Visual training effect evaluation method and system

Abstract

The invention provides a visual training effect evaluation method and system. The method comprises the following steps: acquiring sample visual training effect data; inputting the sample visual training effect data into a preset data expansion model to obtain expansion data; constructing a classifier based on a Monte Carlo algorithm and a preset decision tree model, and constructing an initial evaluation model based on the classifier; training the initial evaluation model according to the expansion data and the sample visual training effect data to obtain a target evaluation model; obtaining to-be-tested vision training effect data, and inputting the vision training effect data into the target evaluation model, so that the target evaluation model performs feature processing on the to-be-tested vision to obtain to-be-tested feature data, and then the target evaluation model classifies the to-be-tested feature data based on the classifier to obtain a vision training effect evaluation result. Therefore, the accuracy of visual training effect evaluation is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of vision assessment, and in particular to a method and system for evaluating the effect of vision training. Background Art

[0002] The evaluation of vision training effectiveness is gaining increasing attention and has a wide range of applications. For example, for patients with myopia, hyperopia, astigmatism, and amblyopia, vision training effectiveness assessment can effectively determine the degree of vision improvement achieved through a series of visual activities and exercises. However, traditional methods for evaluating vision training effectiveness rely primarily on subjective judgment or simple vision tests, which fail to fully reflect the effectiveness of training and lack accuracy and objectivity. The development of artificial intelligence (AI), particularly machine learning and deep learning, has provided new approaches to address the complex data analysis problem of evaluating vision training effectiveness. However, traditional AI still has limitations in model construction. For example, data augmentation may not fully simulate the real-world data distribution, resulting in large bias in the training sample, and overly conservative architecture selection can lead to inaccurate evaluation results. Summary of the Invention

[0003] The present invention aims to provide a method and system for evaluating the effect of vision training, so as to solve the above-mentioned technical problems and improve the accuracy of the evaluation of the effect of vision training.

[0004] In order to solve the above technical problems, the present invention provides a method for evaluating the effect of vision training, comprising the following steps:

[0005] Obtain sample vision training effect data;

[0006] Inputting the sample vision training effect data into a preset data expansion model to obtain expanded data;

[0007] Constructing a classifier based on a Monte Carlo algorithm and a preset decision tree model, and constructing an initial evaluation model based on the classifier;

[0008] Training the initial evaluation model according to the expanded data and the sample vision training effect data to obtain a target evaluation model;

[0009] Obtain the vision training effect data to be tested, and input the vision training effect data into the target evaluation model, so that the target evaluation model performs feature processing on the vision to be tested to obtain the feature data to be tested, and then the target evaluation model classifies the feature data to be tested based on the classifier to obtain the vision training effect evaluation result.

[0010] In the above scheme, the sample vision training effect data is expanded through the data expansion model to improve the quantity and authenticity of the data used to train the initial evaluation model; based on the Monte Carlo algorithm and the preset decision tree model, a classifier is constructed, and the initial evaluation model is constructed based on the classifier, so as to explore more possible tree structures, so that the model structure of the constructed classifier has a more accurate vision training effect prediction ability, further improving the prediction accuracy of the initial evaluation model; the initial evaluation model is trained to obtain the target evaluation model, and then the vision training effect data to be tested is input into the target evaluation model to obtain a more accurate vision training effect evaluation result.

[0011] In this approach, the data augmentation model augments sample vision training effect data, increasing both the quantity and veracity of this data. Furthermore, the Monte Carlo algorithm selects decision tree structures from a pre-set extreme gradient boosting decision tree model. Compared to traditional decision tree structure selection methods that rely on greedy algorithms, this allows for a wider range of model structures to be explored, improving the model's generalization capabilities. All of these factors contribute to improving the accuracy of vision training effect evaluation.

[0012] Furthermore, the data augmentation model includes a generative adversarial network and a variational autoencoder, and the sample vision training effect data is input into a preset data augmentation model to obtain augmented data, including: performing data generation processing on the sample vision training effect data through the generative adversarial network to generate synthetic data; and performing automatic differentiation and variational inference on the synthetic data through the variational autoencoder to obtain the augmented data.

[0013] In the above scheme, the sample vision training effect data is processed by a generative adversarial network to generate synthetic data, thereby providing sufficient and real data for the subsequent training of the initial evaluation model. However, the synthetic data generated by the generative adversarial network may be limited to a certain category. Therefore, the synthetic data is automatically differentiated and inferred by the variational autoencoder to obtain expanded data that is more in line with the real data distribution.

[0014] Furthermore, the sample vision training effect data is input into a preset data expansion model to obtain expanded data, including: the training process of the generative adversarial network is: based on the sample vision training effect data, a preset initial generative adversarial network is trained to obtain the generative adversarial network, and then intermediate training data is obtained based on the generative adversarial network; the training process of the variational autoencoder is: based on the intermediate training data, a preset initial variational autoencoder is trained to obtain the variational autoencoder.

[0015] In the above scheme, since the purpose of the variational autoencoder is to automatically differentiate and perform variational inference on the synthetic data generated by the generative adversarial network, during the training process, the data preset for the initial variational autoencoder for training is also the intermediate training data output by the generative adversarial network, avoiding the separate training of data from different sources, ensuring the uniformity of the training process, and improving the reliability of the final generative adversarial network and variational autoencoder.

