A method and system for establishing lens recognition model

Abstract

The present invention relates to the field of graphic analysis technology, and in particular to a method and system for establishing a lens recognition model. The method comprises the following steps: obtaining an original multi-source lens data set; performing data enhancement processing on the original multi-source lens data set to generate an original enhanced lens data set; performing data feature extraction on the original enhanced lens data set to generate original lens feature data; performing multi-dimensional cross-domain feature fusion on the original lens feature data to generate a multi-dimensional cross-domain fusion data set for the lens; performing integrated learning based on the multi-dimensional cross-domain fusion data set for the lens, and performing model construction to generate an integrated lens recognition model; therefore, the present invention solves the deficiencies of traditional lens recognition models in feature processing and accuracy through multi-source data enhancement, feature fusion and integrated learning strategies, and improves the accuracy, robustness and generalization ability of the recognition model.

Description

Technical Field

[0001] The present invention relates to the field of graphic analysis technology, and in particular to a method and system for establishing a lens recognition model. Background Art

[0002] Traditional methods mostly rely on manual feature extraction and shallow learning, which makes it difficult to effectively process complex lens features such as surface texture, optical properties, and temperature refractive index. These technologies are usually unable to fully integrate the intrinsic relationships of multi-source data, resulting in insufficient recognition accuracy and generalization ability. In addition, traditional data enhancement methods are limited to image transformation and fail to effectively solve the problem of data scarcity. Although ensemble learning has applications in lens recognition, existing technologies fail to fully tap the synergy of each base learner, especially when the lens features are highly heterogeneous, the performance of the ensemble model is still limited. Furthermore, the evaluation and report generation mechanism is too static and lacks dynamic feedback, and cannot provide real-time support for model optimization. Overall, existing technologies have failed to break through existing bottlenecks in multi-source data processing, feature fusion, and model optimization, affecting the performance and reliability of lens recognition systems. Summary of the invention

[0003] Based on this, it is necessary to provide a method and system for establishing a lens recognition model to solve at least one of the above technical problems.

[0004] To achieve the above object, a method for establishing a lens recognition model is provided, the method comprising the following steps:

[0005] Step S1: obtaining an original multi-source lens data set; performing data enhancement processing on the original multi-source lens data set to generate an original enhanced lens data set;

[0006] Step S2: extracting data features from the original enhanced lens data set to generate original lens feature data; performing multi-dimensional cross-domain feature fusion on the original lens feature data to generate a multi-dimensional cross-domain fusion data set for the lens;

[0007] Step S3: performing integrated learning based on the lens multi-dimensional cross-domain fusion dataset, and building a model to generate a lens integrated recognition model;

[0008] Step S4: Perform model evaluation on the lens integrated recognition model and generate a report, thereby completing the operation of establishing the lens recognition model.

[0009] The beneficial effect of the present invention is that by acquiring the original multi-source data set of the lens and performing data enhancement processing, the original enhanced data set of the lens is generated, which provides more abundant and diverse training data for subsequent model training. This enhancement process not only expands the number of samples, but also enhances the robustness of the model by introducing diversified data transformations, so that a high recognition performance can be maintained when facing different data variations. Then, by extracting features from the enhanced data set, the key information of the lens is captured, especially in terms of optical performance and physical properties. Through multi-dimensional cross-domain feature fusion, the model can integrate features from multiple fields, such as optical performance, temperature response, etc., improve the richness of data representation and the recognition ability of the model, thereby enhancing the adaptability of the model to complex data patterns. This process effectively improves the semantic level and relevance of the data by fusing the features of different data sources, and significantly improves the accuracy of recognition and the stability of the model. Through the ensemble learning method, the advantages of multiple base learners are combined to further improve the recognition ability of the model. Ensemble learning not only reduces the risk of overfitting, but also enhances the robustness of the system through the complementarity of multiple models. Finally, the model was rigorously evaluated using a variety of evaluation indicators to ensure its effectiveness in practical applications, and detailed feedback was provided for model improvement and optimization through report generation. This systematic process not only improves the accuracy of lens recognition, but also ensures the generalization ability of the model on different data sets, solving the challenges of traditional methods in data scarcity, feature diversity, and insufficient model generalization ability.

[0010] Preferably, step S1 comprises the following steps:

[0011] Step S11: Acquire original multi-source lens data set;

[0012] Step S12: preprocessing the original multi-source lens data set to generate multi-source preprocessed lens data;

[0013] Step S13: Perform data enhancement processing based on the lens multi-source pre-processed data to generate a lens original enhanced data set.

[0014] The present invention provides a variety of input sources for subsequent analysis by acquiring the original multi-source data set of the lens, ensuring that the data covers the performance of the lens under different working conditions. These data come from different sensors, experimental conditions or external environmental factors, and have strong diversity and complexity. Secondly, by preprocessing the original multi-source data set of the lens to generate multi-source preprocessed data of the lens, the problems of noise, missing values, outliers, etc. in the original data are solved, and the validity of the data in statistics and model building is ensured. Data preprocessing not only cleans and optimizes the data quality, but also quantifies and unifies the data through technical means such as normalization and standardization, providing a solid foundation for the next step of feature extraction and analysis. In addition, the preprocessed data not only removes redundant information and eliminates deviations, but also enhances the consistency and representativeness of the data, effectively avoiding the interference of data noise on the model building process. The multi-source preprocessed data of the lens is processed by data enhancement to generate the original enhanced data set of the lens. Data enhancement is to expand the number of data samples by performing different transformations (such as rotation, scaling, flipping, etc.) on the data set, thereby improving the generalization ability of the model. The enhanced dataset not only increases the sample space, but also introduces a variety of perspectives and deformation methods, further improving the model's ability to recognize lens features and robustness in practical applications. Data enhancement also effectively alleviates the problem of data imbalance and ensures the model's recognition performance on scarce categories. Overall, through data preprocessing and enhancement processing, not only the quality and structure of the dataset are optimized, but also its adaptability to subsequent deep learning models and analysis techniques is improved, providing more reliable and representative training samples for model training, thereby ensuring the efficiency and accuracy of the entire lens recognition system.

[0015] Preferably, step S13 includes the following steps:

[0016] Step S131: classifying the lens multi-source pre-processed data into scarce categories to generate lens multi-source classified data, wherein the lens multi-source classified data includes lens multi-source scarce classified data and lens multi-source sufficient classified data;

[0017] Step S132: performing generative adversarial network processing on the lens multi-source scarce classification data to generate lens adversarial network generated data; performing variational autoencoder transformation on the lens adversarial network generated data to generate lens scarce similarity data; performing synthetic minority class oversampling processing on the lens adversarial network generated data and the lens scarce similarity data to generate lens scarce interpolation sample data; performing synthetic enhancement processing on the lens scarce interpolation sample data to generate lens scarce sample synthetic enhanced data;

[0018] Step S133: performing image flipping processing on the lens multi-source sufficient classification data to generate lens flip sufficient data; performing color transformation on the lens flip sufficient data to generate lens color transformation data;

[0019] Step S134: merging the lens scarce sample synthesis enhancement data and the lens color transformation data to generate a lens original enhancement data set.

