Intelligent identification method and device for shell type based on glass insulator

Abstract

The present invention relates to the field of intelligent identification technology, specifically to a method and device for intelligent identification of shell types based on glass insulators, comprising the following steps: based on the requirements for glass insulator type identification, a differential evolution algorithm is used to perform global search and optimization of parameters, and an optimized parameter configuration is generated by dynamically adjusting the mutation factor and crossover rate, and adjusting the population size and number of iterations according to the characteristics of the identification task. The present invention uses a differential evolution algorithm to optimize the parameters of the intelligent identification system, achieving better adaptability to complex environments and efficient global parameter optimization. This greatly improves the performance and efficiency of the identification system under variable conditions. Secondly, the image noise is processed by a frequency domain filtering method, which effectively improves the image quality under complex backgrounds or low light conditions, thereby improving the recognition accuracy. Combined with information gain evaluation and Bayesian decision methods, the uncertainty and diversity of data can be more accurately evaluated.

Description

Technical Field

[0001] The present invention relates to the field of intelligent identification technology, and in particular to a method and device for intelligently identifying shell types based on glass insulators. Background Art

[0002] The field of intelligent recognition technology involves the use of computer vision, image processing, machine learning and artificial intelligence algorithms to automatically identify and classify objects or patterns. This technology is able to process and analyze batches of image data to identify specific patterns, shapes or features. In the industrial and manufacturing fields, intelligent recognition technology is used to improve production efficiency, quality control, automated inspection, etc. It enables machines to mimic human visual recognition capabilities, thereby improving processing speed and accuracy.

[0003] Among them, glass insulators are a very important component in electronic devices, which are mainly used to provide excellent power supply performance, airtightness and environmental adaptability. The intelligent identification method based on the shell type of glass insulators is a specific application designed to automatically identify and distinguish different types of glass insulator shells. The purpose of this method is to reduce or eliminate the need for manual inspection, thereby improving the efficiency and accuracy of the production and inspection processes. By automatically identifying different types of glass insulators, this method can help ensure that the correct components are used for appropriate applications, reduce errors and improve overall product quality.

[0004] Although the existing technology has demonstrated the ability to process batch image data and recognize specific patterns in the field of intelligent recognition, it still has the problem of insufficient adaptability to complex environments in terms of global parameter optimization. This limitation affects the performance and efficiency of the recognition system under variable conditions. In terms of noise processing and image quality improvement, the existing methods still face challenges in simulating and removing non-periodic noise, especially under complex backgrounds or low light conditions. In addition, the existing technology is also unable to assess data uncertainty and diversity, which is particularly evident in highly complex recognition tasks, affecting the optimization of recognition schemes and the accuracy of decision-making. When processing large amounts of complex data, especially in accurately identifying and excluding abnormal data points, the existing technology has limited effect, which is prone to misidentification and missed judgments. For the processing of constraint satisfaction problems, the existing technology is also insufficient in terms of optimizing the accuracy and efficiency of the recognition process, especially in dealing with specific types of constraints, such as size, shape, color, etc. Finally, when processing highly complex or detailed image data, the existing technology needs to be strengthened in deeply analyzing the local and overall features of the image, which limits the ability to extract richer and more accurate feature information from these data. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings in the prior art and to propose a method and device for intelligently identifying the shell type of a glass insulator.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: a method for intelligently identifying the shell type of a glass insulator, comprising the following steps:

[0007] S1: Based on the requirements for glass insulator type identification, the differential evolution algorithm is used to perform global search and optimization of parameters. The optimized parameter configuration is generated by dynamically adjusting the mutation factor and crossover rate, and adjusting the population size and number of iterations according to the characteristics of the identification task;

[0008] S2: Based on the optimized parameter configuration, a frequency domain filtering method is used to perform noise processing on the image data, including converting the image to the frequency domain, applying a filter to remove the noise frequency component, and restoring the processed image to the time domain, optimizing the effect of noise on recognition accuracy, and generating processed image data;

[0009] S3: Based on the processed image data, a linear programming method is used to define and optimize multiple constraint conditions, including the size, shape and color of the insulator, select a recognition model and match multiple recognition scenarios to generate a structurally optimized recognition model;

[0010] S4: Based on the structurally optimized recognition model, the information gain evaluation and Bayesian decision method are used to analyze the data to be recognized, and the recognition scheme is selected by evaluating the information content of the data features and the probability model of the decision, so as to generate the recognition decision result driven by the data features;

[0011] S5: Based on the recognition decision results driven by the data features, the random forest algorithm and Z-Score outlier detection are used to analyze the data set, and the quality of the data set and the accuracy of recognition are optimized by identifying the importance of data features and abnormal data points, and the recognition decision data after the abnormal points are screened are generated;

[0012] S6: Based on the recognition decision data after the outlier screening, Fourier transform and Riemann geometry analysis are used to perform image analysis operations, and the potential periodicity and structural patterns are revealed by analyzing the frequency components of the image data, and the geometric characteristics and internal structure of the image data are analyzed to generate structural analysis results of the image data;

[0013] S7: Based on the structural analysis results of the recognition decision data and image data after the abnormal point screening, a logistic regression classification method is used to perform data fusion and analysis, and the fused data is classified and processed to generate an insulator shell type recognition result.

[0014] The present invention has the following improvements: the optimized parameter configuration includes a specified mutation rate setting, a crossover operation ratio, and a population size and iteration number adjusted for an insulator identification task; the processed image data includes an image with noise removed in the frequency domain, an image with smoothed edges, and an image with optimized key features; the structurally optimized recognition model includes an insulator size matching, shape recognition accuracy optimization, and an improved model for color recognition; the data feature-driven recognition decision result includes an information gain score for key features, a probability weight for multiple features in a data set, and a probability evaluation of a recognition scheme; the recognition decision data after outlier screening includes an identified key feature set, a list of excluded outlier data points, and a feature strength evaluation of the screened data; the structural analysis results of the image data include a frequency distribution diagram of the image, a recognition result of a periodic pattern, and an overview of the geometric and structural features of the image; and the insulator shell type recognition result includes a type judgment for each insulator sample, a credibility score for the type classification, and a summary result of the comprehensive data analysis.

[0015] The present invention has the following improvements: based on the requirements for glass insulator type identification, a differential evolution algorithm is used to perform global search and optimization of parameters, and the specific steps of generating optimized parameter configurations are as follows by dynamically adjusting the mutation factor and crossover rate, and adjusting the population size and the number of iterations according to the characteristics of the identification task:

[0016] S101: Based on the requirements for glass insulator type identification, the differential evolution algorithm is used to initialize the parameter search process by constructing an initial population, setting the gene length, and selecting a matching encoding method, and the initial parameter range is selected to generate an initial parameter population;

[0017] S102: Based on the initial parameter population, an adaptive strategy is adopted to dynamically adjust the mutation factor and crossover rate in the differential evolution by analyzing the changes of each generation solution and the average solution, so as to optimize the adaptability of the algorithm to the insulator recognition characteristics and generate the adjusted evolutionary parameters;

[0018] S103: Based on the adjusted evolutionary parameters, a population size optimization strategy is adopted to select a population size by comparing the convergence speed and solution quality under multiple population sizes, balance the efficiency and effect of the algorithm, and generate an optimized population size configuration;

[0019] S104: Based on the optimized population size configuration, a convergence detection technology is used to judge the convergence state of parameter optimization by comparing the change range of solutions of consecutive generations, and the iteration is stopped when the termination condition is reached, and a matching parameter configuration of the insulator identification task is selected to generate an optimized parameter configuration.

[0020] The present invention is improved in that, based on the optimized parameter configuration, a frequency domain filtering method is used to perform noise processing on the image data, including converting the image to the frequency domain, applying a filter to remove the noise frequency component, and restoring the processed image to the time domain, optimizing the effect of noise on recognition accuracy, and generating the processed image data in the following specific steps:

[0021] S201: Based on the optimized parameter configuration, using Fourier transform, by calculating the frequency component of each pixel point, converting the image data into a frequency domain representation, and performing edge refinement or feature highlighting to generate first frequency domain image data;

[0022] S202: Based on the first frequency domain image data, using frequency domain filtering technology, by analyzing the frequency spectrum of the image, applying a filter to remove noise components in the image, adjusting the frequency components, improving the designated area of ​​the image, optimizing the interference of image noise on recognition, and generating denoised frequency domain data;

[0023] S203: Based on the denoised frequency domain data, using inverse Fourier transform, by converting the frequency domain signal back to its original time domain representation, restoring the denoised image, and performing color correction and detail optimization to generate restored image data;

[0024] S204: Based on the restored image data, image sharpening and contrast adjustment techniques are used to optimize image quality, and image recognizability and clarity are improved by adjusting image edges and adjusting brightness and contrast, thereby generating processed image data.