[0016] Furthermore, the preset initial generative adversarial network includes a generator and a discriminator, and the preset initial generative adversarial network is trained based on the sample vision training effect data to obtain the generative adversarial network, and then intermediate training data is obtained based on the generative adversarial network, including: initializing the parameters of the generator and the parameters of the discriminator; inputting a preset random noise vector into the generator to generate forged data; constructing the loss function of the discriminator based on the forged data, the sample vision training effect data and the parameters of the discriminator; constructing the loss function of the generator based on the random noise vector, the parameters of the generator and the parameters of the discriminator; updating the parameters of the discriminator based on the gradient of the loss function of the discriminator, so that the preset initial generative adversarial network updates the parameters of the generator based on the gradient of the loss function of the generator, thereby making the loss function of the generator and the loss function of the discriminator converge, and when the loss function of the generator and the loss function of the discriminator converge at the same time, stopping the updating of the parameters of the generator and the parameters of the discriminator to obtain the generative adversarial network.

[0017] In the above scheme, adversarial training is performed between the generator and the discriminator, and the two influence each other, so that the final trained generator and discriminator have high accuracy, and the generated intermediate training data is closer to the real data.

[0018] Furthermore, when the loss function of the generator and the loss function of the discriminator converge at the same time, the updating of the parameters of the generator and the parameters of the discriminator is stopped to obtain the generative adversarial network, including: constructing a dual loss function based on the loss function of the generator and the loss function of the discriminator; updating the parameters of the discriminator and the parameters of the generator in the dual loss function based on the gradient of the dual loss function, and stopping updating the parameters of the generator and the parameters of the discriminator when the dual loss function converges to obtain the generative adversarial network.

[0019] In the above scheme, a dual loss function is constructed based on the loss function of the generator and the loss function of the discriminator. That is, the dual loss function is used to balance the training intensity of the generator and the discriminator to prevent one side from being too strong and causing training imbalance, thereby obtaining a more accurate generative adversarial network, making the intermediate training data it outputs more diverse and similar to the real sample vision training effect data in terms of overall data distribution.

[0020] Furthermore, the preset initial variational autoencoder includes an encoder and a decoder, and the preset initial variational autoencoder is trained based on the intermediate training data to obtain the variational autoencoder, including: constructing a loss function of the preset initial variational autoencoder based on the intermediate training data, the parameters of the encoder, the parameters of the decoder and the KL divergence; updating the parameters of the encoder and the decoder respectively based on the gradient of the loss function of the preset initial variational autoencoder, and stopping the updating of the parameters of the encoder and the decoder when the loss function of the variational autoencoder converges to obtain the variational autoencoder.

[0021] In the above scheme, during the training of the preset initial variational autoencoder, the KL divergence constraint is used so that the variational autoencoder finally trained can generate expanded data whose distribution is closer to the real data distribution and covers the entire data space more evenly.

[0022] Furthermore, the method of constructing a classifier based on the Monte Carlo algorithm and a preset decision tree model, and constructing an initial evaluation model based on the classifier includes: initializing the preset decision tree model; performing upper confidence bound exploration on the sample vision training effect data and the vision training effect features in the expanded data through the Monte Carlo algorithm to obtain a target feature combination; performing segmentation point evaluation on the vision training effect features in the target feature combination through the Monte Carlo algorithm to obtain a target segmentation point; and updating the preset decision tree model based on the target feature combination and the target segmentation point to obtain the classifier.

[0023] In the above solution, because traditional decision tree models typically rely on greedy algorithms for structure selection, they neglect to explore more optimal model structures, limiting the model's generalization ability. Therefore, the above solution uses the Monte Carlo algorithm to explore more possible tree structures, thereby improving the model's generalization ability.

[0024] Furthermore, the preset decision tree model is updated based on the target feature combination and the target segmentation point to obtain the classifier, including: obtaining the pruning value of each decision tree node in the preset decision tree model based on a preset pruning algorithm; when the pruning value is less than a preset pruning threshold, the decision tree node is pruned to obtain the classifier.

[0025] In the above scheme, in order to prevent overfitting, each decision tree node in the preset decision tree model is pruned to remove branches that do not significantly improve the performance of the preset decision tree model, thereby removing branches that do not significantly improve the model performance, thereby improving the stability of the preset decision tree model.

[0026] Furthermore, the target evaluation model includes a feature extraction sub-model and a feature dimensionality reduction sub-model, and the target evaluation model is used to perform feature processing on the vision to be tested to obtain feature data to be tested, including: performing feature extraction on the vision training effect data to be tested by the feature extraction sub-model to obtain initial vision features; and performing feature dimensionality reduction on the initial vision features by the feature dimensionality reduction sub-model to obtain feature data to be tested.

[0027] In the above scheme, the feature extraction sub-model performs feature extraction on the vision training effect data to be tested to obtain initial vision features; the feature dimensionality reduction sub-model performs feature dimensionality reduction on the initial vision features to obtain feature data to be tested, thereby improving the quality and efficiency of feature representation, thereby improving the accuracy of subsequent evaluation and classification of the feature data to be tested by the classifier.