[0020] The present invention classifies the multi-source pre-processed lens data into scarce categories and divides the data into scarce categories and sufficient categories, so as to focus on the scarce categories in the data set. This process provides a clear direction for subsequent processing by identifying the category imbalance problem in the data, ensuring that all types of data are reasonably used in training. Secondly, the generative adversarial network (GAN) technology is used to process the multi-source scarce classification data of the lens. The generative adversarial network enhances the data of the scarce categories by generating realistic samples, thereby improving the model's learning ability for the scarce categories. The generated scarce samples are further transformed by the variational autoencoder (VAE) transformation to increase their diversity and simulate more complex scarce data features. Then, the synthetic minority class oversampling (SMOTE) and ADASYN enhancement technology are used to generate sample data of the scarce categories through interpolation and synthesis methods, thereby improving the recognition accuracy and generalization ability of the model for these categories. This series of enhancement means not only alleviates the category imbalance problem, but also expands the training sample space of the scarce categories and enhances the robustness of the model in the unbalanced data set. In addition, by flipping and color transforming the lens multi-source sufficient classification data, the diversity and robustness of the sufficient category data are further enhanced. These transformation methods increase the richness of the data set by simulating different variations in the data, thereby improving the model's ability to recognize sufficient categories. Finally, step S134 generates the original lens enhanced data set by merging the scarce sample synthetic enhancement data and the sufficient category color transformation data. This step ensures the balance of the final data set in different categories, enhances the diversity and representativeness of the data, and provides more comprehensive, balanced and challenging training samples for subsequent model training.

[0021] Preferably, step S2 comprises the following steps:

[0022] Step S21: extracting data features from the original enhanced lens data set to generate original lens feature data;

[0023] Step S22: performing t-SNE dimension reduction according to the original feature data of the lens to generate original feature dimension reduction data of the lens, wherein the original feature dimension reduction data of the lens includes lens surface optical performance data and lens temperature-refractive index performance data;

[0024] Step S23: Perform multi-dimensional cross-domain feature fusion based on the original feature dimensionality reduction data of the lens to generate a multi-dimensional cross-domain fusion data set of the lens.

[0025] The present invention generates original feature data of the lens through data feature extraction from the original enhanced data set of the lens, providing a more accurate representation for subsequent analysis. Feature extraction helps capture the key attributes of the lens, such as optical performance, refractive index, etc., by converting the original data into high-dimensional and structured feature data, thereby improving the expression quality and operability of the data. This step ensures that all potential important information is fully extracted when processing complex lens data. Then, by performing t-SNE dimensionality reduction on the original feature data of the lens, the original feature dimensionality reduction data of the lens is generated, especially the surface optical performance and temperature-refractive index performance data. This dimensionality reduction process reduces high-dimensional data to a lower dimension through nonlinear mapping, greatly reducing the redundancy of the data, while maintaining the inherent structure and important features of the original data, so that the data does not lose the validity of its information while reducing the dimension. Through t-SNE dimensionality reduction, the complex features of the lens are effectively simplified, which facilitates subsequent analysis and modeling, and improves the efficiency of data processing. Finally, by performing multi-dimensional cross-domain feature fusion based on the original feature dimensionality reduction data of the lens, a multi-dimensional cross-domain fusion data set of the lens is generated. This process fuses the features of different data domains and realizes cross-domain collaboration of data. By integrating multi-dimensional features such as surface optical properties and temperature-refractive index properties, the model can effectively learn in a unified feature space, thereby enhancing the expressiveness of the data set and the recognition ability of the model. This cross-domain fusion not only expands the semantic level of the data, but also improves the comprehensive expression of the data, making the inherent correlation between lens features more prominent, and can better support subsequent machine learning or pattern recognition tasks.

[0026] Preferably, step S21 includes the following steps:

[0027] Step S211: performing texture analysis on the original enhanced data set of the lens to generate lens surface texture characteristics; performing shape outlining based on the lens surface texture characteristics to generate lens overall shape data; performing stereo model construction on the lens surface texture characteristics and the lens overall shape data to generate a three-dimensional model of the lens;

[0028] Step S212: performing an accelerated robust feature algorithm analysis on the three-dimensional model of the lens to generate local feature data of the lens; performing a microstructure analysis on the local feature data of the lens to generate microstructure data of the lens;

[0029] Step S213: coupling the original thermal-optical characteristics of the lens microstructure data to generate original characteristic data of the lens.

[0030] The present invention generates the surface texture characteristics of the lens by performing texture analysis on the original enhanced data set of the lens. The texture analysis technology can extract subtle surface structure and texture information from the surface data of the lens, providing valuable basic data for subsequent shape recognition and performance modeling. Then, the shape is outlined based on the surface texture characteristics of the lens to generate the overall shape data of the lens. This process accurately depicts the macroscopic outline and external shape of the lens through the extraction of geometric features, laying the foundation for subsequent three-dimensional modeling and further analysis. Further, the surface texture characteristics of the lens and the overall shape data are used to construct a three-dimensional model to generate a three-dimensional model of the lens, and the three-dimensional structure of the lens is accurately represented. This step integrates the surface texture and shape features through the three-dimensional modeling technology, and can provide a more comprehensive and accurate data view to ensure that the subsequent analysis and simulation are more in line with reality. Then, the three-dimensional model of the lens is analyzed by an accelerated robust feature algorithm to generate local feature data of the lens. This step identifies the local key features of the lens, such as the tiny geometric deformation and complex structure of the surface, through an efficient and robust feature extraction algorithm, providing more fine-grained information for the refined analysis of the lens. In addition, the microstructure analysis of the local feature data of the lens was performed to generate the lens microstructure data, which further penetrated into the microscopic level of the lens and captured the material details and its microscopic properties, such as granularity, lattice structure and other characteristics, which are of great value for lens performance optimization. Finally, the original feature data of the lens was generated by coupling the lens microstructure data with the original thermal-optical properties. This process integrates the physical properties of the lens under different environmental conditions, such as the effect of temperature changes on the refractive index, and can provide an important reference for the performance prediction of the lens in the actual use environment.

[0031] Preferably, step S23 includes the following steps:

[0032] Step S231: Calculate the mean of the optical performance data on the lens surface to generate the optical mean data on the lens surface; perform spatial mapping on the optical mean data on the lens surface to generate the optical spatial mapping data on the lens surface;

[0033] Step S232: performing time fluctuation analysis on the lens temperature-refractive index performance data, and removing abnormal data to generate lens temperature-refractive index coupling data; performing lens coupling data space mapping on the lens temperature-refractive index coupling data to generate lens photothermal space mapping data;

[0034] Step S233: perform feature matrix splicing on the lens surface optical space mapping data and the lens photothermal space mapping data to generate a multi-dimensional feature matrix of the lens; perform weighted calculation on the multi-dimensional feature matrix of the lens, and simultaneously perform multi-dimensional cross-domain feature fusion of the lens to generate a multi-dimensional cross-domain fusion data set of the lens.

[0035] Preferably, step S3 comprises the following steps:

[0036] Step S31: performing base learning training based on the lens multi-dimensional cross-domain fusion data set to generate lens base learning training data;

[0037] Step S32: performing sampling with replacement on the lens-based learning training data to generate a lens multi-subdataset; performing stacked logistic regression based on the lens multi-subdataset to generate lens multi-based learner analysis data;

[0038] Step S33: Perform hyperparameter tuning on the lens multi-base learner analysis data to generate lens model hyperparameter data; construct a model based on the lens multi-base learner analysis data, and perform Bayesian verification using the lens model hyperparameter data to generate a lens integrated recognition model.