[0025] The present invention is improved in that, based on the processed image data, a linear programming method is used to define and optimize multiple constraint conditions, including the size, shape and color of the insulator, and a recognition model is selected while matching multiple recognition scenes to generate a structurally optimized recognition model. The specific steps are as follows:

[0026] S301: Based on the processed image data, a linear programming algorithm is used to solve the constrained optimization problem by applying the simplex method and the interior point method, including constructing a linear relationship model, defining constraint conditions, and iteratively solving, defining and optimizing the insulator characteristic constraint conditions, and generating constraint condition optimization results;

[0027] S302: Based on the optimization result of the constraint conditions, a multi-objective optimization strategy is adopted to balance multiple objective functions by using Pareto optimization and weight method, including converting the objective into a single objective function, and weighting and optimizing it, designing and adjusting the structure of the insulator identification model, and generating a model structure design scheme;

[0028] S303: Based on the model structure design scheme, a genetic algorithm is used to perform population selection, crossover and mutation operations, including creating an initial population, evaluating the fitness of each individual, applying crossover and mutation to generate a new population, and iterating to find the best solution, optimize the model parameters, and generate parameter adjustment results;

[0029] S304: Based on the parameter adjustment results, a cross-validation technique is used to divide the training set and the test set and perform multiple iterations of verification, including dividing the data set into multiple subsets, alternately using some of them as test sets and the rest as training sets, to evaluate the performance of the model on multiple data and generate a recognition model with optimized structure.

[0030] The present invention is improved in that, based on the structurally optimized recognition model, information gain evaluation and Bayesian decision method are used to analyze the data to be recognized, and the recognition scheme is selected by evaluating the information content of the data features and the probability model of the decision, and the specific steps of generating the recognition decision result driven by the data features are as follows:

[0031] S401: Based on the structurally optimized recognition model, an information gain evaluation method is used to calculate the entropy reduction of features to categories, including analyzing the information contribution of each feature to the output category, selecting features of information volume by quantifying information gain, and analyzing the features of the data to be recognized to generate a feature information volume evaluation result;

[0032] S402: Based on the feature information evaluation result, a Bayesian network is used to construct a decision model by using conditional probability and prior knowledge, including constructing a network structure, whose nodes represent variables and edges represent probabilistic relationships between variables, to optimize the decision of the identification scheme and generate a probabilistic model decision result;

[0033] S403: Based on the decision result of the probability model, a simulated annealing algorithm is used to randomly select a solution and adjust the probability of accepting a new solution, including randomly perturbing the initial solution to generate a new solution, and deciding whether to accept the new solution according to the criteria, optimizing and adjusting the decision plan to generate a recognition decision verification result;

[0034] S404: Based on the recognition decision verification result, an integrated learning optimization method is used to optimize the model by using bagging and boosting methods, including combining the predictions of multiple base learners to optimize the decision plan and generate a data feature-driven recognition decision result.

[0035] The present invention has the following improvements: based on the recognition decision results driven by the data features, the random forest algorithm and Z-Score outlier detection are used to analyze the data set, and the quality of the data set and the accuracy of recognition are optimized by identifying the importance of data features and abnormal data points. The specific steps of generating recognition decision data after outlier screening are as follows:

[0036] S501: Based on the recognition decision results driven by the data features, a random forest algorithm is used to establish multiple decision trees and vote on the importance of features, including randomly extracting sub-samples and features from the original data set to construct multiple decision trees, and aggregating the prediction results of the trees to generate feature importance analysis results;

[0037] S502: Based on the feature importance analysis result, a Z-Score outlier detection method is used to identify abnormal data points by calculating the standard scores of the data points, including calculating the degree of deviation of each data point from the average value of the data set, in units of standard deviation, and generating abnormal data point screening results;

[0038] S503: Based on the abnormal data point screening results, a cluster analysis method is used to group the data set through K-means or hierarchical clustering, identify natural patterns and distributions in the data, mine potential patterns and relationships in the data, and generate cluster analysis results;

[0039] S504: Based on the cluster analysis results, data analysis and recognition accuracy evaluation technology is used to quantitatively evaluate the correctness, coverage and balance of model predictions using precision, recall and F1 score indicators, measure the performance of the model in the application, and perform data analysis and recognition accuracy evaluation to generate recognition decision data after outlier screening.

[0040] The present invention is improved in that, based on the recognition decision data after the outlier point screening, Fourier transform and Riemann geometry analysis are used to perform image analysis operations, and the potential periodicity and structural patterns are revealed by analyzing the frequency components of the image data, and the geometric characteristics and intrinsic structure of the image data are analyzed. The specific steps of generating the structural analysis results of the image data are as follows:

[0041] S601: Based on the recognition decision data after the outlier screening, a discrete Fourier transform is used to convert the image signal in the time domain into a complex representation in the frequency domain by calculating the complex amplitude and phase of each data point, and the frequency components of the image data are analyzed to generate second frequency domain image data;

[0042] S602: Based on the second frequency domain image data, using spectrum analysis, by calculating the amplitude spectrum and phase spectrum of the frequency domain data, screening key frequency components, revealing periodic features and potential structural patterns in the image, and generating a periodic feature image;

[0043] S603: Based on the periodic characteristic image, using the Riemannian geometry method, by analyzing the local curvature and shape of the image data, constructing a geometric model of the image data, evaluating the geometric characteristics and intrinsic structure of the image, and generating a geometric characteristic analysis result;

[0044] S604: Based on the periodic characteristic image and the geometric characteristic analysis results, data fusion technology is used to integrate the frequency components and geometric characteristics of the image through statistical methods and machine learning algorithms, and pattern recognition and structural analysis are performed to generate structural analysis results of the image data.

[0045] The present invention is improved in that, based on the structural analysis results of the identification decision data and image data after the abnormal point screening, a logistic regression classification method is used to perform data fusion and analysis, and the fused data is classified and processed to generate the specific steps of the insulator shell type identification result as follows:

[0046] S701: Based on the identification decision data after the outlier screening, a data fusion technology is used to integrate multi-source data and features, including time series data, frequency domain features, and geometric characteristics, to construct a data view of logistic regression classification and generate a fused data set;

[0047] S702: Based on the structural analysis results of the fused data set and the image data, principal component analysis is used to calculate the covariance matrix of the data, extract key components and variables of the data, optimize the data dimension, retain influential features, and generate feature selection results;

[0048] S703: Based on the feature selection result, a logistic regression classification algorithm is used to construct a logistic function model, estimate the influence and probability of multiple features on the classification result, and optimize weight and bias parameters to generate a classification model training result;

[0049] S704: Based on the classification model training results, confusion matrix and ROC curve analysis methods are used to calculate and analyze the true positive examples, false positive examples, true negative examples and false negative examples of the model, and draw the receiver operating characteristic curve to test and verify the model. By evaluating the classification accuracy and error, the reliability and effectiveness of the model are optimized to generate the insulator shell type recognition result.

[0050] The shell type intelligent identification device based on glass insulators includes a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps of the above-mentioned shell type intelligent identification method based on glass insulators when executing the computer program.

[0051] Compared with the prior art, the advantages and positive effects of the present invention are:

[0052] In the present invention, the differential evolution algorithm is used to optimize the parameters of the intelligent recognition system, achieving better adaptability to complex environments and efficient global parameter optimization. This greatly improves the performance and efficiency of the recognition system under variable conditions. Secondly, the image noise is processed by the frequency domain filtering method, which effectively improves the image quality under complex backgrounds or low light conditions, thereby improving the recognition accuracy. In addition, combined with information gain evaluation and Bayesian decision method, the uncertainty and diversity of the data can be more accurately evaluated, and the selection of recognition schemes can be optimized, thereby improving the decision accuracy of the recognition task. The application of random forest algorithm and Z-Score outlier detection further optimizes the quality of the data set and effectively reduces misidentification and missed judgments. The application of linear programming methods and multi-objective optimization strategies improves the accuracy and efficiency of the recognition process when dealing with constraint satisfaction problems. Finally, the application of Fourier transform and Riemann geometry analysis makes the local and overall feature analysis of highly complex or detailed image data more in-depth, allowing more abundant and accurate feature information to be extracted from these data. In general, these innovative solutions greatly improve the overall performance of intelligent recognition of glass insulator types, especially when dealing with complex data, noise interference and highly uncertain recognition tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A flow chart of a method for intelligently identifying the shell type of a glass insulator is proposed for the present invention;