[0028] The present invention also provides a vision training effect evaluation system, comprising: a first acquisition module for acquiring sample vision training effect data; an input module for inputting the sample vision training effect data into a preset data expansion model to obtain expanded data; a construction module for constructing a classifier based on a Monte Carlo algorithm and a preset decision tree model, and constructing an initial evaluation model based on the classifier; a training module for training the initial evaluation model according to the expanded data and the sample vision training effect data to obtain a target evaluation model; an evaluation module for acquiring vision training effect data to be tested, and inputting the vision training effect data into the target evaluation model, so that the target evaluation model performs feature processing on the vision to be tested to obtain feature data to be tested, and then the target evaluation model classifies the feature data to be tested based on the classifier to obtain a vision training effect evaluation result.

[0029] In the system provided by the above solution, the data augmentation model expands the amount and authenticity of sample vision training effect data. A Monte Carlo algorithm is used to select decision tree structures from a preset extreme gradient boosting decision tree model. Compared to traditional decision tree structure selection methods that rely on greedy algorithms, this method can explore a wider range of model structures and improve the model's generalization ability. All of these contribute to improving the accuracy of vision training effect evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A schematic flow chart of a method for evaluating the effect of vision training provided by one embodiment of the present invention;

[0031] Figure 2 This is a schematic diagram of the architecture of a vision training effect evaluation system provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0032] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0033] See Figure 1 This embodiment provides a method for evaluating the effect of vision training, comprising the following steps:

[0034] Step S1: Obtaining sample vision training effect data;

[0035] Step S2: inputting the sample vision training effect data into a preset data expansion model to obtain expanded data;

[0036] Step S3: constructing a classifier based on the Monte Carlo algorithm and the preset decision tree model, and constructing an initial evaluation model based on the classifier;

[0037] Step S4: training the initial evaluation model based on the expanded data and the sample vision training effect data to obtain a target evaluation model;

[0038] Step S5: Obtain the vision training effect data to be tested, and input the vision training effect data into the target evaluation model, so that the target evaluation model performs feature processing on the vision to be tested to obtain the feature data to be tested, and then the target evaluation model classifies the feature data to be tested based on the classifier to obtain the vision training effect evaluation result.

[0039] In this embodiment, the sample vision training effect data is expanded through a data expansion model to improve the quantity and authenticity of the data used to train the initial evaluation model; based on the Monte Carlo algorithm and the preset decision tree model, a classifier is constructed, and an initial evaluation model is constructed based on the classifier, so as to explore more possible tree structures, so that the model structure of the constructed classifier has a more accurate vision training effect prediction ability, further improving the prediction accuracy of the initial evaluation model; the initial evaluation model is trained to obtain a target evaluation model, and then the vision training effect data to be tested is input into the target evaluation model to obtain a more accurate vision training effect evaluation result.

[0040] In this embodiment, the data augmentation model augments sample vision training effect data, increasing both the quantity and accuracy of the data. Furthermore, the Monte Carlo algorithm selects a decision tree structure from a pre-set extreme gradient boosting decision tree model. Compared to traditional decision tree structure selection methods that rely on greedy algorithms, this allows for a wider range of model structures to be explored, improving the model's generalization capabilities. All of these factors contribute to improving the accuracy of vision training effect assessments.

[0041] It should be noted that the sample vision training effect data can be derived from actual records of clinical amblyopia training and training data simulated under laboratory conditions, and the acquired data is stored in vector form. The features in the sample vision training effect data can be divided into multiple categories. For example, in the evaluation of vision training effect for amblyopia, a1 represents the vision level before training; a2 represents the vision level after training; a3 represents the duration of training; a4 represents the frequency of training; a5 represents the intensity of training; a6 represents the training method; a7 represents the age of the patient; a8 represents the gender of the patient; a9 represents the initial amblyopia degree of the patient; a1 represents the age of the patient; a10 represents the gender of the patient; a110 represents the gender of the patient; a120 represents the gender of the patient; a130 represents the gender of the patient; a140 represents the gender of the patient; a150 represents the gender of the patient; a160 represents the gender of the patient; a170 represents the gender of the patient; a180 represents the gender of the patient; a190 represents the gender of the patient; a1 .... 10 Indicates other auxiliary treatment methods, etc. The sample vision training effect data is manually labeled. For example, three evaluation categories are labeled: obvious training effect, average training effect, and insignificant training effect.

[0042] Furthermore, the data augmentation model includes a generative adversarial network and a variational autoencoder, and the sample vision training effect data is input into a preset data augmentation model to obtain augmented data, including:

[0043] Generate synthetic data by processing the sample vision training effect data through generative adversarial networks;

[0044] The augmented data is obtained by performing automatic differentiation and variational inference on the synthetic data through a variational autoencoder.