[0039] The present invention generates lens base learning training data by performing base learning training based on a multi-dimensional cross-domain fusion data set of lenses. This step constructs a multi-dimensional data set by fusing data from different dimensions, ensuring that all types of features in the learning process can fully reflect the multivariate performance of the lens. On this basis, the base learning training process can extract important patterns and relationships in the data, providing a rich information basis for subsequent model training. Next, the lens base learning training data is sampled with replacement to generate a lens multi-subdata set. Through the sampling technology with replacement (such as the bootstrap method), samples can be repeatedly extracted from the original training data to generate multiple different sub-data sets. This process enhances the diversity of data samples, avoids the risk of overfitting, and provides data support for the construction of multi-base learners. Based on these sub-data sets, stacked logistic regression is used to generate lens multi-base learner analysis data. Stacked logistic regression is an effective integrated learning method. It combines the outputs of multiple base learners to obtain a more powerful prediction model, effectively improving the prediction accuracy and generalization ability of the model. Hyperparameter tuning is performed on the lens multi-base learner analysis data to generate lens model hyperparameter data. Hyperparameter tuning improves the predictive performance of the model by searching for the optimal combination of parameters, ensuring that the model can perform optimally under different conditions. This process makes the learning model more refined and can make more accurate predictions based on the complexity and diversity of the data. Finally, the model is constructed based on the lens multi-base learner analysis data, and Bayesian verification is performed using the lens model hyperparameter data to generate a lens integrated recognition model. Bayesian verification provides further performance evaluation for the model through probabilistic reasoning, ensuring the reliability and accuracy of the ultimately generated integrated recognition model in practical applications. Through this series of steps, the present invention successfully improves the modeling capabilities of lens data, solves the shortcomings of traditional methods in data diversity, model training, and prediction accuracy, and enhances the robustness and practicality of the integrated learning model.

[0040] Preferably, step S32 includes the following steps:

[0041] Step S311: performing sampling with replacement on the lens-based learning training data to generate a lens multi-sub-dataset;

[0042] Step S312: performing vector machine base learner prediction based on the lens multi-subdatasets to generate lens stacking base learner data; performing meta-learner training on the lens stacking base learner data, and performing regularization processing to generate lens regularized data;

[0043] Step S313: Perform stacked logistic regression on the lens regularization data to generate lens multi-base learner analysis data.

[0044] The present invention generates a lens multi-subdataset by sampling with replacement on the lens base learning training data. This process utilizes sampling techniques such as the bootstrap method, so that the original data set can generate multiple different data subsets, thereby enhancing the diversity and representativeness of the training data. In this way, the model can learn more potential information from different sub-datasets, reduce data deviation to a certain extent, and improve the generalization ability of the data. Then, the vector machine base learner prediction is performed based on the lens multi-subdataset to generate lens stacking base learner data. Support vector machine (SVM) is a powerful supervised learning method. By classifying and regressing the data, it can effectively extract boundary information in the data and provide high-quality basic data for subsequent stacking learning. At the same time, the outputs of multiple independent base learners are combined by stacking the base learner, which effectively improves the performance and accuracy of the model. Subsequently, the lens stacking base learner data is trained by the meta-learner and regularized to generate lens regularized data. The meta-learner further optimizes the prediction results by training on the outputs of multiple learners. Regularization avoids the problem of overfitting by introducing penalty terms during the training process, so that the learner can still maintain strong generalization ability when facing complex data, thereby improving the applicability of the model in different scenarios. Stacked logistic regression is performed on the lens regularization data to generate lens multi-base learner analysis data. As an integration method, stacked logistic regression effectively integrates the outputs of multiple base learners to provide a more accurate and stable prediction result for the final model. Through stacking technology, the advantages of different models are complemented to form a more efficient and accurate comprehensive model. This process significantly improves the accuracy and robustness of the lens recognition model, enabling it to maintain a high level of prediction performance in complex environments.

[0045] Preferably, step S4 comprises the following steps:

[0046] Step S41: performing model evaluation on the lens integrated recognition model to generate lens integrated recognition model evaluation data;

[0047] Step S42: generating a report on the lens integrated recognition model evaluation data to generate a lens integrated recognition model report;

[0048] Step S43: Perform data report visualization processing on the lens integrated recognition model report, thereby completing the operation of establishing the lens recognition model.

[0049] The present invention generates lens integrated recognition model evaluation data by evaluating the lens integrated recognition model. This step systematically analyzes the performance of the model in terms of accuracy, stability and generalization ability by evaluating the performance of the model on different data sets. By calculating various performance indicators of the model (such as accuracy, recall rate, F1 value, etc.), the evaluation data can provide key feedback information for model optimization, help researchers understand the performance of the model when facing different feature data, and provide a basis for subsequent adjustments. Next, a report is generated for the lens integrated recognition model evaluation data to generate a lens integrated recognition model report. This process converts the results of the model evaluation into a systematic and structured data report, so that researchers can intuitively understand the performance of the model in different test scenarios. The report not only covers various evaluation indicators of the model, but also presents the advantages and disadvantages of the model through data visualization, thereby providing a scientific basis for subsequent improvements. Finally, the lens integrated recognition model report is visualized, thereby completing the operation of establishing the lens recognition model. Through data report visualization, various evaluation results in the report are presented in the form of charts, curves, heat maps, etc., making the interpretation of the report more intuitive and easy to understand. This visualization facilitates rapid analysis of complex assessment results, helping researchers make decisions in the shortest possible time, while also providing non-specialists with easy-to-understand analysis results.

[0050] In this specification, a system for establishing a lens recognition model is provided, which is used to execute the above-mentioned method for establishing a lens recognition model. The system for establishing a lens recognition model includes:

[0051] The data acquisition and enhancement module is used to acquire the original multi-source data set of the lens; perform data enhancement processing on the original multi-source data set of the lens to generate the original enhanced data set of the lens;

[0052] The feature extraction and feature fusion module is used to extract data features from the original enhanced data set of the lens to generate the original feature data of the lens; perform multi-dimensional cross-domain feature fusion on the original feature data of the lens to generate a multi-dimensional cross-domain fusion data set of the lens;

[0053] The integrated learning and model building module is used to perform integrated learning and model building based on the multi-dimensional cross-domain fusion data set of lenses to generate an integrated lens recognition model;

[0054] The model evaluation and report generation module is used to evaluate the lens integrated recognition model and generate a report, thereby completing the task of establishing the lens recognition model. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A schematic diagram of a method for establishing a lens recognition model;

[0056] Figure 2 for Figure 1 Detailed implementation steps of step S2 in the flowchart;

[0057] Figure 3 for Figure 1 Detailed implementation steps of step S3 in FIG.

[0058] Figure 4 for Figure 1 Detailed implementation steps of step S4 in FIG.