[0054] Figure 2 A schematic diagram of the detailed flow of step S1 in the intelligent identification method of the shell type of glass insulator proposed in the present invention;

[0055] Figure 3 A schematic diagram of a detailed flow chart of step S2 in the intelligent identification method for shell types of glass insulators proposed in the present invention;

[0056] Figure 4 A schematic diagram of a detailed flow chart of step S3 in the intelligent identification method for shell types of glass insulators proposed in the present invention;

[0057] Figure 5 A schematic diagram of a detailed flow chart of step S4 in the intelligent identification method for shell types of glass insulators proposed in the present invention;

[0058] Figure 6 A schematic diagram of a detailed flow chart of step S5 in the intelligent identification method for shell types of glass insulators proposed in the present invention;

[0059] Figure 7 A schematic diagram of a detailed flow chart of step S6 in the intelligent identification method for shell types of glass insulators proposed in the present invention;

[0060] Figure 8 The present invention provides a detailed flow chart of step S7 in the method for intelligently identifying the shell type of glass insulators. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0062] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating positions or positional relationships, are based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0063] Example

[0064] See also Figure 1 The present invention provides a technical solution: a method for intelligently identifying the shell type of a glass insulator, comprising the following steps:

[0065] S1: Based on the requirements for glass insulator type identification, the differential evolution algorithm is used to perform global search and optimization of parameters. The optimized parameter configuration is generated by dynamically adjusting the mutation factor and crossover rate, and adjusting the population size and number of iterations according to the characteristics of the identification task;

[0066] S2: Based on the optimized parameter configuration, a frequency domain filtering method is used to perform noise processing on the image data, including converting the image to the frequency domain, applying a filter to remove the noise frequency component, and restoring the processed image to the time domain, optimizing the effect of noise on recognition accuracy, and generating processed image data;

[0067] S3: Based on the processed image data, a linear programming method is used to define and optimize multiple constraints, including the size, shape and color of the insulator, select the recognition model and match multiple recognition scenarios to generate a structurally optimized recognition model;

[0068] S4: Based on the structural optimization recognition model, the information gain evaluation and Bayesian decision method are used to analyze the data to be recognized. By evaluating the information content of the data features and the probability model of the decision, the recognition scheme is selected to generate the recognition decision results driven by the data features;

[0069] S5: Based on the recognition decision results driven by data features, the random forest algorithm and Z-Score outlier detection are used to analyze the data set. By identifying the importance of data features and abnormal data points, the quality of the data set and the accuracy of recognition are optimized, and recognition decision data after outlier screening is generated;

[0070] S6: Based on the recognition decision data after outlier screening, Fourier transform and Riemann geometry analysis are used to perform image analysis operations, reveal potential periodicity and structural patterns by analyzing the frequency components of the image data, and analyze the geometric characteristics and intrinsic structure of the image data to generate structural analysis results of the image data;

[0071] S7: Based on the structural analysis results of the recognition decision data and image data after outlier screening, a logistic regression classification method is used to perform data fusion and analysis, and the fused data is classified to generate an insulator shell type recognition result.

[0072] The optimized parameter configuration includes the specified mutation rate setting, the crossover operation ratio, and the population size and iteration times adjusted for the insulator recognition task. The processed image data includes the image with noise removed in the frequency domain, the image with smoothed edges, and the image with optimized key features. The structurally optimized recognition model includes the matching of insulator size, the accuracy optimization of shape recognition, and the improved model of color recognition. The data feature-driven recognition decision results include the information gain score of key features, the probability weights of multiple features in the data set, and the probability evaluation of the recognition scheme. The recognition decision data after outlier screening includes the key feature set of recognition, the list of excluded abnormal data points, and the feature strength evaluation of the screened data. The structural analysis results of the image data include the frequency distribution diagram of the image, the recognition results of the periodic pattern, and the overview of the geometric and structural features of the image. The insulator shell type recognition results include the type judgment of each insulator sample, the credibility score of the type classification, and the summary results of the comprehensive data analysis.

[0073] In step S1, the differential evolution algorithm is used to globally optimize the parameters for glass insulator type identification. This involves comprehensive adjustments to key parameters in the identification algorithm, such as thresholds and algorithm-specific variables. Specifically, the algorithm first creates an initial population, where each individual represents a set of parameter configurations. Next, new parameter solutions are generated through mutation and crossover operations, and a selection mechanism is used to retain solutions with better performance. During this process, the mutation factor and crossover rate are dynamically adjusted to ensure that the algorithm can effectively explore the parameter space while avoiding premature convergence. According to the characteristics of the identification task, the population size and number of iterations are adjusted to adapt to different identification scenarios. Finally, this step generates parameter configurations optimized for specific identification tasks, which help improve the accuracy and efficiency of the identification system.

[0074] In step S2, the image data is processed for noise by frequency domain filtering to improve recognition accuracy. First, the image is converted from the time domain to the frequency domain using Fourier transform, so that the image data can be analyzed at the frequency level. Then, the designed filter is applied to remove the noise frequency components, especially the non-periodic noise. In this process, the filter is carefully designed to retain the main features of the image while removing background noise. The processed image is converted back to the time domain through inverse Fourier transform, and the image at this time has reduced noise interference, thus providing clearer and more reliable data for subsequent steps. The generated image data is not only visually clearer, but also more suitable for high-precision recognition at the data level.

[0075] In step S3, linear programming methods are applied to define and optimize multiple constraints, such as the size, shape, and color of the insulator. This involves building a linear relationship model that contains a mathematical description of the insulator characteristics. Then, algorithms such as the simplex method and the interior point method are used to solve the constrained optimization problem to find the best solution that meets all conditions. This approach ensures that the recognition model is not only effective in a single scenario, but can also adapt to a variety of different recognition environments. The optimization results include precise matching of insulator sizes, accuracy of shape recognition, and improvement of color recognition. These optimization results directly affect the performance of the recognition model, making it more accurate and reliable in practical applications.

[0076] In step S4, the data to be identified is analyzed by combining information gain evaluation and Bayesian decision method. First, the contribution of each feature to the output category is analyzed through information gain evaluation, and the amount of information is quantified to determine which features are most important in the identification process. Then, a decision model is constructed using Bayesian network, which helps to select the best identification scheme based on conditional probability and prior knowledge. With this method, the most informative features can be found in complex and uncertain data, and accurate identification decisions can be made based on them. The generated identification decision results not only improve the accuracy of identification, but also ensure the scientificity and rationality of the decision-making process.

[0077] In step S5, the dataset is analyzed by combining the random forest algorithm and Z-Score outlier detection. The random forest algorithm effectively processes and classifies data by building multiple decision trees and voting on the importance of features. In addition, Z-Score outlier detection is used to identify and exclude abnormal data points in the dataset by calculating the degree of deviation of each data point from the mean of the dataset. The quality of the dataset is optimized, the accuracy and reliability of the recognition process are improved, and the data generated in the end is not only optimized in quantity, but also significantly improved in quality.

[0078] In step S6, Fourier transform and Riemann geometry analysis are used for image analysis. The image data is converted to the frequency domain by Fourier transform, and its frequency components are analyzed to reveal the periodicity and structural patterns of the image. Subsequently, the Riemann geometry analysis method is applied to deeply analyze the local curvature and shape of the image data and construct a geometric model of the image data. Not only the structural characteristics of the image are displayed, but also the intrinsic geometric characteristics of the image are revealed. The generated image data structure analysis results provide rich information and help to more accurately identify and classify insulator types.

[0079] In step S7, based on the data from the previous steps, the logistic regression classification method is used for data fusion and analysis. First, by integrating multi-source data and features, a data view of logistic regression classification is constructed. Then, the principal component analysis method is used to extract the key components and variables of the data, optimize the data dimensions, and retain influential features. Then, the logistic regression classification algorithm is used to estimate the impact of multiple features on the classification results and optimize the weight and bias parameters. Finally, the model is tested and verified by the confusion matrix and ROC curve analysis methods to ensure its classification accuracy and effectiveness, thereby generating accurate insulator shell type recognition results.

[0080] See also Figure 2 ,Based on the requirements of glass insulator type identification, the differential evolution algorithm is used to ,globally search and optimize the parameters.,The specific steps of generating the optimized parameter configuration ,are as follows by dynamically adjusting the mutation factor and ,crossover rate, and adjusting the population size and iteration times according to the ,characteristics of the identification task.