[0045] It should be noted that the sample vision training effect data is used to train the initial evaluation model, but the amount of sample vision training effect data obtained is insufficient for model training, which easily leads to poor generalization ability of the target evaluation model obtained by training, and at the same time affects the evaluation accuracy of the target evaluation model. Therefore, it is necessary to expand the data volume of the vision training effect data. The present application expands the vision training effect data through a data expansion model: the sample vision training effect data is input into a preset data expansion model, and the sample vision training effect data is processed by a generative adversarial network to generate synthetic data close to the real sample vision training effect data. The synthetic data is used together with the sample vision training effect data to train the initial evaluation model, which ensures both the adequacy of the data volume and the authenticity of the data. However, the data distribution of the synthetic data generated by the generative adversarial network is uncertain and may be limited to a certain category, which will lead to inaccurate evaluation of the target evaluation model finally constructed. Therefore, it is also necessary to perform automatic differentiation and variational inference on the synthetic data through a variational autoencoder to make the generated expanded data closer to the distribution of the real data, further improving the authenticity of the expanded data. Generative adversarial networks optimize the authenticity of generated synthetic data, while variational autoencoders optimize the data distribution matching of generated augmented data. Together, they improve the quality of augmented data.

[0046] Furthermore, the sample vision training effect data is input into a preset data expansion model to obtain expanded data, including:

[0047] The training process of the generative adversarial network is:

[0048] Training a preset initial generative adversarial network based on the sample vision training effect data to obtain a generative adversarial network, and then obtaining intermediate training data based on the generative adversarial network;

[0049] The training process of the variational autoencoder is:

[0050] The preset initial variational autoencoder is trained based on the intermediate training data to obtain a variational autoencoder.

[0051] It should be noted that the variational autoencoder processes the output of the generative adversarial network (GAN). Therefore, during training, the initial variational autoencoder needs to obtain intermediate training data based on the GAN. This intermediate training data is then used to train the initial variational autoencoder. The training of the GAN and the initial variational autoencoder is repeated until the preset stop condition is met, indicating that the data augmentation model training is complete.

[0052] Furthermore, the preset initial generative adversarial network includes a generator and a discriminator, and the preset initial generative adversarial network is trained based on the sample vision training effect data to obtain a generative adversarial network, and then intermediate training data is obtained based on the generative adversarial network, including:

[0053] Initialize the parameters of the generator and the discriminator;

[0054] Input the preset random noise vector into the generator to generate fake data;

[0055] Construct the loss function of the discriminator based on the fake data, the sample vision training effect data and the discriminator parameters;

[0056] Construct the generator’s loss function based on the random noise vector, the generator’s parameters, and the discriminator’s parameters.

[0057] Based on the gradient of the discriminator's loss function, the parameters of the discriminator are updated so that the preset initial generative adversarial network updates the parameters of the generator based on the gradient of the generator's loss function, thereby making the generator's loss function and the discriminator's loss function converge. When the generator's loss function and the discriminator's loss function converge at the same time, the updating of the generator's parameters and the discriminator's parameters is stopped to obtain a generative adversarial network.

[0058] It should be noted that the preset initial generative adversarial network includes a generator (G) and a discriminator (D), wherein the generator is responsible for generating forged data, and the discriminator is responsible for distinguishing between real data and forged data. During the training process, the generator is continuously optimized so that the generated forged data can deceive the discriminator, while the discriminator continuously improves its identification ability so that it can distinguish whether the forged data output by the generator is real data, that is, the generator and the discriminator are in adversarial training, and the two influence each other, which can ultimately make the generated data closer to the real data. Specifically: the parameters of the generator and the discriminator can be initialized with random weights, and then a preset random noise vector is input into the generator to generate forged data, wherein the random noise vector can be sampled from a Gaussian distribution. For example, the random noise vector can be expressed as:

[0059] bz~N(0,I);

[0060] Where bz is the random noise vector, N represents the normal distribution, 0 is the mean vector, and I is the identity matrix.

[0061] For example, the forged data can be expressed as:

[0062] bx=G(bz;b G );

[0063] Where bx is the generated fake data, G() is the existing generating function, bz is the random noise vector, b G For example, in the task of expanding sample vision training effect data for amblyopia, the generator can be used to simulate the changes in vision levels of different patients after training, and generate forged data similar to the sample vision training effect data with different training times, training methods, and auxiliary treatments.

[0064] It should be noted that the loss function of the discriminator is constructed based on the forged data, the sample vision training effect data and the parameters of the discriminator. For example, the loss function of the discriminator can be expressed as:

[0065]

[0066] Where, L D is the loss function of the discriminator, m is the batch size, D() is the existing discriminant function, bx real is the sample vision training effect data, bx is the generated fake data, b D is the parameter of the discriminator. During the training process, the dataset corresponding to the sample vision training effect data is divided into multiple small subsets, and the batch represents the subset number.

[0067] It should be noted that the loss function of the generator is constructed based on the random noise vector, the parameters of the generator and the parameters of the discriminator. For example, the loss function of the generator can be expressed as:

[0068]

[0069] Where, L G is the loss function of the generator, m is the batch size, D() is the discriminant function, G() is the generating function, bz is the random noise vector, b G is the parameter of the generator, b D are the parameters of the discriminator.

[0070] It should be noted that the parameters of the discriminator are updated based on the gradient of the loss function of the discriminator. For example, the parameters of the discriminator can be updated by back propagation, which is expressed as:

[0071]

[0072] Where η is the learning rate, is the gradient of the discriminator loss function. The network parameters are adjusted inversely according to the learning rate ratio to make the discrimination result more accurate.

