session and calculating the time difference between the end timestamps of the i-th and i+1-th cognitive training sessions. If the time difference is less than or equal to a preset time threshold, the training data corresponding to the i+1-th session is removed.

[0043] It is understandable that repeatedly training patients within a short period could distort the data. For example, if a patient trains continuously within a short timeframe, residual memory or incomplete digestion of the training content may cause subsequent training scores to fail to accurately reflect changes in cognitive ability, instead being influenced by short-term memory or fatigue. By removing training data with excessively short intervals, it is ensured that the training scores included in the analysis are obtained under the premise that the patient has a reasonable cognitive recovery period. This avoids interference from "invalid training data" on the cognitive training score change curve, ensuring that when assessing the training progress based on curve similarity, the data more accurately reflects the actual changes in the patient's cognitive function, rather than being affected by false improvements caused by inappropriate training frequency or intervals.

[0044] In another preferred embodiment of the present invention, the specific calculation process of the similarity value in step S3 is as follows:

[0045] S11, calculate the mean training score A' of the first change curve and the mean training score B' of the second change curve respectively;

[0046] S12, construct a deviation sequence corresponding to any task region based on the mean training score, calculate the covariance based on the deviation sequence, and calculate the similarity value between the change curves based on the covariance.

[0047] S13, the formula for calculating similarity values ​​is:

[0048] ;

[0049] Where n is the total number of training sessions for the patient in any task region, Ai is the patient's training score in the i-th training session in the first task region, and Bi is the patient's training score in the i-th training session in the second task region.

[0050] Normalization eliminates the influence of score scale differences across different task regions, making the two sets of data comparable. Calculating the mean and bias sequences extracts the fluctuation characteristics of the curves relative to the average level. Covariance measures the overall trend of two variables; when the trends of the two curves are consistent, the covariance is positive and large, and vice versa. By standardizing the covariance and converting it into a similarity value, the degree of similarity between the two curves can be quantified. This process accurately captures the correlation between training score changes in the first task region (fixed task format) and the second task region (variable task format). A high similarity value indicates that the patient's performance trends in the two types of tasks are similar, potentially indicating a significant interference of proficiency in the assessment. A low similarity value indicates that the variation in the second task region effectively reduces the influence of proficiency, thus providing a scientific basis for subsequently judging the degree of interference of proficiency on the assessment results based on curve similarity. This ensures that the assessment model can accurately separate "cognitive ability contribution" from "proficiency contribution," improving the reliability of training process assessment.

[0051] In another preferred embodiment of the present invention, if the total number of training iterations in the first task region is not equal to the total number of training iterations in the second task region, then the training score sequence A=[A1,A2,...,Az] corresponding to the first task region is obtained; similarly, the training score sequence B=[b1,b2,...,bm] corresponding to the second task region is obtained, a cumulative distance matrix Cost is constructed based on the training score sequences A and B, and the minimum distance value Cost[z,m] of the cumulative distance matrix Cost is calculated according to the calculation formula. The similarity value is calculated.

[0052] It is understandable that when patients have different training counts in two task regions, directly comparing curves can lead to assessment distortion due to unequal sample sizes. The Dynamic Time Warping (DTW) algorithm, by constructing a cumulative distance matrix, allows sequences of different lengths to be flexibly aligned on the time axis, finding the optimal matching path so that each point in the short sequence can reasonably match its corresponding point in the long sequence. By calculating the minimum cumulative distance and converting it into a similarity function, it ensures that training curves with different counts are quantified for similarity under a unified standard, thereby accurately determining whether the trends of the two curves are consistent. This process avoids the "interference of proficiency on assessment" or the generation of "spurious similarity" due to differences in the number of training counts. This allows subsequent similarity-based assessments to truly reflect the changes in patients' cognitive abilities and the degree of influence of proficiency under different task formats, providing key technical support for scientifically separating the contribution of cognitive function improvement and task process familiarity.

[0053] In another preferred embodiment of the present invention, if the patient's training behavior data within a set period is less than a preset training behavior data threshold, the set period [G, G+T] is adjusted to [G, G+T+t'], where t' is a preset time length and T is the time length of the set period; if the patient's training behavior data within the adjusted set period continues to be less than the preset training behavior data threshold, the above process is repeated until the patient's training behavior data within the adjusted set period is greater than or equal to the preset training behavior data threshold.

[0054] In practical applications, if the preset period T is 7 days, the preset training behavior data threshold is 5 training sessions, and the preset extension time t' is 3 days, when the patient completes only 3 training sessions (less than 5) within the first preset period [G, G+7], the system will automatically adjust the preset period to [G, G+7+3], i.e., [G, G+10]. If the patient completes 4 training sessions (still less than 5) within the adjusted period, the period will be extended again by 3 days to [G, G+13], until the patient completes 5 or more training sessions within a certain adjusted period, for example, 6 training sessions within the [G, G+16] period. At this point, the period adjustment will stop, and these 6 training sessions will be included in the analysis.

[0055] Ensure that the amount of training data used for analysis meets the statistically valid sample size requirements to avoid biases in curve similarity calculations due to insufficient data. For example, if there is too little data within a set period, accidental factors may prevent the training score curve from accurately reflecting the trend of changes in patients' cognitive abilities, just as a small sample cannot accurately infer overall characteristics. By dynamically extending the set period until the data volume meets the standard, it is ensured that subsequent similarity calculations and proficiency impact assessments based on training score curves are built on a reliable data foundation. This allows the assessment model to accurately capture the differences in cognitive function changes and proficiency contributions in different task areas, thereby avoiding misjudgments of "improved cognitive ability" and "enhanced task proficiency" due to data sparsity. This provides solid data support for scientifically evaluating the training process and developing personalized training programs.