[0059] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0060] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0061] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0062] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0063] To achieve this, please refer to Figures 1 to 4 , a method for establishing a lens recognition model, the method comprising the following steps: step S1: obtaining an original multi-source lens data set; performing data enhancement processing on the original multi-source lens data set to generate an original enhanced lens data set;

[0064] Step S2: extracting data features from the original enhanced lens data set to generate original lens feature data; performing multi-dimensional cross-domain feature fusion on the original lens feature data to generate a multi-dimensional cross-domain fusion data set for the lens;

[0065] Step S3: performing integrated learning based on the lens multi-dimensional cross-domain fusion dataset, and building a model to generate a lens integrated recognition model;

[0066] Step S4: Perform model evaluation on the lens integrated recognition model and generate a report, thereby completing the operation of establishing the lens recognition model.

[0067] In the embodiment of the present invention, reference Figure 1 FIG. 1 is a schematic flow chart of the steps of a method for establishing a lens recognition model according to the present invention. In this example, the method for establishing a lens recognition model includes the following steps:

[0068] Step S1: obtaining an original multi-source lens data set; performing data enhancement processing on the original multi-source lens data set to generate an original enhanced lens data set;

[0069] In an embodiment of the present invention, the original multi-source lens data set is obtained by collecting lens data from different sources (such as optical imaging, physical measurement, environmental monitoring, etc.) through a variety of sensors and data acquisition devices. These data cover multiple characteristics of the lens, such as surface texture, optical performance, refractive index, temperature change, etc. The original data usually has strong diversity and noise, and there are problems such as data scarcity and bias. Therefore, in order to improve the quality and richness of the data, data enhancement processing is required. Data enhancement technology generates enhanced data with similar statistical characteristics to the original data by performing certain transformation operations on the original data, such as rotation, scaling, cropping, color transformation, adding noise, etc. These enhanced data increase the diversity of the data set without changing the essential information of the data, thereby reducing the risk of overfitting in model training. For the lens data set, the specific data enhancement method includes but is not limited to the translation and rotation of the lens surface texture, the brightness adjustment and contrast change of the lens color, the slight distortion or scale transformation of the lens morphology, and even the inversion and mirroring of the lens image. Through these data augmentation techniques, the sample size of the original data set can be effectively expanded, ensuring that each data augmentation strategy can retain the essential characteristics of the lens while increasing the robustness and generalization ability of the model under various conditions.

[0070] Step S2: extracting data features from the original enhanced lens data set to generate original lens feature data; performing multi-dimensional cross-domain feature fusion on the original lens feature data to generate a multi-dimensional cross-domain fusion data set for the lens;

[0071] In the embodiment of the present invention, data feature extraction is to extract information that can effectively represent the important features of the lens from the original enhanced data set of the lens. This process usually relies on a variety of computer vision technologies and feature selection methods. At the data level, the goal of feature extraction is to extract effective features that can describe various physical and optical properties of the lens from the enhanced data, such as texture features, edge features, shape features, color features, and optical performance features. Feature extraction methods can include classic image processing algorithms, such as edge detection (such as Canny edge detection), texture analysis (such as local binary pattern (LBP)), and feature extraction techniques based on deep learning, such as convolutional layer features in convolutional neural networks (CNN). In addition, feature extraction can also use statistical methods, such as principal component analysis (PCA) to reduce data dimensions and extract the most discriminative features. These methods can effectively compress important information related to lens performance from high-dimensional data. Next, the process of multi-dimensional cross-domain feature fusion is carried out, the main purpose of which is to fuse multi-dimensional features from different data sources (such as optical imaging data, temperature data, refractive index data, etc.) to obtain a more comprehensive description of the lens. At the data level, this fusion process usually involves feature selection, weighted averaging, splicing, mapping, and data fusion after mapping. Specifically, by constructing a feature matrix, data from different domains can be spliced ​​horizontally, and different data sources can be given different weights through weighting to further enhance their contribution to model training. At the same time, some algorithms, such as weighted sum, convolutional fusion, and multi-input fusion networks based on neural networks, can be used to achieve effective fusion of cross-domain features.

[0072] Step S3: performing integrated learning based on the lens multi-dimensional cross-domain fusion dataset, and building a model to generate a lens integrated recognition model;

[0073] In the embodiment of the present invention, the core technical means of ensemble learning is to reduce the deviation and variance of the model and improve the overall performance by combining the prediction results of multiple base learners. At the data level, ensemble learning can adopt a variety of methods, such as voting, weighted average, stacking, and random forest. In this step, a common practice is to integrate multiple models trained on different subsets of the data set and combine multiple learning algorithms (such as decision trees, support vector machines, neural networks, etc.) to build base learners. These base learners train the lens data independently, capture different patterns in the data by dividing the data set differently or using different algorithm frameworks, and thus provide diversified prediction results. In the process of model construction, the prediction results of these base learners will be integrated through the ensemble algorithm to obtain a more accurate, robust and generalized lens recognition model. For example, the stacking method can further improve the performance of the model by using a meta-learner to re-learn the output results of the base learner. In the selection of base learners, algorithms with strong generalization ability and adaptability to multi-dimensional data input are usually selected according to the characteristics of the lens data set, such as support vector machine (SVM), random forest (RF), gradient boosting tree (GBDT), etc. These algorithms can make full use of the nonlinear and high-dimensional information in the lens data. Through the ensemble learning method, the advantages of multiple base learners are combined, reducing the overfitting or underfitting problems of a single model and improving the stability and accuracy of the recognition results. In the process of model construction, in order to further improve the performance, a cross-validation strategy is also added to divide the data set for multiple training to ensure the generalization ability of the model on unseen data. At the same time, the robustness and accuracy of the model are further improved by properly standardizing and regularizing the features of the data. Ensemble learning can effectively combine the advantages of different features and handle the complex relationships in multi-dimensional cross-domain fusion data sets, so that the lens integrated recognition model has stronger expressiveness when processing multi-source data, and finally generates a comprehensive and efficient recognition model.

[0074] Step S4: Perform model evaluation on the lens integrated recognition model and generate a report, thereby completing the operation of establishing the lens recognition model.

[0075] In the embodiment of the present invention, the technical means of model evaluation mainly include using a set of standardized evaluation indicators to measure the prediction performance of the model, and these indicators usually include accuracy, recall, F1 value, area under the ROC curve (AUC), mean square error (MSE), etc. At the data level, the evaluation process requires a data set that is not involved in the training, that is, the test set, which is usually separated from the overall data set when the data is divided. The role of the test set is to simulate the model's prediction ability for new data in the actual environment, to avoid the model performing well only on the training data, thereby avoiding overfitting. During the evaluation process, the lens integrated recognition model will first be applied to the test set to obtain the prediction results of the model, and various evaluation indicators will be calculated based on the difference between the actual label and the predicted label to evaluate the classification performance and regression ability of the model. In addition, in actual operation, multiple rounds of cross-validation will be performed to divide and train the data set multiple times to further verify the stability and generalization ability of the model. While evaluating the model, report generation is also an important part of step S4. The technical means of report generation usually include summarizing and visualizing the model evaluation results through automated tools so that users can intuitively understand the performance of the model. These reports usually contain multiple parts, including basic information of the model, details of the training process, statistical data of the evaluation results, and numerical and graphical presentations of various evaluation indicators. Data visualization technology is particularly critical in this process. Common visualization methods include ROC curves, confusion matrices, learning curves, etc. These charts help to show the recognition ability of the model in different categories and its performance under different thresholds. At the same time, the report should also include potential improvement directions for the model, limitations of the current model, and applicability analysis in different practical scenarios. Report generation not only provides quantitative evaluation results, but also provides guidance for further optimization and application of the model. Ultimately, through model evaluation and report generation, the performance of the lens integrated recognition model can be fully understood to ensure that it has sufficient accuracy and robustness to meet the needs of practical applications.