[0073] It should be noted that based on the gradient of the generator's loss function, the parameters of the generator are updated. For example, the parameters of the generator can be updated by backpropagation, which is expressed as:

[0074]

[0075] Where η is the learning rate, is the gradient of the generator's loss function. Based on the error gradient fed back by the discriminator, the generator adjusts its own parameters inversely, making the generated fake data closer to the discriminator's assessment of the actual visual acuity training data. The discriminator and generator parameters are then continuously updated until the generator and discriminator loss functions converge. When both converge, updates to the generator and discriminator parameters cease, resulting in a generative adversarial network.

[0076] Furthermore, when the loss function of the generator and the loss function of the discriminator converge at the same time, the updating of the parameters of the generator and the discriminator is stopped, and a generative adversarial network is obtained, including:

[0077] Based on the loss function of the generator and the loss function of the discriminator, a dual loss function is constructed;

[0078] Based on the gradient of the dual loss function, the parameters of the discriminator and the generator in the dual loss function are updated. When the dual loss function converges, the updating of the parameters of the generator and the discriminator is stopped to obtain a generative adversarial network.

[0079] It should be noted that based on the loss function of the generator and the loss function of the discriminator, the constructed dual loss function can be expressed as:

[0080] L GD =L G +λ vg L D ;

[0081] Where, L GD is the joint loss function, L G is the loss function of the generator, L D is the loss function of the discriminator, λ vg is the balance coefficient. For example, λ vg Set to 0.3. Dual loss function L GD This is used when jointly updating the parameters of the generator and discriminator to balance their training intensity, preventing imbalanced training caused by excessive strength in one. By balancing the training of the generator and discriminator, a more accurate generative adversarial network is achieved, resulting in more diverse intermediate training data output. This not only closely resembles the overall distribution of real-world vision training data, but also aligns with the different vision training methods of different patient types.

[0082] Furthermore, the preset initial variational autoencoder includes an encoder and a decoder, and the preset initial variational autoencoder is trained based on the intermediate training data to obtain the variational autoencoder, including:

[0083] Based on the intermediate training data, the encoder parameters, the decoder parameters and the KL divergence, a loss function of the preset initial variational autoencoder is constructed;

[0084] Based on the gradient of the loss function of the preset initial variational autoencoder, the parameters of the encoder and the parameters of the decoder are updated respectively. When the loss function of the variational autoencoder converges, the updating of the parameters of the encoder and the decoder is stopped to obtain a variational autoencoder.

[0085] It should be noted that the preset initial variational autoencoder is used to perform automatic differentiation and variational inference on the intermediate training data output by the generative adversarial network. Through the KL divergence constraint of the variational autoencoder, the variational autoencoder finally trained can generate expanded data with a distribution closer to the real data distribution and more evenly covering the entire data space. Specifically, the loss function of the preset initial variational autoencoder can be expressed as:

[0086]

[0087] Where, L VAE To preset the initial variational autoencoder loss function, represents the existing expectation function, b enc is the encoder parameter, b dec is the parameter of the decoder, bx is the intermediate training data generated by the generator, q(b enc |bx) is the posterior distribution of the encoder, p(bx|b dec ) is the generation distribution of the decoder, D KL is the KL divergence, which is used to characterize the degree of difference between the posterior distribution of the encoder and the generative distribution of the decoder.

[0088] It should be noted that based on the gradient of the loss function of the preset initial variational autoencoder, the parameters of the encoder and the parameters of the decoder are updated respectively. When the loss function of the variational autoencoder converges, the updating of the parameters of the encoder and the decoder is stopped to obtain a variational autoencoder. Exemplarily, the parameters of the encoder and the decoder are updated by backpropagation, which can be expressed as:

[0089]

[0090] Where η bf is the learning rate, is the gradient of the variational autoencoder loss function with respect to the encoder parameters, is the gradient of the variational autoencoder loss function with respect to the decoder parameters. For example, η bf The encoder and decoder update parameters inversely at a fixed learning rate based on the variational loss gradient, so that the final generated data conforms to the original distribution while retaining key features.

[0091] Furthermore, based on the Monte Carlo algorithm and the preset decision tree model, a classifier is constructed, and an initial evaluation model is constructed based on the classifier, including:

[0092] Initialize the preset decision tree model;

[0093] The Monte Carlo algorithm is used to explore the upper confidence bound of the vision training effect features in the sample vision training effect data and the expanded data to obtain the target feature combination;

[0094] The Monte Carlo algorithm is used to evaluate the segmentation point of the vision training effect features in the target feature combination to obtain the target segmentation point;

[0095] Based on the target feature combination and target segmentation point, the preset decision tree model is updated to obtain a classifier.

[0096] It should be noted that traditional decision tree models usually rely on greedy algorithms in structural selection, thus ignoring the exploration of better model structures and limiting the generalization ability of the model. In this solution, the Monte Carlo algorithm is used to explore more possible tree structures, thereby improving the generalization ability of the model. Specifically, the parameters of the preset decision tree model are initialized. For example, the preset decision tree model can adopt an extreme gradient boosting decision tree model. The parameters of the preset decision tree model include the learning rate, the maximum depth of the tree, and the maximum number of trees. The preset decision tree model can be expressed as:

[0097] θ (0) =(λ q , Y q , α q ,σ q );

[0098] Where θ (0) is the preset decision tree model, λ q represents the learning rate; γ q Indicates the maximum depth of the tree; α q represents the sum of the minimum leaf node sample weights; σ q Indicates the maximum number of trees. Preferably, the learning rate λ of the extreme gradient boosting decision tree model q is set to 0.1, the maximum depth of the tree γ q Set to 6, the maximum number of trees σ q Set to 100.