[0081] S101: Based on the requirements for glass insulator type identification, the differential evolution algorithm is used to initialize the parameter search process by constructing an initial population, setting the gene length, and selecting a matching encoding method, and the initial parameter range is selected to generate an initial parameter population;

[0082] S102: Based on the initial parameter population, an adaptive strategy is adopted to dynamically adjust the mutation factor and crossover rate in differential evolution by analyzing the changes in each generation of solutions and the average solution, optimize the adaptability of the algorithm to the insulator recognition characteristics, and generate the adjusted evolution parameters;

[0083] S103: Based on the adjusted evolutionary parameters, a population size optimization strategy is adopted to select a population size by comparing the convergence speed and solution quality under multiple population sizes, balancing the efficiency and effect of the algorithm, and generating an optimized population size configuration;

[0084] S104: Based on the optimized population size configuration, the convergence detection technology is used to judge the convergence state of parameter optimization by comparing the change range of solutions of consecutive generations, and the iteration is stopped when the termination condition is reached. The matching parameter configuration of the insulator identification task is selected to generate the optimized parameter configuration.

[0085] In sub-step S101, the process of parameter optimization for the glass insulator type identification task is carried out by the differential evolution algorithm. First, the range and initial conditions of the parameter search are defined, including key parameters in the identification algorithm, such as thresholds and specific algorithm variables. The data format at this stage is mainly the numerical representation of the parameters, such as real numbers or integers. Starting from these initial settings, the differential evolution algorithm constructs an initial population. Each population member represents a set of parameter configurations. The size of the population, the encoding method of the individuals, and the gene length are determined at this stage. The choice of encoding method directly affects the algorithm's search efficiency and the quality of the solution. Then, the algorithm enters an iterative process, continuously optimizing the parameter configuration by performing mutation, crossover, and selection operations on the individuals in the population. In this process, the fitness evaluation of the individuals is based on the performance of their corresponding parameter configurations in specific identification tasks. Finally, a set of preliminarily optimized parameter configurations is generated, laying the foundation for subsequent more detailed optimization.

[0086] In substep S102, the application of the differential evolution algorithm is refined, with special attention paid to the implementation of the adaptive strategy. In this step, the algorithm analyzes the changes in each generation of solutions and the average solution generated by the initial population, and dynamically adjusts the mutation factor and crossover rate accordingly. The mutation factor determines the diversity of the solution, while the crossover rate affects the global search ability of the solution. This enables the algorithm to avoid premature convergence to the local optimal solution while maintaining diversity. The implementation of the adaptive strategy is based on the continuous monitoring and evaluation of the population performance, ensuring that the algorithm remains efficient and accurate throughout the search process. In this way, the algorithm provides a more refined and adaptable parameter configuration for the insulator identification characteristics. Finally, the adjusted evolution parameters are generated, which are more suitable for complex and changing identification tasks.

[0087] In sub-step S103, the algorithm further optimized the population size and number of iterations. By comparing the convergence speed and solution quality under different population sizes, the most suitable population size was found. The choice of population size is related to the search efficiency of the algorithm and the global optimization ability of the solution. Too large a population size leads to a waste of computing resources, while too small a population size causes the algorithm to fall into a local optimum. In addition, the determination of the number of iterations is also based on the same considerations, aiming to balance the search depth and computational cost of the algorithm. Through carefully designed population size and number of iterations, the algorithm can find the optimal parameter configuration within an effective computing time. An optimized population size configuration is generated, which plays a decisive role in the efficiency and effect of the entire parameter optimization process.

[0088] In sub-step S104, a convergence detection technique is introduced to determine whether the parameter optimization process has reached the expected end point. By comparing the change amplitude of consecutive generations of solutions, it is determined whether the algorithm has converged to a stable state. If the change amplitude of the solution is lower than the preset threshold after a certain number of iterations, the algorithm believes that an approximate global optimal solution has been found and stops iterating. This method ensures that the algorithm will not continue invalid iterations after reaching a valid solution, thereby improving the efficiency of the entire optimization process. Through this convergence detection, the algorithm finally generates the parameter configuration that best matches the specific insulator identification task, which can significantly improve the accuracy and efficiency of identification in practical applications.

[0089] See also Figure 3 ,Based on the optimized parameter configuration, the frequency domain filtering method is used to perform noise processing on the image data, including converting the image to the frequency domain, applying filters to remove the noise frequency components, and restoring the processed image to the time domain to optimize the impact of noise on recognition accuracy. The specific steps for generating processed image data are as follows,

[0090] S201: Based on the optimized parameter configuration, Fourier transform is used to convert the image data into a frequency domain representation by calculating the frequency component of each pixel point, and edge refinement or feature highlighting is performed to generate first frequency domain image data;

[0091] S202: Based on the first frequency domain image data, using frequency domain filtering technology, by analyzing the frequency spectrum of the image, applying a filter to remove noise components in the image, adjusting the frequency components, improving the designated area of ​​the image, optimizing the interference of image noise on recognition, and generating denoised frequency domain data;

[0092] S203: Based on the denoised frequency domain data, an inverse Fourier transform is used to convert the frequency domain signal back to its original time domain representation, thereby restoring the denoised image, and performing color correction and detail optimization to generate restored image data;

[0093] S204: Based on the restored image data, image sharpening and contrast adjustment techniques are used to optimize the image quality, and the image recognizability and clarity are improved by adjusting the image edge and adjusting the brightness and contrast, thereby generating processed image data.

[0094] In sub-step S201, the image data is converted from the time domain to the frequency domain by the Fourier transform method, and the preliminary processing of the image noise is performed. First, it involves processing the format of the image data. The image exists in the form of a pixel matrix, and each pixel contains color and brightness information. The Fourier transform calculates the frequency component of each pixel and converts this data into a frequency domain representation. After the conversion, the noise and signal components in the image are separated in frequency, making the subsequent filtering process more effective. In addition, this step also includes refining the edges of the image or highlighting the features, which is achieved by adjusting the amplitude of specific frequency components in the frequency domain. The first frequency domain image data finally generated provides the basis for denoising and image enhancement, which is crucial to improving the accuracy of the recognition algorithm.

[0095] In sub-step S202, the first frequency domain image data is further processed using frequency domain filtering technology. The core is to design and apply filters, the purpose of which is to remove noise components in the image while retaining key image information. By analyzing the spectrum of the image, it is possible to determine which frequency components belong to noise and which are important features of the image. The filter is designed based on this information to only remove frequency components identified as noise, while retaining or enhancing important signal frequency components. This processing not only reduces the noise of the image, but also improves the specified area of ​​the image to a certain extent, such as enhancing the contrast and clarity of the image. The generated denoised frequency domain data provides a cleaner and more accurate basis for image restoration and subsequent processing.

[0096] In sub-step S203, the denoised frequency domain data is converted back to its original time domain representation through an inverse Fourier transform to restore the denoised image. This process is the inverse operation of the Fourier transform, and its purpose is to convert the processed frequency domain signal back into an image format that can be directly recognized and analyzed. During this process, color correction and detail optimization operations are also performed to further improve the image quality. Color correction adjusts the color balance of the image to make it closer to the real scene or more suitable for subsequent recognition tasks. Detail optimization focuses on the high-frequency part of the image and enhances the detail performance of the image. The generated restored image data is visually clearer and technically more suitable for intelligent recognition of insulator types.

[0097] In sub-step S204, image sharpening and contrast adjustment techniques are used to optimize the quality of the restored image data. Image sharpening improves the clarity of the image by enhancing the edge information in the image, which is mainly achieved by enhancing the details of the high-frequency part of the image. Contrast adjustment changes the brightness of the image and enhances the visual differences between different areas in the image, thereby improving the recognizability of the image. These processes not only improve the visual effect of the image, but more importantly, improve the usability of the image in the intelligent recognition system. After these processing steps, the image data finally generated has been significantly improved in clarity and recognition accuracy, laying a solid foundation for the accurate recognition of insulator types.

[0098] See also Figure 4 ,Based on the processed image data, a linear programming method is used to define and optimize multiple constraints, including the size, shape and color of the insulator, and the recognition model is selected while matching multiple recognition scenarios. The specific steps of generating a structurally optimized recognition model are as follows,

[0099] S301: Based on the processed image data, a linear programming algorithm is used to solve the constraint optimization problem by applying the simplex method and the interior point method, including building a linear relationship model, defining constraint conditions, and iteratively solving, defining and optimizing the insulator characteristic constraint conditions, and generating constraint optimization results;

[0100] S302: Based on the optimization results of the constraint conditions, a multi-objective optimization strategy is adopted to balance multiple objective functions by using Pareto optimization and weight method, including converting the objectives into a single objective function, and weighting and optimizing it, designing and adjusting the structure of the insulator identification model, and generating a model structure design scheme;

[0101] S303: Based on the model structure design scheme, a genetic algorithm is used to perform population selection, crossover and mutation operations, including creating an initial population, evaluating the fitness of each individual, applying crossover and mutation to generate a new population, and iterating to find the best solution, optimize the model parameters, and generate parameter adjustment results;

[0102] S304: Based on the parameter adjustment results, a cross-validation technique is used to divide the training set and the test set and perform multiple iterations of verification, including dividing the data set into multiple subsets, alternately using some of them as test sets and the rest as training sets, to evaluate the performance of the model on multiple data and generate a structurally optimized recognition model.