[0076] Preferably, step S1 comprises the following steps:

[0077] Step S11: Acquire original multi-source lens data set;

[0078] Step S12: preprocessing the original multi-source lens data set to generate multi-source preprocessed lens data;

[0079] Step S13: Perform data enhancement processing based on the lens multi-source pre-processed data to generate a lens original enhanced data set.

[0080] In the embodiment of the present invention, obtaining the original multi-source data set of the lens is the starting point of the whole process. Usually, this data set includes data from different sources and different sensors, such as optical performance, physical properties, image data, etc. These data come from different experimental devices, sensors or external environments, so there are certain differences in data type, format and quality. The diversity and heterogeneity of the data set bring challenges to subsequent processing, so the acquired data set needs to have high quality and high consistency to ensure the accuracy and effectiveness of subsequent processing. The original multi-source data set of the lens is preprocessed to generate multi-source preprocessed data of the lens. This process includes multiple technical means, such as data cleaning, missing value filling, outlier detection, denoising and data normalization. The purpose of data cleaning is to delete or correct redundant or erroneous information in the data. Missing value filling can be completed by mean interpolation, nearest neighbor interpolation or other interpolation methods, and outlier detection is usually performed with the help of statistical analysis or model-based anomaly detection algorithm. Data denoising is to remove unnecessary noise through filtering techniques such as Gaussian filtering or median filtering to ensure the quality of the data. In terms of data normalization, standardization (Z-score) or minimum-maximum normalization (Min-Max) methods are often used to ensure that different features are at the same scale to avoid the unbalanced impact of features of different scales on model training. Through this series of operations, the quality of the original data set is improved and prepared for subsequent analysis. Data enhancement is performed on the multi-source preprocessed data of the lens to generate the original enhanced lens data set. The purpose of data enhancement is to improve the generalization ability and robustness of the model and prevent overfitting by increasing the diversity of data. For lens data, data enhancement methods can include geometric transformations (such as rotation, flipping, translation, scaling, etc.), color enhancement (such as changes in brightness, contrast, saturation, etc.), and image noise addition. These operations can generate different variants based on the original data, thereby greatly expanding the scale of the data set without increasing the acquisition cost. In addition, for different types of data, synthetic techniques can also be used, such as SMOTE (Synthetic Minority Oversampling Technology) to balance the class imbalance problem in the data.

[0081] Preferably, step S13 includes the following steps:

[0082] Step S131: classifying the lens multi-source pre-processed data into scarce categories to generate lens multi-source classified data, wherein the lens multi-source classified data includes lens multi-source scarce classified data and lens multi-source sufficient classified data;

[0083] Step S132: performing generative adversarial network processing on the lens multi-source scarce classification data to generate lens adversarial network generated data; performing variational autoencoder transformation on the lens adversarial network generated data to generate lens scarce similarity data; performing synthetic minority class oversampling processing on the lens adversarial network generated data and the lens scarce similarity data to generate lens scarce interpolation sample data; performing synthetic enhancement processing on the lens scarce interpolation sample data to generate lens scarce sample synthetic enhanced data;

[0084] Step S133: performing image flipping processing on the lens multi-source sufficient classification data to generate lens flip sufficient data; performing color transformation on the lens flip sufficient data to generate lens color transformation data;

[0085] Step S134: merging the lens scarce sample synthesis enhancement data and the lens color transformation data to generate a lens original enhancement data set.

[0086] In an embodiment of the present invention, by classifying the lens multi-source pre-processed data into scarce categories, lens multi-source classified data is generated, wherein the data is divided into scarce categories and sufficient categories. Scarce categories generally refer to categories with fewer samples in the original data set, while sufficient categories refer to categories with larger sample sizes. Through this classification, it is possible to effectively distinguish which categories need to be oversampled, thereby balancing the category distribution in the data set and avoiding insufficient learning of the model for minority categories during the training process. The classification process can be automatically classified by using traditional classification algorithms such as decision trees, random forests or support vector machines, or by using deep learning models such as convolutional neural networks (CNNs). Next, the generative adversarial network (GAN) technology is used to process the lens multi-source scarce classified data to generate lens adversarial network generated data. The generative adversarial network consists of a generator and a discriminator. The generator generates new samples by learning the distribution characteristics of scarce categories, while the discriminator judges the generated samples to ensure that the generated data has sufficient authenticity. Next, the generated adversarial data is transformed using a variational autoencoder (VAE) to generate lens scarce similar data. The variational autoencoder generates new data samples by encoding and decoding the input data. These samples are similar to the original data in the feature space and have a certain degree of diversity, which helps improve the model's recognition ability for scarce categories. In order to further enhance the samples of the scarce categories, the synthetic minority oversampling technology (SMOTE) is used to perform sample interpolation to generate lens scarce interpolation sample data. SMOTE expands the number of minority class samples by finding adjacent scarce class samples in the feature space and generating new samples between them. Next, the ADASYN (Adaptive Synthetic Sampling) algorithm is used to further perform synthetic enhancement processing on the scarce interpolation sample data to solve the skew problem of minority class samples and further improve the model's learning ability for scarce categories. The image flipping processing is performed on the lens multi-source sufficient classification data to generate lens flipping sufficient data. In addition, the flipped data is color transformed to generate lens color transformation data. Color transformation is usually achieved by changing the image's attributes such as hue, saturation or brightness, which not only increases the diversity of the data, but also improves the model's generalization ability under different visual conditions. Finally, the lens scarce sample synthetic enhancement data and lens color transformation data are merged to generate the final lens original enhanced dataset.

[0087] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes:

[0088] Step S21: extracting data features from the original enhanced lens data set to generate original lens feature data;

[0089] Step S22: performing t-SNE dimension reduction according to the original feature data of the lens to generate original feature dimension reduction data of the lens, wherein the original feature dimension reduction data of the lens includes lens surface optical performance data and lens temperature-refractive index performance data;

[0090] Step S23: Perform multi-dimensional cross-domain feature fusion based on the original feature dimensionality reduction data of the lens to generate a multi-dimensional cross-domain fusion data set of the lens.