[0099] It should be noted that since the sample vision training effect data and the expanded data have many vision training effect features, illustratively including features such as “vision level before and after training, training duration, training frequency, training method, and patient age”, it is necessary to screen the vision training effect features that are most valuable for evaluation and classification, thereby improving the evaluation accuracy of the initial evaluation model. In this application, the upper confidence bound of different vision training effect features in the sample vision training effect data and the expanded data is explored by the Monte Carlo algorithm to select the optimal target feature combination. Specifically, for each vision training effect feature, its contribution to the model's predictive ability is evaluated, and the feature combination that maximizes the model performance improvement is selected. The evaluation function of this process can be expressed as:

[0100]

[0101] Where U(bm) is the upper confidence bound of the vision training effect feature bm; represents the average improvement effect of the vision training effect feature bm in historical simulations; c is the preset exploration parameter used to balance exploration and utilization. In this task, since some vision training effect features, such as patient age, may have a greater impact on the training effect, the adjustment of the exploration parameter ensures that the preset decision tree model can not only utilize known effective vision training effect features, such as training time, but also try new vision training effect features, such as training intensity; N is the total number of simulations; n(bm) is the number of times the vision training effect feature bm is simulated.

[0102] Among them, historical simulation uses the contribution of vision training effect features in historical data for evaluation. For example, if "training frequency" has always had a greater impact on the results in past evaluation tasks, then its probability of being selected will be higher. The calculation method can be expressed as:

[0103]

[0104] Where ΔL j (bm) represents the loss reduction brought about by using the vision training effect feature bm as the segmentation feature in the jth simulation.

[0105] It should be noted that the evaluation and classification of vision training effects involves multiple continuous variables, such as vision changes, training time, etc. Therefore, how to find the best segmentation point is the key to determining the evaluation and classification effect. The Monte Carlo method simulates different segmentation points, calculates the expected total loss reduction, and selects the optimal segmentation scheme. For each decision tree in the preset decision tree model, the Monte Carlo method is used to determine the best segmentation point. In the amblyopia assessment data, the choice of segmentation point directly affects the evaluation and classification results. For example: if the "training time" exceeds a certain threshold, the possibility of vision improvement increases significantly, then a segmentation point can be set at this point; by calculating the loss reduction, the dividing point where vision improvement is significant is found to better predict the training effect. Therefore, the Monte Carlo algorithm is used to evaluate the segmentation point of the vision training effect features in the target feature combination to obtain the target segmentation point. Specifically, for each candidate segmentation point s, the expected total loss reduction is calculated, which can be expressed as:

[0106]

[0107] Where ΔL(s) is the expected total loss at split point s, G L and G R Represents the gradient sum of the left and right child nodes respectively; H L and H R Represents the second-order derivatives of the left and right child nodes respectively; λ q is the regularization parameter. The child nodes in the decision tree are the vision training effect features in the target feature combination. The target segmentation point is obtained by comparing the expected total loss with the preset threshold.

[0108] It should be noted that based on the target feature combination and target segmentation point, each decision tree in the preset decision tree model is updated. In each iteration, the loss of the decision tree in the current preset decision tree model is calculated, including considerations of classification error and model complexity. Furthermore, the parameters of the decision tree are adjusted to minimize this loss, thereby finding a balance between the accuracy and complexity of the decision tree. The calculation method of the loss function can be expressed as:

[0109]

[0110] Where, is the loss between the predicted value and the true value of the decision tree output in the preset decision tree model; Ω(θ) is the complexity of the current preset decision tree model; n is the number of samples input in the batch; λ qThis is a preset regularization coefficient used to adjust the balance between model complexity and goodness of fit. In the sample vision training effect data and augmented data, certain vision training effect features can cause the model to overfit. For example, age may appear as a strong feature in the training set but not stable in the test set. Therefore, the regularization parameter is used to control complexity and prevent overfitting.

[0111] In one embodiment, The calculation method of logical loss can be adopted, which can be expressed as:

[0112]

[0113] Where y i is the actual label of the i-th data point; is the probability predicted by the decision tree, calculated using the logistic function.

[0114] In this embodiment, the model complexity Ω(θ) includes the number of trees and the depth of the trees in the preset decision tree model, and can be calculated as follows:

[0115]

[0116] Where T is the total number of trees in the preset decision tree model; ω t is the vector of leaf node values of the t-th tree; γ q and λ q are parameters that control the complexity of the tree.

[0117] It should be noted that after training each decision tree in the preset decision tree model, all constructed decision trees are fused, and the outputs of all decision trees are weighted fused in a weighted manner, with the weight w t The calculation can be expressed as:

[0118] w t =exp(-α q ·R(t));

[0119] Where exp() is an exponential function with a natural constant as the base; R(t) is the loss of decision tree t on the validation set; α q is the weight attenuation factor. Preferably, the weight attenuation factor α q Set to 0.9. The calculation method of R(t) can be expressed as:

[0120]

[0121] Where G t is the sum of all sample gradients of the node; H t is the sum of the second-order derivatives of all samples at the node, λ qis the preset regularization parameter.