[0103] In sub-step S301, the processed image data is applied to a linear programming algorithm to solve the constraint optimization problem in glass insulator identification. First, it involves building a linear relationship model that defines the key constraints in the insulator identification task, such as size, shape, and color. These constraints are converted into linear equations or inequalities to form a solvable linear programming problem. The data format mainly exists in numerical form, including the size parameters, color information, etc. of the insulator. Next, the algorithm iteratively solves these constraints through two solution strategies, the simplex method and the interior point method, to optimize these constraints to achieve the best recognition performance. In this process, each iteration adjusts the variable values ​​in the model to find the optimal solution that satisfies all constraints. Finally, the constraint optimization results for insulator features are generated, which directly affect the accuracy and adaptability of the recognition model.

[0104] In sub-step S302, based on the optimization results of the constraint conditions, a multi-objective optimization strategy is used to further improve the structure of the recognition model. Multiple objective functions are balanced through Pareto optimization and weighting methods. These objective functions include multiple aspects such as recognition accuracy, processing speed, and resource consumption. The algorithm first converts these multiple objectives into a single objective function, and then assigns appropriate weights to each sub-objective. The data format represents the performance indicators of each objective in numerical form. The weight allocation is based on a deep understanding of the characteristics and requirements of the insulator recognition task, ensuring that the recognition model can achieve the optimal balance in all aspects. The final model structure design scheme not only ensures recognition accuracy, but also takes into account the efficiency and practicality of the algorithm.

[0105] In sub-step S303, the parameters of the recognition model are deeply optimized using a genetic algorithm. The process begins with the creation of an initial population, where each individual represents a set of model parameter configurations. Individuals in the population are selected based on their fitness assessment, which reflects the performance of the parameter configuration in the recognition task. Then, new populations are generated through crossover and mutation operations. These genetic operations promote the diversity of the understanding space and the global search ability of the algorithm. Throughout the process, iterations are repeated to continuously optimize the model parameters. The data format still exists in the form of numerical values ​​in the steps, representing the size and type of various parameters. Finally, optimized model parameters are generated. These parameter configurations are the result of multiple generations of selection and optimization, and are more suitable for complex insulator recognition tasks.

[0106] In sub-step S304, cross-validation technology is used to verify and evaluate the performance of the model. This process involves dividing the entire data set into multiple subsets, and then alternately using a portion of them as a test set and the rest as a training set. This method can effectively evaluate the generalization ability and stability of the model on different data. Each iterative validation will generate data about the performance of the model, such as true positive examples, false positive examples, true negative examples, and false negative examples. These data exist in numerical form and are used to calculate performance indicators such as accuracy, recall rate, and F1 score of the model. Through multiple iterative validations, the algorithm can comprehensively evaluate the performance of the model on multiple data, ensuring that the generated recognition model structure is highly reliable and effective in practical applications.

[0107] See also Figure 5 Based on the structural optimization recognition model, the information gain evaluation and Bayesian decision method are used to analyze the data to be recognized. By evaluating the information content of the data features and the probability model of the decision, the recognition scheme is selected and the specific steps of generating the recognition decision results driven by the data features are as follows.

[0108] S401: Based on the structural optimization recognition model, the information gain evaluation method is used to calculate the entropy reduction of the feature to the category, including analyzing the information contribution of each feature to the output category, selecting the feature of the information amount by quantifying the information gain, and analyzing the features of the data to be recognized to generate the feature information amount evaluation result;

[0109] S402: Based on the feature information evaluation result, a Bayesian network is used to construct a decision model by using conditional probability and prior knowledge, including constructing a network structure, whose nodes represent variables and edges represent the probability relationship between variables, to optimize the decision of the identification scheme and generate a probability model decision result;

[0110] S403: Based on the decision result of the probability model, a simulated annealing algorithm is used to randomly select solutions and adjust the probability of accepting new solutions, including randomly perturbing the initial solution to generate new solutions, and deciding whether to accept the new solution according to the criteria, optimizing and adjusting the decision plan, and generating identification decision verification results;

[0111] S404: Based on the recognition decision verification result, an integrated learning optimization method is used to optimize the model by using the bagging and boosting methods, including combining the predictions of multiple base learners to optimize the decision plan and generate a data feature-driven recognition decision result.

[0112] In sub-step S401, the information gain evaluation method is used to perform feature analysis on the recognition model after structural optimization. First, the algorithm calculates the entropy reduction of each feature to the category, that is, information gain. Information gain is a method to measure the importance of features in the classification process. The data format mainly exists in numerical form, representing the statistical data of different features and categories. By analyzing the information contribution of each feature to the output category, the algorithm can identify the most critical features for the recognition task. This involves a large amount of data calculation and analysis, including probability calculation and entropy calculation. The information gain of each feature is quantified to determine its weight in the recognition model. Finally, the feature information evaluation results are generated, which guide the construction of subsequent decision models, ensuring that the recognition model focuses on the most informative features, thereby improving the accuracy and efficiency of recognition.

[0113] In sub-step S402, a decision model is constructed using a Bayesian network based on the feature information evaluation results. A Bayesian network is a graphical model based on probabilistic relationships, in which nodes represent variables and edges represent probabilistic relationships between variables. In this step, the data format is converted into probability values ​​and conditional probability values ​​to represent the relationship between different features and categories. The process of constructing a Bayesian network involves determining the network structure, setting prior probabilities, and updating these probabilities based on data. Through conditional probabilities and prior knowledge, Bayesian networks can provide effective decision support for uncertain and complex data. The final generated probability model decision result not only reflects the impact of each feature on the recognition decision, but also provides an overall decision optimization direction, making the recognition scheme more scientific and reasonable.

[0114] In sub-step S403, the decision model is optimized using a simulated annealing algorithm. The simulated annealing algorithm is a global optimization algorithm that imitates the annealing process in the physical process to find the global optimal solution. The algorithm starts with an initial solution, generates new solutions through random perturbations, and decides whether to accept the new solutions based on specific criteria. In this process, the probability of accepting new solutions gradually decreases as the iterations proceed, similar to the annealing process of material cooling. The data format exists in this step as the parameters of the decision model and its performance indicators. Through repeated iterations, the simulated annealing algorithm can avoid falling into local optimality and find the optimal decision solution. Finally, the identification decision verification results are generated, which provide effective verification of the decision solution and ensure the global optimality and stability of the final decision solution.

[0115] In sub-step S404, the recognition model is finally optimized by applying an ensemble learning optimization method. Ensemble learning optimizes the decision plan by combining the predictions of multiple base learners. Two ensemble learning methods, Bagging and Boosting, are mainly used here. Bagging constructs multiple base learners by randomly sampling on the original data set, and then merges their prediction results. Boosting trains the base learners sequentially, and each base learner tries to correct the errors of the previous learner. The data format mainly exists in the prediction results of each base learner and their weights. This ensemble learning method can significantly improve the performance of the recognition model, especially in reducing overfitting and improving model robustness. The recognition decision result driven by the data features finally generated integrates the advantages of multiple models and provides more accurate and stable recognition decisions.

[0116] See also Figure 6 ,Based on the recognition decision results driven by data features, the random forest algorithm and Z-Score outlier detection are used to analyze the data set.,By identifying the importance of data features and abnormal data points, the quality of the data set and the accuracy of recognition are optimized.,The specific steps for generating recognition decision data after outlier screening are as follows,

[0117] S501: Based on the recognition decision results driven by data features, the random forest algorithm is used to build multiple decision trees and vote on the importance of features, including randomly extracting sub-samples and features from the original data set to build multiple decision trees, and aggregating the prediction results of the trees to generate feature importance analysis results;

[0118] S502: Based on the feature importance analysis results, the Z-Score outlier detection method is used to identify abnormal data points by calculating the standard scores of the data points, including calculating the degree of deviation of each data point from the average value of the data set, with the standard deviation as the unit, and generating abnormal data point screening results;

[0119] S503: Based on the abnormal data point screening results, a cluster analysis method is used to group the data set through K-means or hierarchical clustering, identify natural patterns and distributions in the data, mine potential patterns and relationships in the data, and generate cluster analysis results;

[0120] S504: Based on the cluster analysis results, data analysis and recognition accuracy evaluation technology are used to quantitatively evaluate the correctness, coverage and balance of model predictions by using precision, recall and F1 score indicators, measure the performance of the model in the application, and conduct data analysis and recognition accuracy evaluation to generate recognition decision data after outlier screening.

