Article label identification method based on fuzzy logic control

Abstract

The invention discloses an article label identification method based on fuzzy logic control. The method comprises the following steps: S1, data acquisition; s2, preprocessing the image data and the environment data; s3, performing classification according to an improved clustering algorithm to generate an initial fuzzy rule base; s4, using an improved genetic algorithm to optimize the initial fuzzy rule base, adopting adaptive variation and crossover operation, introducing an elitist retention mechanism, designing a multi-objective optimized fitness function, and generating an optimized fuzzy rule base; s5, constructing a three-layer fuzzy system; s6, fusing three layers of reasoning results through a fuzzy operator to obtain a final label identification result, and if the reasoning confidence is relatively low, performing reasoning again; and S7, storing related data, and if the user corrects the identification result, automatically triggering the improved genetic algorithm to update the fuzzy rule base and the reasoning parameters. Article label identification is optimized through multi-level fuzzy reasoning and the improved genetic algorithm, and the method has the advantages of high adaptability, high precision and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and pattern recognition, and in particular to an article label recognition method based on fuzzy logic control. Background Art

[0002] In the application fields of modern industrial automation, intelligent logistics and product traceability, the recognition of item labels is a key technology. Traditional methods of item label recognition usually rely on the acquisition of high-resolution images and pattern matching algorithms to perform label recognition. However, with the complexity of application scenarios, the difficulty of label recognition also increases. Label images may be affected by lighting conditions, surface materials of objects, color changes and environmental noise, which makes traditional methods inadequate when dealing with blurred, unclear or highly variable labels. Especially when the environment changes significantly, the edges of the labels are blurred or the surface is reflective, traditional rule-based image processing methods are prone to recognition failure or reduced accuracy.

[0003] At present, the commonly used label recognition technologies mainly include feature matching-based methods, deep learning-based methods, and methods combining multimodal information. The feature matching-based method extracts and recognizes the features of the label by means of edge detection, shape matching, template matching, etc. However, this method often does not work well when dealing with blurred labels or blurred edges, is easily affected by noise and illumination changes, and the recognition result depends on the preset template matching accuracy. In addition, when the color of the label is close to the background color, this method is also difficult to accurately separate the label from the background. Although the deep learning-based method can improve the recognition accuracy to a certain extent, it often requires a large amount of label sample data for training, and the training and reasoning process of the model requires high computing resources, which is not suitable for resource-constrained embedded devices. In addition, when facing new environments or unseen label types, the generalization ability of the deep learning model may be insufficient, resulting in a decrease in recognition accuracy.

[0004] Some existing improved methods attempt to introduce multimodal information fusion technology, such as combining image data and environmental data (light, temperature, humidity, etc.) to improve the robustness of recognition. However, these methods usually process each modal feature independently and then simply fuse them in the decision stage, failing to fully utilize the correlation information between different modal features. This feature processing method limits the adaptive ability of the system, making it unable to dynamically adjust the feature weights and processing strategies when facing complex environmental changes, thereby affecting the recognition effect. In addition, the design of traditional multimodal fusion methods mostly lacks a flexible adjustment mechanism and cannot optimize the feature fusion strategy according to environmental changes in real time, resulting in poor recognition performance in scenarios with drastic environmental changes.

[0005] The main drawback of the prior art is that the rule settings of its label recognition process are relatively fixed, and it is difficult to adjust the rule parameters in real time according to environmental changes, resulting in a decrease in recognition accuracy in complex or changeable scenarios. The present invention generates an initial fuzzy rule base through an improved clustering algorithm, and combines the adaptive evolution strategy of the genetic algorithm to perform multi-level optimization of the rule base, thereby enhancing the system's adaptability to complex environments. In addition, the existing methods are usually unable to effectively handle the complex interactive relationships between multiple features, while the present invention adopts a multi-level fuzzy reasoning system to hierarchically process the shape, color and material features and then perform fusion reasoning, so that the correlation information between different features can be fully utilized. By adaptively adjusting the weights and reasoning rules of the fuzzy operators, the present invention can still maintain a high recognition accuracy under conditions such as fuzzy label features, poor lighting conditions and complex materials.

[0006] In addition, the recognition effect of traditional methods often decreases significantly when dealing with problems such as blurred labels, blurred edges or surface reflections. The present invention further optimizes the fuzzy rule base and reasoning parameters by dynamically adjusting the shape of the membership function during the reasoning process and re-executing the evolution process of the genetic algorithm when the reasoning confidence is low, thereby achieving real-time adaptive adjustment to different features and environmental changes. Compared with the prior art, the present invention provides a more robust and flexible label recognition solution that can significantly improve the accuracy and stability of label recognition under changing environmental conditions.