[0122] Furthermore, based on the target feature combination and the target segmentation point, the preset decision tree model is updated to obtain a classifier, including:

[0123] Based on a preset pruning algorithm, obtain the pruning value of each decision tree node in a preset decision tree model;

[0124] When the pruning value is less than the preset pruning threshold, the decision tree nodes are pruned to obtain a classifier.

[0125] It should be noted that after updating the parameters of each decision tree in the preset decision tree model and subsequently training it, the preset decision tree model can be pruned, and then all constructed decision trees can be fused. Because in vision training effectiveness evaluation tasks, the sample vision training effectiveness evaluation data and augmented data used for training often contain noise. For example, the unique treatment methods of individual patients may interfere with model learning. Therefore, post-pruning is necessary to remove branches that affect the stability of the decision tree model. To prevent overfitting, the generated decision tree is post-pruned. Based on the performance of the validation data, branches that do not significantly improve the performance of the preset decision tree model are removed. A preset pruning algorithm can be used to obtain the pruning value of each decision tree node in the preset decision tree model. The pruning algorithm is the formula for obtaining R(t) described above. The obtained R(t) is used as the pruning value, and δ is set as the preset pruning threshold. For example, the preset pruning threshold δ is set to 0.01. If R(t) is less than the set threshold δ, the node is pruned.

[0126] Furthermore, the target evaluation model includes a feature extraction sub-model and a feature dimension reduction sub-model. The target evaluation model is used to perform feature processing on the vision to be measured to obtain feature data to be measured, including:

[0127] The feature extraction sub-model is used to extract features from the visual acuity training effect data to be tested, thereby obtaining initial visual acuity features;

[0128] The feature dimension reduction sub-model is used to reduce the initial visual features to obtain the feature data to be tested.

[0129] It should be noted that the feature extraction sub-model can adopt a method of dynamically adjusting the network topology structure, allowing the weights and biases to be adjusted during the learning process, thereby improving the adaptability and training stability of the feature extraction sub-model. The trained feature extraction sub-model is used to extract features from the vision training effect data to obtain the initial vision features. The feature dimensionality reduction sub-model can adopt an autoencoder network with a micro-adjustment strategy, which not only reconstructs the input data, but also optimizes the matching degree between the encoded features and the preset low-dimensional target features, thereby improving the quality and efficiency of feature representation. The trained feature dimensionality reduction sub-model is used to reduce the initial vision features to obtain the feature data to be tested. The vision training effect is evaluated using the trained target evaluation model. In one embodiment, for a newly collected piece of vision training effect data to be tested, feature extraction is first performed using the trained feature extraction sub-model. Further, the extracted initial vision features are input into the feature dimensionality reduction model for feature dimensionality reduction. Further, the reduced feature data to be tested is input into a classifier for classification. For example, the classification categories include: obvious training effect, general training effect, and insignificant training effect, for a total of three categories.

[0130] See Figure 2 This embodiment also provides a vision training effect evaluation system, including: a first acquisition module, used to obtain sample vision training effect data; an input module, used to input the sample vision training effect data into a preset data expansion model to obtain expanded data; a construction module, used to construct a classifier based on the Monte Carlo algorithm and a preset decision tree model, and to construct an initial evaluation model based on the classifier; a training module, used to train the initial evaluation model according to the expanded data and the sample vision training effect data to obtain a target evaluation model; an evaluation module, used to obtain the vision training effect data to be tested, and input the vision training effect data into the target evaluation model, so that the target evaluation model performs feature processing on the vision to be tested to obtain feature data to be tested, and then the target evaluation model classifies the feature data to be tested based on the classifier to obtain a vision training effect evaluation result.

[0131] The system provided in this embodiment can effectively implement the aforementioned vision training effectiveness evaluation method. It uses a data augmentation model to expand sample vision training effectiveness data, increasing both the quantity and accuracy of this data. Furthermore, it uses a Monte Carlo algorithm to select a decision tree structure from a preset extreme gradient boosting decision tree model. Compared to traditional decision tree structure selection methods that rely on greedy algorithms, this method can explore a wider range of model structures and improve the model's generalization capabilities. All of these contribute to improving the accuracy of vision training effectiveness evaluation.

[0132] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for evaluating the effect of vision training, characterized in that: include: Obtain sample vision training effect data; Inputting the sample vision training effect data into a preset data expansion model to obtain expanded data; Constructing a classifier based on a Monte Carlo algorithm and a preset decision tree model, and constructing an initial evaluation model based on the classifier; Training the initial evaluation model according to the expanded data and the sample vision training effect data to obtain a target evaluation model; Obtain the vision training effect data to be tested, and input the vision training effect data into the target evaluation model, so that the target evaluation model performs feature processing on the vision to be tested to obtain the feature data to be tested, and then the target evaluation model classifies the feature data to be tested based on the classifier to obtain the vision training effect evaluation result.