[0121] In sub-step S501, the random forest algorithm is used to optimize the quality of the data set and the accuracy of recognition by building multiple decision trees and voting to evaluate the importance of features. First, it involves randomly extracting subsamples and features from the original data set to build multiple decision trees. The core of the random forest algorithm is to enhance the robustness and accuracy of the model by building multiple decision trees. The data and features used by each tree are randomly selected from the original data set. This method aims to reduce the risk of overfitting and improve the generalization ability of the model. The establishment of each tree involves selecting the best split point, which is evaluated by information gain or Gini impurity. After the decision tree is built, the random forest evaluates the importance of the feature through a voting mechanism, that is, by counting the frequency of each feature as a split point in the decision tree to determine its importance. It not only reveals the importance of data features in classification problems, but also increases the interpretability of the model. By evaluating the importance of features in this way, random forests can effectively identify and exclude features that have little impact on the final recognition results, thereby improving the prediction accuracy and efficiency of the model. Finally, the feature importance analysis results generated by this sub-step not only provide an important basis for subsequent outlier detection, but also optimize the performance of the entire recognition process.

[0122] In sub-step S502, the Z-Score outlier detection method is used to identify abnormal data points by calculating the standard scores of data points. This step involves calculating the degree of deviation of each data point from the mean of the data set in units of standard deviation. The Z-Score method is a standardization technique that converts the value of each data point into a score related to the mean and standard deviation of the entire data set. Specifically, the Z-Score of each data point is the result of subtracting the mean of the data set from its value and dividing it by the standard deviation. It reveals the degree of deviation of each data point relative to the entire data set, and a higher or lower Z-Score value indicates that the data point is an outlier. By setting a threshold (for example, the absolute value of the Z-Score is greater than 2 or 3), outliers in the data set can be effectively identified. The key to this step is to accurately calculate the Z-Score of each data point and make reasonable judgments based on these scores. The screened out abnormal data points will be removed from the data set to improve data quality and ensure the accuracy of subsequent analysis and model building. Not only does it improve the reliability of the data, but it also provides a more accurate data basis for accurately identifying and classifying insulators.

[0123] In sub-step S503, based on the results of abnormal data point screening, cluster analysis methods are used to group the data set to identify natural patterns and distributions in the data. In this step, K-means or hierarchical clustering algorithms are used to group data points based on their characteristics to reveal potential patterns and relationships in the data. K-means clustering is performed by specifying the number of clusters (K value) and iteratively assigning data points to the nearest cluster center until the convergence condition is met. Hierarchical clustering, on the other hand, forms a cluster hierarchy by gradually merging or splitting data points. Both methods aim to identify natural groupings in the data, which is crucial for understanding the inherent structure of insulator data. The key to cluster analysis is to select appropriate clustering algorithms and parameters, such as the selection of K value or the determination of distance metric, to ensure the effectiveness and accuracy of data grouping. Through cluster analysis, groups of data points with similar characteristics can be identified to reveal patterns and trends in the data. It not only provides important insights for subsequent data analysis and model building, but also helps to optimize the identification process and improve the accuracy of insulator type identification.

[0124] In sub-step S504, based on the cluster analysis results, data analysis and recognition accuracy evaluation techniques are used to quantitatively evaluate the correctness, coverage and balance of the model predictions. This step involves using indicators such as precision, recall and F1 score to measure the performance of the model in the application. Precision refers to the proportion of positive examples correctly identified by the model to all positive examples, while recall refers to the proportion of positive examples correctly identified by the model to all actual positive examples. The F1 score is the harmonic mean of precision and recall, which is used to evaluate the overall effectiveness of the model. The calculation of these indicators involves comparative analysis of the model prediction results and actual data to quantify the performance of the model. This allows a comprehensive evaluation of the performance of the model in different aspects, including its ability to correctly identify positive examples and its effect in avoiding false identification. The key to this step is to accurately calculate these evaluation indicators and conduct a comprehensive evaluation of the model based on these indicators. Ensure the effectiveness and reliability of the model in practical applications, thereby improving the overall accuracy and efficiency of insulator type identification.

[0125] See also Figure 7 ,Based on the recognition decision data after outlier screening, the Fourier transform and Riemann geometry analysis method are used to perform image analysis operations.,By analyzing the frequency components of the image data, the potential periodicity and structural patterns are revealed, and the geometric characteristics and intrinsic structure of the image data are analyzed.,The specific steps for generating the structural analysis results of the image data are as follows,

[0126] S601: Based on the recognition decision data after outlier screening, a discrete Fourier transform is used to convert the image signal in the time domain into a complex representation in the frequency domain by calculating the complex amplitude and phase of each data point, and the frequency components of the image data are analyzed to generate second frequency domain image data;

[0127] S602: Based on the second frequency domain image data, spectrum analysis is performed to screen key frequency components by calculating the amplitude spectrum and phase spectrum of the frequency domain data, revealing periodic features and potential structural patterns in the image, and generating a periodic feature image;

[0128] S603: Based on the periodic characteristic image, a Riemannian geometry method is used to analyze the local curvature and shape of the image data, construct a geometric model of the image data, evaluate the geometric characteristics and internal structure of the image, and generate a geometric characteristic analysis result;

[0129] S604: Based on the periodic characteristic image and geometric characteristic analysis results, data fusion technology is used to integrate the frequency components and geometric characteristics of the image through statistical methods and machine learning algorithms, and pattern recognition and structural analysis are performed to generate structural analysis results of the image data.

[0130] In sub-step S601, the frequency domain analysis of the image data is performed by discrete Fourier transform, which realizes the conversion of the image signal from the time domain to the frequency domain. It involves calculating the complex amplitude and phase of each point in the image data to obtain its frequency domain representation. The discrete Fourier transform reveals the periodic characteristics in the image by decomposing the image signal into a series of frequency components. In the implementation process, the brightness value of each pixel is regarded as the signal intensity, and the DFT converts these values ​​into a set of frequency components, each with a corresponding amplitude and phase. This operation not only provides a deep understanding of the frequency characteristics of the image data, but also helps to identify repetitive patterns and periodic structures in the image. The generated second frequency domain image data, i.e., the spectrogram, is an intuitive representation of the frequency domain analysis of the original image. It provides a basis for the subsequent spectrum analysis, so that the image data can be analyzed from the frequency domain perspective, thereby deeply understanding the periodicity and structural characteristics of the image content.

[0131] In sub-step S602, the second frequency domain image data is further processed by spectral analysis, with a focus on identifying and screening key frequency components. This process involves calculating the amplitude spectrum and phase spectrum of the frequency domain data, which helps to reveal important periodic features and potential structural patterns in the image. The amplitude spectrum shows the intensity of different frequency components, while the phase spectrum reveals the phase information of these frequency components. By analyzing these spectral lines, it can be determined which frequency components contribute most to the structure and characteristics of the image. For example, high-amplitude frequency components correspond to major edges or patterns in the image. By screening these key frequency components, the most important information in the image can be effectively extracted, thereby generating an image that reflects the periodic characteristics of the image. Not only does it enhance the understanding of the image structure, but it also provides an accurate frequency information basis for subsequent geometric property analysis.

[0132] In sub-step S603, the Riemannian geometry method is used to conduct an in-depth analysis of the periodic characteristic image. This step aims to construct a geometric model of the image data and evaluate the geometric characteristics and intrinsic structure of the image by analyzing the local curvature and shape of the image data. The Riemannian geometry method reveals the intrinsic structure of the image by considering the local geometric properties of the image data, such as curvature and shape changes. For example, by analyzing the curvature changes in different regions of the image, the object boundaries and characteristic shapes in the image can be identified. This process not only provides important geometric information about the image structure, but also helps to better understand the essential characteristics of the image content. The generated geometric characteristic analysis result is a comprehensive description of the image structure, including the shape, size and relationship of the objects in the image. It is crucial for accurately identifying and understanding the image content because it provides the ability to parse the image from a geometric perspective.