[0091] In an embodiment of the present invention, by mining various types of information in the original enhanced lens data set, feature data that is crucial to subsequent analysis is identified. This process usually relies on traditional feature engineering methods, such as statistical analysis, Fourier transform, edge detection, etc., or uses deep learning technology to automatically extract visual features of image data through convolutional neural networks (CNN). The core goal of feature extraction is to extract the most recognizable features from the original data, so that the subsequent model can perform classification or regression tasks more accurately. t-SNE (t-Distributed Stochastic Neighbor Embedding) dimensionality reduction is performed based on the original feature data of the lens. This is a nonlinear dimensionality reduction technology that is mainly used for visualization of high-dimensional data. t-SNE maintains the local structure between high-dimensional data points so that the reduced-dimensional data can retain the neighboring relationship of similar data points in the original data. Its advantage is that it can effectively reduce noise and redundant information in high-dimensional space, thereby providing a clearer low-dimensional data representation. In this process, the original feature data of the lens will be converted into lens surface optical performance data and lens temperature-refractive index performance data, representing different feature dimensions of the data. The surface optical performance data reflects the reflectivity, refractive index and other optical properties of the lens, while the temperature-refractive index performance data shows the changes in the optical behavior of the lens at different temperatures. These reduced-dimensional data can more clearly express the performance of the lens under different conditions, providing valuable information for subsequent analysis. Multi-dimensional cross-domain feature fusion is performed using the reduced-dimensional data of the original features of the lens. This technical means mainly fuses data from different fields (such as optical performance, thermal performance, etc.) to build a unified feature set containing multi-dimensional information. The key to multi-dimensional cross-domain feature fusion is to integrate feature data from different sources through reasonable fusion strategies (such as weighted averaging, splicing or deep fusion, etc.) to form a more comprehensive and informative feature set. This process not only retains the diversity of the original data, but also enhances the model's perception of the behavior of the lens under different conditions. At the data level, this fusion enables the model to not only identify the basic optical properties of the lens, but also consider the impact of temperature changes on the lens performance, thereby improving the model's prediction accuracy and generalization ability.

[0092] Preferably, step S21 includes the following steps:

[0093] Step S211: performing texture analysis on the original enhanced data set of the lens to generate lens surface texture characteristics; performing shape outlining based on the lens surface texture characteristics to generate lens overall shape data; performing stereo model construction on the lens surface texture characteristics and the lens overall shape data to generate a three-dimensional model of the lens;

[0094] Step S212: performing an accelerated robust feature algorithm analysis on the three-dimensional model of the lens to generate local feature data of the lens; performing a microstructure analysis on the local feature data of the lens to generate microstructure data of the lens;

[0095] Step S213: coupling the original thermal-optical characteristics of the lens microstructure data to generate original characteristic data of the lens.

[0096] In an embodiment of the present invention, the texture analysis technology uses image processing algorithms, such as gray level co-occurrence matrix (GLCM), Gabor filter or convolutional neural network (CNN) to extract the texture characteristics of the lens surface. These methods can identify the tiny details and structures of the lens surface, such as tiny texture changes, gloss, scratches, granularity, etc., by analyzing the texture patterns in the lens surface image. Through in-depth analysis of these texture characteristics, the surface quality of the lens can be quantitatively described. Next, based on the extracted surface texture characteristics, the shape is outlined by a graphic modeling algorithm (such as edge detection, curve fitting or image segmentation) to generate the overall shape data of the lens. This process not only takes into account the surface texture of the lens, but also combines information such as the geometric shape and thickness of the lens to generate a relatively complete three-dimensional structure model of the lens. By combining the surface texture characteristics of the lens with the overall shape data, step S211 finally generates a three-dimensional model of the lens, which becomes the basic data set for subsequent analysis and calculation. Accelerated robust feature algorithm analysis is performed based on the three-dimensional model. This technical approach extracts local features from the three-dimensional data of the lens by using efficient algorithms, such as accelerated principal component analysis (PCA), independent component analysis (ICA) or feature extraction models based on deep learning. Local feature data includes the geometric features, optical properties and relationships with other regions of the lens in the local area, which help reveal subtle differences and potential problems of the lens (for example, local surface defects, small refractive index changes, etc.). On this basis, the local feature data of the lens is subjected to microstructural analysis to generate lens microstructural data. This process uses analysis methods in image processing and materials science, such as microstructural analysis technology (such as X-ray imaging, scanning electron microscope SEM image analysis, etc.) combined with machine learning technology to analyze the material composition, structural density and lattice distribution of the lens at the microscopic level, thereby generating fine-grained lens microstructural data. The microstructural data of the lens is coupled with the temperature-refractive index characteristics. Through physical modeling and data-driven methods (such as finite element analysis, physical parameter regression, etc.), the refractive index change characteristics of the lens at different temperatures can be combined with its microstructural characteristics to generate the original feature data of the lens.

[0097] Preferably, step S23 includes the following steps:

[0098] Step S231: Calculate the mean of the optical performance data on the lens surface to generate the optical mean data on the lens surface; perform spatial mapping on the optical mean data on the lens surface to generate the optical spatial mapping data on the lens surface;

[0099] Step S232: performing time fluctuation analysis on the lens temperature-refractive index performance data, and removing abnormal data to generate lens temperature-refractive index coupling data; performing lens coupling data space mapping on the lens temperature-refractive index coupling data to generate lens photothermal space mapping data;

[0100] Step S233: perform feature matrix splicing on the lens surface optical space mapping data and the lens photothermal space mapping data to generate a multi-dimensional feature matrix of the lens; perform weighted calculation on the multi-dimensional feature matrix of the lens, and simultaneously perform multi-dimensional cross-domain feature fusion of the lens to generate a multi-dimensional cross-domain fusion data set of the lens.

[0101] In the embodiment of the present invention, the optical mean data of the lens surface is obtained by mean calculation. This process uses traditional statistical methods. By averaging the optical performance data of multiple samples, the overall trend of the optical properties of the lens surface is extracted, and the noise and outliers in the data are removed. Subsequently, these mean data are spatially mapped. The spatial mapping technology here generally uses an interpolation algorithm based on multidimensional data (such as Kriging interpolation, radial basis function RBF, etc.) or a dimensionality reduction algorithm (such as t-SNE, PCA, etc.). By mapping the data into a unified space, the distribution relationship of different optical properties in space is clearer, thereby generating optical space mapping data of the lens surface. The temperature-refractive index performance data of the lens is subjected to time fluctuation analysis to identify the time series pattern, periodic fluctuation and trend change in the data. Common time fluctuation analysis methods include Fourier transform, time series analysis, sliding window method, etc., which aim to reveal the temporal behavior of the refractive index of the lens changing with temperature. At the same time, as part of data cleaning, abnormal data removal uses statistical methods such as Z-Score, box plot method, IQR method, etc. to remove those abnormal values ​​that are obviously deviated from the normal range from the data set, so as to ensure the accuracy and reliability of subsequent analysis. After these processes, the lens temperature-refractive index coupling data is obtained and further spatially mapped. In this process, spatial mapping also uses the interpolation algorithm or nonlinear mapping method mentioned above, such as polynomial fitting, graphic nesting, etc., to finally generate lens photothermal spatial mapping data. The lens surface optical spatial mapping data and lens photothermal spatial mapping data are merged by feature matrix splicing technology to form a multi-dimensional feature matrix. Feature matrix splicing generates a richer data set by merging data matrices from different sources. Each dimension in the matrix represents different physical properties and performance of the lens. Next, weighted calculation is performed on this multi-dimensional feature matrix to give it different importance in different feature dimensions. Weighted calculation usually uses certain weight algorithms (such as information gain, mutual information, correlation analysis, etc.) to determine the contribution value of each feature in the final analysis, thereby achieving differentiated processing of the importance of different features. Finally, the availability and information density of the dataset were further improved by performing multi-dimensional cross-domain feature fusion. Cross-domain feature fusion refers to synthesizing features from different data sources (such as optical properties and temperature-refractive index properties) using weighted averages, convolutional layers in convolutional neural networks (CNNs), or other data fusion techniques to form a more comprehensive feature set, providing richer data support for subsequent machine learning or pattern recognition.