[0056] In another preferred embodiment of the present invention, the process of obtaining the target curve is as follows:

[0057] The first and second change curves are spatially superimposed, and an aligned coordinate system is constructed based on the training number sequence dimension. For each training number K, the training scores of the two curves under that number are extracted, and the maximum value is selected as the optimal data point corresponding to number K by numerical comparison. All optimal data points are arranged in the time series order of the training number sequence and connected to obtain the target curve.

[0058] When the similarity value is greater than or equal to a preset threshold, it indicates that the training of the two different task formats has produced similar improvement effects on the patient. This suggests that the patient's cognitive training was not significantly affected by task proficiency, but rather that their cognitive abilities themselves were genuinely improved. Based on this, by obtaining the intersection point through curve overlay, the balance point of the two task training effects can be determined, reflecting the patient's ability transfer nodes between different tasks. Identifying inflection points can capture changes in the rate of improvement of the patient's abilities during training, dividing the curve into more meaningful sub-segments. Using a difference function to select the optimal curve segment ensures that the best performance under both task formats is considered, avoiding evaluation bias caused by the limitations of a single task format. The final target curve formed by splicing the curves more comprehensively and accurately reflects the patient's true cognitive ability level after eliminating the interference of proficiency, providing a reliable basis for subsequent adjustments to training difficulty and determination of the optimal training method. This makes the training plan more precisely targeted at the patient's cognitive ability improvement needs, enhancing the effectiveness and scientific rigor of rehabilitation training.

[0059] In another preferred embodiment of the present invention, step S4 further includes obtaining the training method for the task region corresponding to the target curve, and determining the training method as the optimal training method for the patient.

[0060] The optimal training method for the patient is determined by identifying the training method corresponding to the target curve within the task region. This is because the target curve itself is obtained by scientifically comparing the training curves of the first task region (fixed task format) and the second task region (variable task format), thus accurately reflecting the trend of cognitive ability changes after eliminating the interference of task proficiency. When the similarity value is greater than or equal to a preset threshold, the target curve is formed by splicing the high segments of the two curves, integrating the advantages of both task formats. When the similarity value is less than the preset threshold, the target curve directly adopts the second task region curve, which is resistant to proficiency interference. In this case, the training method corresponding to the target curve is either the mode that better stimulates the patient's cognitive ability among the two task formats, or a task design that effectively avoids the influence of proficiency. Determining it as the optimal training method can accurately match the patient's current cognitive state and learning characteristics, enabling subsequent training tasks to maximize the improvement of cognitive function while avoiding misjudgment of proficiency due to a single task format. This improves training efficiency and provides data support for the development of personalized training plans, ultimately helping to achieve accurate assessment and scientific intervention of the patient's cognitive training process.

[0061] In another preferred embodiment of the present invention, step S3 further includes issuing an early warning message and reducing the difficulty of subsequent training tasks for patients if the average slope of all the change curves is less than 0.

[0062] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A virtual kitchen environment construction and model training method for cognitive training, characterized by, The method comprises the following steps: S1, constructing a first task area and a second task area in a virtual kitchen scene; the first task area is used to provide training tasks with the same form and difficulty; the second task area is used to provide training tasks with different forms but the same difficulty; the training tasks of the first task area and the second task area have the same difficulty; S2, drawing a change curve of the training score of the patient in the first task area and a change curve of the training score of the patient in the second task area within a set period, to obtain a first change curve and a second change curve; S3, calculating the average slope of the two change curves, and if the average slope of any change curve is greater than 0, calculating the similarity value between the two change curves; S4, if the similarity value is less than a preset threshold, selecting the second change curve as a target curve; if the similarity value is greater than or equal to the preset threshold, superimposing the first change curve and the second change curve, selecting a curve segment with a higher training task score each time, marking the curve segment as an optimal curve segment, and splicing the optimal curve segment and marking it as a target curve; the target curve is obtained by: superimposing the first change curve and the second change curve in space, constructing an alignment coordinate system based on the training number sequence dimension; for each training number sequence K, extracting the training scores of the two curves under the sequence, and selecting the maximum value as the optimal data point corresponding to the sequence K through numerical comparison; arranging all optimal data points in the time sequence order of the training number sequence and connecting them to obtain the target curve; adjusting the difficulty of the subsequent training task of the patient according to the average training score of the target curve; obtaining the training method of the task area corresponding to the target curve, and determining the training method as the optimal training method of the patient. 2.The virtual kitchen environment construction and model training method for cognitive training of claim 1, wherein, In S2, the end time stamp of each cognitive training of the patient is obtained, the time difference between the end time stamp of the i-th cognitive training and the end time stamp of the i+1-th cognitive training is calculated, and if the time difference is less than or equal to a preset time threshold, the training data corresponding to the i+1-th cognitive training is excluded. 3.The virtual kitchen environment construction and model training method for cognitive training of claim 2, wherein, If the training behavior data of the patient within a set period is less than a preset training behavior data threshold, the set period [G, G+T] is adjusted to [G, G+T+t'], where t' is a preset time length and T is the time length of the set period; if the training behavior data of the patient within the adjusted set period continues to be less than the preset training behavior data threshold, the above process is repeated until the training behavior data of the patient within the adjusted set period is greater than or equal to the preset training behavior data threshold. 4.The virtual kitchen environment construction and model training method for cognitive training of claim 1, wherein, In S3, if the average slope of the change curve is less than 0, an early warning information is sent, and the difficulty of the subsequent training task of the patient is reduced.

Images