[0133] In sub-step S604, the periodic characteristic image and geometric characteristic analysis results are integrated through data fusion technology. It involves the use of statistical methods and machine learning algorithms to fuse the frequency components and geometric characteristics of the image to achieve pattern recognition and structural analysis. The key to data fusion is to comprehensively consider multiple aspects of the image data, including its frequency properties and geometric characteristics, to obtain a more comprehensive and in-depth understanding of the image. By combining these different analysis results, more accurate and comprehensive image structure analysis results can be generated. For example, combining the periodic information obtained from frequency domain analysis and the shape characteristics revealed by geometric analysis can more accurately identify and interpret objects and patterns in the image. It not only improves the accuracy and depth of image analysis, but also provides strong support for the final image recognition and classification. Therefore, the image data can be fully understood from multiple dimensions, providing key support for the intelligent recognition method of shell type based on glass insulators.

[0134] See also Figure 8 ,Based on the structural analysis results of the recognition decision data and image data after the abnormal point screening, the logistic regression classification method is used to perform data fusion and analysis, and the fused data is classified and processed to generate the insulator shell type recognition results. The specific steps are as follows,

[0135] S701: Based on the identification decision data after outlier screening, data fusion technology is used to integrate multi-source data and features, including time series data, frequency domain features, and geometric characteristics, to construct a data view of logistic regression classification and generate a fused data set;

[0136] S702: Based on the structural analysis results of the fused data set and the image data, principal component analysis is used to extract key components and variables of the data by calculating the covariance matrix of the data, optimize the data dimension, retain influential features, and generate feature selection results;

[0137] S703: Based on the feature selection result, a logistic regression classification algorithm is used to construct a logistic function model, estimate the influence and probability of multiple features on the classification result, and optimize the weight and bias parameters to generate a classification model training result;

[0138] S704: Based on the classification model training results, the confusion matrix and ROC curve analysis methods are used to calculate and analyze the true positives, false positives, true negatives and false negatives of the model, and draw the receiver operating characteristic curve to test and verify the model. By evaluating the classification accuracy and error, the reliability and effectiveness of the model are optimized to generate the insulator shell type recognition results.

[0139] In sub-step S701, data fusion technology is used to integrate the identification decision data after outlier screening, as well as multi-source data and features such as time series data, frequency domain features, and geometric properties. The key to this process is to extract valuable information from different data sources and combine them together to form a unified data view for logistic regression classification. For example, time series data provides dynamic information that changes over time, frequency domain features reveal the periodicity and repetitive patterns of data, and geometric properties reflect the spatial structure of data. These different types of data are first converted into a consistent format and then fused through algorithms to ensure that the information of each data source is effectively utilized without conflicting with each other. This comprehensive data view allows the logistic regression model to analyze data from multiple perspectives and improve the accuracy and robustness of classification. The resulting fused dataset is a rich data source with comprehensive features, providing a comprehensive foundation for logistic regression classification.

[0140] In sub-step S702, the fused dataset is further processed by principal component analysis (PCA), aiming to optimize the data dimension and extract key components and variables. PCA identifies the main variables in the data by calculating the covariance matrix of the data, which represent the main information and structure in the data. Specifically, PCA first determines the covariance in the data and then finds the directions with the largest covariance, which are the principal components of the data. In this way, PCA is able to reduce the dimensionality of the data while retaining the most critical features. This is especially important for processing high-dimensional data because it can significantly reduce the amount of computation while avoiding overfitting. Ultimately, the resulting feature selection result is a streamlined and effective dataset that contains only the features that are most influential for classification, providing an ideal input for the application of the logistic regression classification algorithm.

[0141] In sub-step S703, the logistic regression classification algorithm is applied to the feature selection results to build a model and perform classification. Logistic regression is a widely used classification technique that builds a logistic function model by estimating the impact and probability of different features on the classification results. Specifically, the model calculates the contribution of each feature to the classification result and adjusts these contributions by optimizing weights and bias parameters. It involves using statistical methods to estimate the impact of features on the probability of the result and continuously adjusting the model parameters through an iterative algorithm until the best classification effect is achieved. The final classification model training result is a tool that can effectively distinguish different types of insulators with a high degree of accuracy and reliability.

[0142] In substep S704, confusion matrix and ROC curve analysis methods are used to test and validate the classification model. The confusion matrix provides a detailed error analysis showing the performance of the model on different categories, including the number of true positives, false positives, true negatives, and false negatives. The ROC curve (Receiver Operating Characteristic Curve) evaluates the overall performance of the model by depicting the true positive rate and false positive rate at different thresholds. Together, these two methods provide a comprehensive perspective for the performance evaluation of the model. Through these analyses, the classification ability of the model can be accurately judged and areas for improvement can be identified. Ultimately, a comprehensive evaluation of the model performance is generated to ensure the reliability and validity of the insulator shell type recognition results.

[0143] The shell type intelligent identification device based on glass insulators includes a memory and a processor, and is characterized in that a computer program is stored in the memory, and the processor implements the steps of the above-mentioned shell type intelligent identification method based on glass insulators when executing the computer program.

[0144] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. An intelligent identification method for shell type based on glass insulators, characterized in that: The following steps are involved: Based on the requirements for glass insulator type identification, the differential evolution algorithm is used to perform global search and optimization of parameters. The optimized parameter configuration is generated by dynamically adjusting the mutation factor and crossover rate, and adjusting the population size and iteration times according to the characteristics of the identification task. Based on the optimized parameter configuration, a frequency domain filtering method is used to perform noise processing on the image data, including converting the image to the frequency domain, applying a filter to remove noise frequency components, and restoring the processed image to the time domain, optimizing the effect of noise on recognition accuracy, and generating processed image data; Based on the processed image data, a linear programming method is used to define and optimize multiple constraint conditions, including the size, shape and color of the insulator, select a recognition model and match multiple recognition scenarios to generate a structurally optimized recognition model; Based on the structurally optimized recognition model, the information gain evaluation and Bayesian decision method are used to analyze the data to be recognized, and the recognition scheme is selected by evaluating the information content of the data features and the probability model of the decision, so as to generate the recognition decision result driven by the data features; Based on the recognition decision results driven by the data features, the random forest algorithm and Z-Score outlier detection are used to analyze the data set, optimize the quality of the data set and the accuracy of recognition by identifying the importance of data features and abnormal data points, and generate recognition decision data after outlier screening; Based on the recognition decision data after the outlier screening, Fourier transform and Riemann geometry analysis are used to perform image analysis operations, reveal potential periodicity and structural patterns by analyzing the frequency components of the image data, and analyze the geometric characteristics and internal structure of the image data to generate structural analysis results of the image data; Based on the identification decision data after the abnormal point screening and the structural analysis results of the image data, a logistic regression classification method is used to perform data fusion and analysis, and the fused data is classified and processed to generate an insulator shell type identification result.

2. The intelligent identification method of shell type based on glass insulator according to claim 1 is characterized in that: The optimized parameter configuration includes the specified mutation rate setting, the crossover operation ratio, and the population size and iteration number adjusted for the insulator identification task. The processed image data includes the image with noise removed in the frequency domain, the image with smoothed edges, and the image with optimized key features. The structurally optimized recognition model includes the matching of insulator size, the accuracy optimization of shape recognition, and the improved model of color recognition. The data feature-driven recognition decision results include the information gain score of the key features, the probability weights of multiple features in the data set, and the probability evaluation of the recognition scheme. The recognition decision data after outlier screening includes the identified key feature set, the excluded outlier data point list, and the feature strength evaluation of the screened data. The structural analysis results of the image data include the frequency distribution diagram of the image, the recognition results of the periodic pattern, and the overview of the geometric and structural features of the image. The insulator shell type recognition results include the type judgment of each insulator sample, the credibility score of the type classification, and the summary results of the comprehensive data analysis.

3. The intelligent identification method of shell type based on glass insulator according to claim 1 is characterized in that: Based on the requirements for glass insulator type identification, the differential evolution algorithm is used to perform global search and optimization of parameters. By dynamically adjusting the mutation factor and crossover rate, and adjusting the population size and number of iterations according to the characteristics of the identification task, the specific steps for generating the optimized parameter configuration are as follows: Based on the requirements of glass insulator type identification, the differential evolution algorithm is used to initialize the parameter search process by constructing the initial population, setting the gene length and selecting the matching encoding method, and then the initial parameter range is selected to generate the initial parameter population; Based on the initial parameter population, an adaptive strategy is adopted to dynamically adjust the mutation factor and crossover rate in the differential evolution by analyzing the changes of each generation solution and the average solution, optimize the adaptability of the algorithm to the insulator recognition characteristics, and generate the adjusted evolutionary parameters; Based on the adjusted evolutionary parameters, a population size optimization strategy is adopted to select a population size by comparing the convergence speed and solution quality under multiple population sizes, balance the efficiency and effect of the algorithm, and generate an optimized population size configuration; Based on the optimized population size configuration, a convergence detection technology is used to judge the convergence state of parameter optimization by comparing the change range of solutions of consecutive generations, and the iteration is stopped when the termination condition is reached. The matching parameter configuration of the insulator identification task is selected to generate the optimized parameter configuration.