[0102] As an example of the present invention, refer to Figure 3 As shown, in this example, step S3 includes:

[0103] Step S31: performing base learning training based on the lens multi-dimensional cross-domain fusion data set to generate lens base learning training data;

[0104] Step S32: performing sampling with replacement on the lens-based learning training data to generate a lens multi-subdataset; performing stacked logistic regression based on the lens multi-subdataset to generate lens multi-based learner analysis data;

[0105] Step S33: Perform hyperparameter tuning on the lens multi-base learner analysis data to generate lens model hyperparameter data; construct a model based on the lens multi-base learner analysis data, and perform Bayesian verification using the lens model hyperparameter data to generate a lens integrated recognition model.

[0106] In the embodiment of the present invention, base learning training is performed based on the multi-dimensional cross-domain fusion data set of the lens, which is a process of constructing a multi-base learner using different feature space information. The purpose of base learning is to perform preliminary training on the multi-dimensional features of the lens through a variety of different learning algorithms, such as decision trees, support vector machines (SVM), random forests, etc., to generate multiple independent learning models, which capture the complex features of the data from different angles, thereby improving the diversity and accuracy of the prediction. The lens base learning training data is sampled with replacement, that is, a multi-sub-dataset of the lens is generated by bootstrap. The bootstrap method is a method of randomly extracting samples from the original data set and resetting them to generate multiple different training subsets. This process can effectively enhance the generalization ability of the model, reduce overfitting, and ensure the robustness of the model. Subsequently, based on these sub-datasets, the stacking logistic regression method is used to generate lens multi-base learner analysis data. Stacking learning is an integrated learning technology that trains multiple different types of base learners and inputs their prediction results as new features into the next layer of learners, usually a meta-learner (such as logistic regression, SVM, etc.). This method can take advantage of the advantages of different base learners and combine their prediction results to improve the accuracy and robustness of the overall model. Perform hyperparameter tuning on the lens multi-base learner analysis data. This process usually automatically finds the optimal hyperparameter combination through methods such as grid search or Bayesian optimization to improve the performance of the model. The goal of hyperparameter tuning is to adjust the key parameters in the base learner (such as learning rate, tree depth, regularization parameter, etc.) so that the model can show the best prediction ability on training data and verification data. Finally, the model is built based on the adjusted lens multi-base learner analysis data, and the model is verified using the Bayesian verification method. Bayesian verification uses Bayesian theory to calculate the posterior distribution of model parameters to evaluate the predictive ability of the model, reduce the risk of overfitting in traditional verification methods, and ensure the generalization performance of the model.

[0107] Preferably, step S32 includes the following steps:

[0108] Step S321: performing sampling with replacement on the lens-based learning training data to generate a lens multi-sub-dataset;

[0109] Step S322: performing vector machine base learner prediction based on the lens multi-sub-datasets to generate lens stacking base learner data; performing meta-learner training on the lens stacking base learner data, and performing regularization processing to generate lens regularized data;

[0110] Step S323: Perform stacked logistic regression on the lens regularization data to generate lens multi-base learner analysis data.

[0111] In an embodiment of the present invention, a lens multi-subdataset is generated by sampling the lens base learning training data with replacement. The process uses the bootstrap method to resample the data, that is, randomly select samples from the original data set, allowing the same data point to be repeatedly selected. This technical means can effectively overcome the noise problem in the data set, increase the robustness of the model to various data patterns, and provide a diverse training data subset for the training of subsequent models, reducing the risk of overfitting. Based on the lens multi-subdataset, a support vector machine (SVM) base learner is used for prediction to generate lens stacking base learner data. Support vector machine is an effective classification and regression analysis tool that performs classification by finding the optimal hyperplane in a high-dimensional feature space, which can handle complex and nonlinear data distributions. Based on the multi-subdataset training, the lens features can be effectively modeled by the support vector machine, providing powerful base learner support for the stacking learning method. Afterwards, the lens stacking base learner data is meta-learned and regularized to generate lens regularized data. Regularization is to control the complexity of the model by introducing additional constraints (such as L1 or L2 regularization), avoid overfitting, and improve the generalization ability of the model. Regularization ensures that the model can remain simple when facing complex data and effectively improves its prediction performance. Stacked logistic regression is performed on the lens regularized data to generate lens multi-base learner analysis data. Stacked logistic regression is a classic ensemble learning method in which the outputs of multiple base learners are passed as input features to the meta-learner (here is logistic regression). The logistic regression model fits the relationship between the output of the base learner and the actual label, and finally generates an integrated prediction result. Through the stacked logistic regression method, the model can integrate the prediction results of multiple base learners, thereby reducing the bias and variance of a single model and improving the overall prediction ability.

[0112] As an example of the present invention, refer to Figure 4 As shown, in this example, step S4 includes:

[0113] Step S41: performing model evaluation on the lens integrated recognition model to generate lens integrated recognition model evaluation data;

[0114] Step S42: generating a report on the lens integrated recognition model evaluation data to generate a lens integrated recognition model report;

[0115] Step S43: Perform data report visualization processing on the lens integrated recognition model report, thereby completing the operation of establishing the lens recognition model.

[0116] In an embodiment of the present invention, the lens integrated recognition model is evaluated to generate lens integrated recognition model evaluation data. This process usually relies on model performance evaluation indicators, such as accuracy, recall rate, F1 value, ROC curve and AUC value. The goal of model evaluation is to objectively measure the performance of the integrated recognition model in practical applications, especially the reliability and accuracy when processing lens data. Through different performance evaluation methods, a systematic statistical analysis of the model output results can reveal the performance of the model under different categories and different environmental conditions, providing a basis for subsequent model optimization. The evaluation process also uses cross-validation technology, that is, by dividing the data set into multiple subsets, training and testing are performed in turn, so as to further reduce the deviation of the evaluation results. By generating a report for the lens integrated recognition model evaluation data, a lens integrated recognition model report is generated. The report generation process usually generates a detailed model evaluation report based on the evaluation results through an automated tool or script. This report not only includes the numerical analysis of each evaluation indicator, but also includes information such as the model training process, parameter tuning, error analysis, and the adaptability of the model to different data sets. The technical means of report generation usually include data processing and summary algorithms, such as data aggregation, statistical inference, and automated template filling, to ensure that the generated report information is comprehensive and easy to understand. The lens recognition model is established by visualizing the data report of the lens integrated recognition model report. The visualization of the data report is to transform complex evaluation data into an intuitive visual display through data visualization techniques, such as charts, curves, and heat maps, so that decision makers can quickly and accurately understand the performance and shortcomings of the model. The technical means of data visualization include using graphical tools (such as Matplotlib, Seaborn, Tableau, etc.) to generate charts, or to achieve interactive data display through a web interface. Through these visualization methods, key information in the report (such as the trend of evaluation indicators, model error distribution, etc.) can be presented to users in a clear and easy-to-understand form, further improving the readability and decision-making value of the report.