[0007] Therefore, how to provide an object label recognition method based on fuzzy logic control is a problem that those skilled in the art need to solve urgently. Summary of the invention

[0008] One purpose of the present invention is to propose an article label recognition method based on fuzzy logic control. The present invention adopts multi-level fuzzy reasoning and improved genetic algorithm to optimize the label recognition method, and realizes the comprehensive processing of shape, color and material characteristics by dynamically adjusting the fuzzy rule base and feature weights. Combined with multimodal data fusion strategy and adaptive evolution mechanism, the adaptability of the system to complex environments is enhanced, and high-precision recognition effects can still be maintained in situations such as significant changes in lighting, blurred labels or complex backgrounds. Compared with the prior art, the present invention significantly improves the robustness, accuracy and reliability of label recognition, and is suitable for changing application scenarios and complex practical needs.

[0009] The method for identifying an item number based on fuzzy logic control according to an embodiment of the present invention comprises the following steps:

[0010] S1, collect image data with item labels and record environmental information to form a multi-dimensional data set;

[0011] S2, grayscale, denoise, edge detect and image enhance the image data of the multi-dimensional data set, and standardize the environmental data;

[0012] S3, classifying the labeled sample data according to the improved clustering algorithm to generate an initial fuzzy rule base, wherein the initial fuzzy rule base includes fuzzy sets and membership functions corresponding to the input features;

[0013] S4. Use the improved genetic algorithm to optimize the initial fuzzy rule base, adopt adaptive mutation and crossover operations, dynamically adjust the mutation rate and crossover rate according to the evolution stage, introduce the elite retention mechanism, combine the label recognition accuracy and rule simplicity to design the multi-objective optimization fitness function, and generate the optimized fuzzy rule base;

[0014] S5. Construct a three-layer fuzzy system, wherein the first layer processes shape feature reasoning, and infers the shape features of the label through the optimized fuzzy rule base; the second layer processes color feature reasoning, and performs fuzzy reasoning on the color features in combination with the environmental data; the third layer processes object material reasoning, and performs fuzzy reasoning on the object material in combination with the environmental data;

[0015] S6. The three-layer reasoning results are integrated through the fuzzy operator to obtain the final label recognition result. If the reasoning confidence is low, the genetic algorithm evolution process is re-executed and reasoning is performed again;

[0016] S7. The label content, reasoning confidence and label recognition result are stored in the database. If the user corrects the recognition result, the improved genetic algorithm is automatically triggered to update the fuzzy rule base and reasoning parameters.

[0017] Optionally, the S3 specifically includes:

[0018] S31, performing feature extraction on the labeled sample data of the preprocessed multidimensional data set, extracting shape features, color features and texture features of the image data, and normalizing the feature data;

[0019] S32. Use the improved K-Means clustering algorithm based on Gaussian process regression kernel function to cluster and classify the sample data after feature extraction, and use the Gaussian process kernel function to calculate the similarity between data points:

[0020] ;

[0021] in, Representation sample To cluster center The distance represents the total number of features, Indicates The weight of the feature, Representation sample No. eigenvalues, Represents the cluster center No. eigenvalues, Represents the Gaussian process kernel function:

[0022] ;

[0023] in, represents the exponential function, represents the kernel function width, and represents the adjustment parameter;

[0024] S33. In the clustering process, an optimization strategy based on GPR kernel entropy is used to adaptively update feature weights. ;

[0025] S34. Construct an initial fuzzy rule base based on the clustering results, set the fuzzy set and membership function, and the membership function adopts the dual Gaussian-Laplace function fused with the kernel function:

[0026] ;

[0027] in, Indicates the input value Compared with fuzzy sets The degree of affiliation, Indicates The center value of the peak, Indicates The width of the peak, represents the number of local peaks in the membership function, represents the parameter that controls the shape of the membership function, represents the weight of the Laplace term that controls the membership function;

[0028] S35. Redundancy detection and optimization are performed on the rule entries in the fuzzy rule base, and the correlation between the rules is calculated using the kernel entropy interaction formula:

[0029] ;

[0030] in, Representation Rules and rules The amount of fuzzy mutual information between Representation Rules In the The membership function value on the features is Representation Rules In the The membership function value on the feature.

[0031] Optionally, the S4 specifically includes:

[0032] S41. An improved genetic algorithm is used to optimize the initial fuzzy rule base, and the population is divided into a global exploration layer and a local fine optimization layer. The global exploration layer focuses on a wide search of the solution space, and the local fine optimization layer focuses on fine-tuning and optimizing high-fitness individuals, focusing on the refinement of the rule structure and local optimization operations.