4. The intelligent identification method of shell type based on glass insulator according to claim 1 is characterized in that: Based on the optimized parameter configuration, a frequency domain filtering method is used to perform noise processing on the image data, including converting the image to the frequency domain, applying a filter to remove the noise frequency component, and restoring the processed image to the time domain to optimize the effect of noise on recognition accuracy. The specific steps of generating the processed image data are as follows: Based on the optimized parameter configuration, Fourier transform is used to convert the image data into frequency domain representation by calculating the frequency component of each pixel point, and edge refinement or feature highlighting is performed to generate first frequency domain image data; Based on the first frequency domain image data, a frequency domain filtering technique is used to analyze the frequency spectrum of the image, and a filter is applied to remove noise components in the image, adjust the frequency components, improve the designated area of ​​the image, optimize the interference of image noise on recognition, and generate denoised frequency domain data; Based on the denoised frequency domain data, an inverse Fourier transform is used to restore the denoised image by converting the frequency domain signal back to its original time domain representation, and color correction and detail optimization are performed to generate restored image data; Based on the restored image data, image sharpening and contrast adjustment techniques are used to optimize image quality, and image recognizability and clarity are improved by adjusting image edges and adjusting brightness and contrast, thereby generating processed image data.

5. The intelligent identification method of shell type based on glass insulator according to claim 1 is characterized in that: Based on the processed image data, a linear programming method is used to define and optimize multiple constraints, including the size, shape and color of the insulator, select a recognition model and match multiple recognition scenarios to generate a structurally optimized recognition model. The specific steps are as follows: Based on the processed image data, a linear programming algorithm is used to solve the constrained optimization problem by applying the simplex method and the interior point method, including constructing a linear relationship model, defining constraint conditions, and iteratively solving, defining and optimizing the insulator characteristic constraint conditions, and generating constraint condition optimization results; Based on the optimization results of the constraint conditions, a multi-objective optimization strategy is adopted to balance multiple objective functions by using Pareto optimization and weight method, including converting the objectives into a single objective function and assigning and optimizing the weights thereof, designing and adjusting the structure of the insulator identification model, and generating a model structure design scheme; Based on the model structure design scheme, a genetic algorithm is used to optimize the model parameters by performing population selection, crossover and mutation operations, including creating an initial population, evaluating the fitness of each individual, applying crossover and mutation to generate a new population, and iterating the process to find the best solution, thereby generating parameter adjustment results; Based on the parameter adjustment results, cross-validation technology is used to divide the training set and the test set and perform multiple iterative verifications, including dividing the data set into multiple subsets, alternately using some of them as test sets and the rest as training sets, to evaluate the performance of the model on multiple data and generate a structurally optimized recognition model.

6. The intelligent identification method of shell type based on glass insulator according to claim 1 is characterized in that: Based on the structurally optimized recognition model, the information gain evaluation and Bayesian decision method are used to analyze the data to be recognized. By evaluating the information content of the data features and the probability model of the decision, the recognition scheme is selected and the specific steps of generating the recognition decision results driven by the data features are as follows: Based on the structurally optimized recognition model, an information gain evaluation method is used to calculate the entropy reduction of features to categories, including analyzing the information contribution of each feature to the output category, selecting features of information volume by quantifying information gain, and analyzing the features of the data to be recognized to generate feature information volume evaluation results; Based on the characteristic information evaluation result, a Bayesian network is used to construct a decision model by using conditional probability and prior knowledge, including constructing a network structure, whose nodes represent variables and edges represent probabilistic relationships between variables, to optimize the decision of the identification scheme and generate a probabilistic model decision result; Based on the decision result of the probability model, a simulated annealing algorithm is used to randomly select solutions and adjust the probability of accepting new solutions, including randomly perturbing the initial solution to generate new solutions, and deciding whether to accept the new solution according to the criteria, optimizing and adjusting the decision plan to generate identification decision verification results; Based on the recognition decision verification results, an integrated learning optimization method is used to optimize the model by using Bagging and Boosting methods, including combining the predictions of multiple base learners to optimize the decision plan and generate data feature-driven recognition decision results.

7. The intelligent identification method of shell type based on glass insulator according to claim 1 is characterized in that: Based on the recognition decision results driven by the data features, the random forest algorithm and Z-Score outlier detection are used to analyze the data set. By identifying the importance of data features and abnormal data points, the quality of the data set and the accuracy of recognition are optimized. The specific steps for generating recognition decision data after outlier screening are as follows: Based on the recognition decision results driven by the data features, a random forest algorithm is used to establish multiple decision trees and vote on the importance of features, including randomly extracting sub-samples and features from the original data set to build multiple decision trees, and aggregating the prediction results of the trees to generate feature importance analysis results; Based on the feature importance analysis results, the Z-Score outlier detection method is used to identify abnormal data points by calculating the standard scores of the data points, including calculating the degree of deviation of each data point from the average value of the data set, in units of standard deviation, and generating abnormal data point screening results; Based on the abnormal data point screening results, a cluster analysis method is used to group the data set through K-means or hierarchical clustering, identify natural patterns and distributions in the data, mine potential patterns and relationships in the data, and generate cluster analysis results; Based on the cluster analysis results, data analysis and recognition accuracy evaluation technology are adopted to measure the performance of the model in the application by using precision, recall rate and F1 score indicators, including quantitative evaluation of the correctness of model prediction, coverage and balance, and data analysis and recognition accuracy evaluation to generate recognition decision data after outlier screening.

8. The intelligent identification method of shell type based on glass insulator according to claim 1 is characterized in that: Based on the recognition decision data after the outlier screening, Fourier transform and Riemann geometry analysis are used to perform image analysis operations, and the potential periodicity and structural patterns are revealed by analyzing the frequency components of the image data, and the geometric characteristics and intrinsic structure of the image data are analyzed. The specific steps for generating the structural analysis results of the image data are as follows: Based on the recognition decision data after the outlier screening, a discrete Fourier transform is used to convert the image signal in the time domain into a complex representation in the frequency domain by calculating the complex amplitude and phase of each data point, and the frequency components of the image data are analyzed to generate second frequency domain image data; Based on the second frequency domain image data, spectrum analysis is used to calculate the amplitude spectrum and phase spectrum of the frequency domain data, screen key frequency components, reveal periodic features and potential structural patterns in the image, and generate a periodic feature image; Based on the periodic characteristic image, a Riemannian geometry method is used to analyze the local curvature and shape of the image data, construct a geometric model of the image data, evaluate the geometric characteristics and intrinsic structure of the image, and generate a geometric characteristic analysis result; Based on the periodic characteristic images and geometric property analysis results, data fusion technology is used to integrate the frequency components and geometric properties of the image through statistical methods and machine learning algorithms, and pattern recognition and structural analysis are performed to generate structural analysis results of the image data.

9. The intelligent identification method of shell type based on glass insulator according to claim 1 is characterized in that: Based on the structural analysis results of the identification decision data and image data after the abnormal point screening, the logistic regression classification method is used to perform data fusion and analysis, and the fused data is classified and processed to generate the specific steps of the insulator shell type recognition result as follows: Based on the identification decision data after the outlier screening, data fusion technology is used to integrate multi-source data and features, including time series data, frequency domain features and geometric characteristics, to construct a data view of logistic regression classification and generate a fused data set; Based on the structural analysis results of the fused data set and the image data, principal component analysis is used to extract key components and variables of the data, optimize the data dimension, retain influential features, and generate feature selection results by calculating the covariance matrix of the data; Based on the feature selection results, a logistic regression classification algorithm is used to estimate the influence and probability of multiple features on the classification results by constructing a logistic function model, and weight and bias parameters are optimized to generate a classification model training result; Based on the classification model training results, confusion matrix and ROC curve analysis methods are used to calculate and analyze the true positive examples, false positive examples, true negative examples and false negative examples of the model, and draw the receiver operating characteristic curve to test and verify the model. By evaluating the classification accuracy and error, the reliability and effectiveness of the model are optimized to generate the insulator shell type recognition results.

10. A housing type intelligent identification device based on a glass insulator, comprising a memory and a processor, characterized in that: The memory stores a computer program, and when the processor executes the computer program, the steps of the method for intelligently identifying the shell type of a glass insulator based on any one of claims 1 to 9 are implemented.

Images