[0117] In this specification, a system for establishing a lens recognition model is provided, which is used to execute the above-mentioned method for establishing a lens recognition model. The system for establishing a lens recognition model:

[0118] The data acquisition and enhancement module is used to acquire the original multi-source data set of the lens; perform data enhancement processing on the original multi-source data set of the lens to generate the original enhanced data set of the lens;

[0119] The feature extraction and feature fusion module is used to extract data features from the original enhanced lens data set to generate original lens feature data; perform multi-dimensional cross-domain feature fusion on the original lens feature data to generate a multi-dimensional cross-domain fusion data set for the lens;

[0120] The integrated learning and model building module is used to perform integrated learning and model building based on the multi-dimensional cross-domain fusion data set of lenses to generate an integrated lens recognition model;

[0121] The model evaluation and report generation module is used to evaluate the lens integrated recognition model and generate a report, thereby completing the task of establishing the lens recognition model.

[0122] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0123] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for establishing a lens recognition model, characterized in that: The following steps are involved: Step S1: obtaining an original multi-source lens dataset; performing data enhancement processing on the original multi-source lens dataset and image flipping processing to generate an original enhanced lens dataset; Step S2: extracting data features from the original enhanced lens data set to generate original lens feature data; performing multi-dimensional cross-domain feature fusion on the original lens feature data to generate a multi-dimensional cross-domain fusion lens data set; wherein step S2 includes: Step S21: extracting data features from the original enhanced lens data set to generate original lens feature data; Step S22: performing t-SNE dimension reduction according to the original feature data of the lens to generate original feature dimension reduction data of the lens, wherein the original feature dimension reduction data of the lens includes lens surface optical performance data and lens temperature-refractive index performance data; Step S23: performing multi-dimensional cross-domain feature fusion based on the original feature dimensionality reduction data of the lens to generate a multi-dimensional cross-domain fusion data set of the lens; wherein step S23 includes: step S231: performing mean calculation on the optical performance data of the lens surface to generate optical mean data of the lens surface; performing spatial mapping on the optical mean data of the lens surface to generate optical spatial mapping data of the lens surface; Step S232: performing time fluctuation analysis on the lens temperature-refractive index performance data, and removing abnormal data to generate lens temperature-refractive index coupling data; performing lens coupling data space mapping on the lens temperature-refractive index coupling data to generate lens photothermal space mapping data; Step S233: Perform feature matrix splicing on the lens surface optical space mapping data and the lens photothermal space mapping data to generate a lens multi-dimensional feature matrix; perform weighted calculation on the lens multi-dimensional feature matrix, and simultaneously perform multi-dimensional cross-domain feature fusion on the lens to generate a lens multi-dimensional cross-domain fusion data set; Step S3: performing integrated learning based on the lens multi-dimensional cross-domain fusion dataset, and building a model to generate a lens integrated recognition model; Step S4: Perform model evaluation on the lens integrated recognition model and generate a report, thereby completing the operation of establishing the lens recognition model.

2. The method for establishing a lens recognition model according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: Acquire original multi-source lens data set; Step S12: preprocessing the original multi-source lens data set to generate multi-source preprocessed lens data; Step S13: Perform data enhancement processing based on the lens multi-source pre-processed data to generate a lens original enhanced data set.

3. The method for establishing a lens recognition model according to claim 2, characterized in that: Step S13 includes the following steps: Step S131: classifying the lens multi-source pre-processed data into scarce categories to generate lens multi-source classified data, wherein the lens multi-source classified data includes lens multi-source scarce classified data and lens multi-source sufficient classified data; Step S132: performing generative adversarial network processing on the lens multi-source scarce classification data to generate lens adversarial network generated data; performing variational autoencoder transformation on the lens adversarial network generated data to generate lens scarce similarity data; performing synthetic minority class oversampling processing on the lens adversarial network generated data and the lens scarce similarity data to generate lens scarce interpolation sample data; performing synthetic enhancement processing on the lens scarce interpolation sample data to generate lens scarce sample synthetic enhanced data; Step S133: performing image flipping processing on the lens multi-source sufficient classification data to generate lens flip sufficient data; performing color transformation on the lens flip sufficient data to generate lens color transformation data; Step S134: merging the lens scarce sample synthesis enhancement data and the lens color transformation data to generate a lens original enhancement data set.

4. The method for establishing a lens recognition model according to claim 1, characterized in that: Step S21 includes the following steps: Step S211: performing texture analysis on the original enhanced data set of the lens to generate lens surface texture characteristics; performing shape outlining based on the lens surface texture characteristics to generate lens overall shape data; performing stereo model construction on the lens surface texture characteristics and the lens overall shape data to generate a three-dimensional model of the lens; Step S212: performing an accelerated robust feature algorithm analysis on the three-dimensional model of the lens to generate local feature data of the lens; performing a microstructure analysis on the local feature data of the lens to generate microstructure data of the lens; Step S213: coupling the original thermal-optical characteristics of the lens microstructure data to generate original characteristic data of the lens.

5. The method for establishing a lens recognition model according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: performing base learning training based on the lens multi-dimensional cross-domain fusion data set to generate lens base learning training data; Step S32: performing sampling with replacement on the lens-based learning training data to generate a lens multi-subdataset; performing stacked logistic regression based on the lens multi-subdataset to generate lens multi-based learner analysis data; Step S33: Perform hyperparameter tuning on the lens multi-base learner analysis data to generate lens model hyperparameter data; construct a model based on the lens multi-base learner analysis data, and perform Bayesian verification using the lens model hyperparameter data to generate a lens integrated recognition model.

6. The method for establishing a lens recognition model according to claim 5, characterized in that: Step S32 includes the following steps: Step S321: performing sampling with replacement on the lens-based learning training data to generate a lens multi-sub-dataset; Step S322: performing vector machine base learner prediction based on the lens multi-sub-datasets to generate lens stacking base learner data; performing meta-learner training on the lens stacking base learner data, and performing regularization processing to generate lens regularized data; Step S323: Perform stacked logistic regression on the lens regularization data to generate lens multi-base learner analysis data.

7. The method for establishing a lens recognition model according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: performing model evaluation on the lens integrated recognition model to generate lens integrated recognition model evaluation data; Step S42: generating a report on the lens integrated recognition model evaluation data to generate a lens integrated recognition model report; Step S43: Perform data report visualization processing on the lens integrated recognition model report, thereby completing the operation of establishing the lens recognition model.

8. A system for establishing a lens recognition model, characterized in that: Used to execute the method for establishing a lens recognition model as claimed in claim 1, the system for establishing a lens recognition model comprises: The data acquisition and enhancement module is used to acquire the original multi-source data set of the lens; perform data enhancement processing on the original multi-source data set of the lens to generate the original enhanced data set of the lens; The feature extraction and feature fusion module is used to extract data features from the original enhanced data set of the lens to generate the original feature data of the lens; perform multi-dimensional cross-domain feature fusion on the original feature data of the lens to generate a multi-dimensional cross-domain fusion data set of the lens; The integrated learning and model building module is used to perform integrated learning and model building based on the multi-dimensional cross-domain fusion data set of lenses to generate an integrated lens recognition model; The model evaluation and report generation module is used to evaluate the lens integrated recognition model and generate a report, thereby completing the task of establishing the lens recognition model.

Images