2. A method for evaluating the effect of vision training according to claim 1, characterized in that: The data expansion model includes a generative adversarial network and a variational autoencoder, and the sample vision training effect data is input into a preset data expansion model to obtain expanded data, including: Performing data generation processing on the sample vision training effect data through the generative adversarial network to generate synthetic data; Automatic differentiation and variational inference are performed on the synthetic data by the variational autoencoder to obtain the augmented data.

3. A method for evaluating the effect of vision training according to claim 2, characterized in that: The step of inputting the sample vision training effect data into a preset data expansion model to obtain expanded data includes: The training process of the generative adversarial network is as follows: Training a preset initial generative adversarial network based on the sample vision training effect data to obtain the generative adversarial network, and then obtaining intermediate training data based on the generative adversarial network; The training process of the variational autoencoder is: A preset initial variational autoencoder is trained based on the intermediate training data to obtain the variational autoencoder.

4. A method for evaluating the effect of vision training according to claim 3, characterized in that: The preset initial generative adversarial network includes a generator and a discriminator. The training of the preset initial generative adversarial network based on the sample vision training effect data to obtain the generative adversarial network, and further obtaining intermediate training data based on the generative adversarial network, includes: Initialize the parameters of the generator and the parameters of the discriminator; Inputting a preset random noise vector into the generator to generate forged data; Constructing a loss function of the discriminator based on the forged data, the sample vision training effect data, and the parameters of the discriminator; Constructing a loss function of the generator based on the random noise vector, parameters of the generator, and parameters of the discriminator; Based on the gradient of the loss function of the discriminator, the parameters of the discriminator are updated so that the preset initial generative adversarial network updates the parameters of the generator based on the gradient of the loss function of the generator, thereby making the loss function of the generator and the loss function of the discriminator converge. When the loss function of the generator and the loss function of the discriminator converge at the same time, the updating of the parameters of the generator and the parameters of the discriminator is stopped to obtain the generative adversarial network.

5. A method for evaluating the effect of vision training according to claim 4, characterized in that: When the loss function of the generator and the loss function of the discriminator converge simultaneously, the updating of the parameters of the generator and the parameters of the discriminator is stopped to obtain the generative adversarial network, including: Constructing a dual loss function based on the loss function of the generator and the loss function of the discriminator; Based on the gradient of the dual loss function, the parameters of the discriminator and the parameters of the generator in the dual loss function are updated. When the dual loss function converges, the updating of the parameters of the generator and the parameters of the discriminator is stopped to obtain the generative adversarial network.

6. A method for evaluating the effect of vision training according to claim 3, characterized in that: The preset initial variational autoencoder includes an encoder and a decoder, and the training of the preset initial variational autoencoder based on the intermediate training data to obtain the variational autoencoder includes: Constructing a loss function of the preset initial variational autoencoder based on the intermediate training data, the parameters of the encoder, the parameters of the decoder, and the KL divergence; Based on the gradient of the loss function of the preset initial variational autoencoder, the parameters of the encoder and the parameters of the decoder are updated respectively. When the loss function of the variational autoencoder converges, the updating of the parameters of the encoder and the parameters of the decoder is stopped to obtain the variational autoencoder.

7. A method for evaluating the effect of vision training according to claim 1, characterized in that: The method of constructing a classifier based on the Monte Carlo algorithm and the preset decision tree model, and constructing an initial evaluation model based on the classifier includes: Initializing the preset decision tree model; Performing upper confidence bound exploration on the sample vision training effect data and the vision training effect features in the expanded data by using a Monte Carlo algorithm to obtain a target feature combination; Performing segmentation point evaluation on the vision training effect feature in the target feature combination by using a Monte Carlo algorithm to obtain a target segmentation point; Based on the target feature combination and the target segmentation point, the preset decision tree model is updated to obtain the classifier.

8. A method for evaluating the effect of vision training according to claim 7, characterized in that: The updating of the preset decision tree model based on the target feature combination and the target segmentation point to obtain the classifier includes: Based on a preset pruning algorithm, obtaining a pruning value for each decision tree node in the preset decision tree model; When the pruning value is less than a preset pruning threshold, the decision tree nodes are pruned to obtain the classifier.

9. A method for evaluating the effect of vision training according to claim 1, characterized in that: The target evaluation model includes a feature extraction sub-model and a feature dimension reduction sub-model. The target evaluation model is used to perform feature processing on the visual acuity to be measured to obtain feature data to be measured, including: Performing feature extraction on the vision training effect data to be tested using the feature extraction sub-model to obtain initial vision features; The feature dimension reduction sub-model is used to perform feature dimension reduction on the initial vision feature to obtain feature data to be measured.

10. A vision training effect evaluation system, characterized in that: include: A first acquisition module is used to obtain sample vision training effect data; An input module, configured to input the sample vision training effect data into a preset data expansion model to obtain expanded data; A construction module is used to construct a classifier based on a Monte Carlo algorithm and a preset decision tree model, and to construct an initial evaluation model based on the classifier; A training module, configured to train the initial evaluation model based on the expanded data and the sample vision training effect data to obtain a target evaluation model; The evaluation module is used to obtain the vision training effect data to be tested, and input the vision training effect data into the target evaluation model, so that the target evaluation model performs feature processing on the vision to be tested to obtain the feature data to be tested, and then the target evaluation model classifies the feature data to be tested based on the classifier to obtain the vision training effect evaluation result.

Images