[0033] S42: In the adaptive mutation operation, a multi-level dynamic mutation strategy based on the rule association matrix is ​​adopted to adaptively adjust the mutation rate according to the correlation between rules and the interactivity between features:

[0034] ;

[0035] in, represents the current mutation rate, represents the number of rule sets, represents the total number of features, Representation Rules and rules The similarity matrix elements between Representation Rules and rules The weight adjustment factor between Represents an individual The fitness value of Represents the maximum fitness in the population;

[0036] S43. In the multi-level crossover operation, the overall structure of the fuzzy rules and the feature combination within the rules are cross-reorganized, and the crossover probability is adjusted in combination with the multi-objective optimization strategy. The crossover operation combines the traditional global crossover and the high-order asymmetry and correlation between the features for dynamic reorganization:

[0037] ;

[0038] in, represents the current crossover rate, represents the number of high-order feature combinations, Indicates The weight of the high-order feature combination, represents the exponential function, and represents the adjustment parameter, Indicates The distance metric for feature combinations is represents an asymmetric correlation measure for high-order feature combinations, represents the number of structural complexity indicators, Indicates The weight of the complexity index, represents the inverse tangent function, Indicates The value of a complexity indicator;

[0039] S44, introduce elite retention and structured perturbation mechanisms, give priority to retaining individuals with higher fitness in each generation, and perform structured perturbations on elite individuals based on rule complexity and population diversity. The perturbation strategy adopts a dual perturbation mechanism;

[0040] S45, using multi-objective fitness function for comprehensive evaluation, combined with the label recognition accuracy, simplicity of fuzzy rules and complex interactions between rules for optimization:

[0041] ;

[0042] in, represents the fitness value of an individual, represents the number of error indicators considered during the optimization process, Indicates error terms, represents the total number of features, Indicates The weight of the feature interaction, Indicates The interaction measure between features is represents the adjustment parameter, represents the number of overlap indicators considered during the optimization process, Indicates The degree of overlap of rules;

[0043] S46. After multi-level evolution, adaptive operation and multi-objective optimization strategy, an optimized fuzzy rule base is finally generated.

[0044] Optionally, the S44 specifically includes:

[0045] S441. For the disturbance amplitude of elite individuals, a dual disturbance strategy is used for adaptive adjustment, combining the fitness differences between individuals, the nonlinear correlation between characteristics and the diversity of the population:

[0046] ;

[0047] in, Represents an individual The disturbance amplitude, represents the basic disturbance intensity, Represents an individual The difference between the fitness and the maximum fitness, represents the maximum fitness in the population, and represents the adjustment parameter that controls the disturbance intensity, represents the number of individuals in the population, Represents an individual and individuals The characteristic differences between represents the nonlinear adjustment parameter, Represents an individual Diversity measures;

[0048] S442. After the disturbance operation, the elite individuals are adaptively repaired, and the characteristic values ​​are adjusted by combining the fitness gradient, the nonlinear interaction between the characteristics and the multi-objective optimization strategy:

[0049] ;

[0050] in, represents the repaired eigenvalue, represents the eigenvalue before perturbation, represents the learning rate, represents the fitness gradient, represents the weight adjustment coefficient, Indicates The interaction measure between features is represents the adjustment parameter, Indicates The regular overlap of features, Indicates A measure of change in a feature;

[0051] S443. Introduce a multi-scale perturbation control strategy for elite individuals. In the early stage of evolution, increase the perturbation intensity to expand the search range; in the middle stage of evolution, gradually reduce the perturbation intensity, increase the perturbation frequency, and use more frequent small-amplitude perturbations to optimize the performance of elite individuals; in the late stage of evolution, further reduce the perturbation intensity and gradually make fine adjustments to the elite individuals;

[0052] S444, elite selection is based on nonlinear correlation and feature coverage, and uses a nonlinear correlation measurement formula to calculate the distribution of individuals in the feature space.

[0053] Optionally, the S5 specifically includes:

[0054] S51, construct a three-layer fuzzy system, the first layer is used to process shape feature reasoning, the shape features of the labels are input into the optimized fuzzy rule base, the shape features are mapped to language variables in the fuzzy set through the membership function, and the fuzzy reasoning is performed in combination with the geometric features and boundary information of the shape;

[0055] S52, in the first-level reasoning process, the shape features are fuzzified by using a multi-layer membership function to decompose the complex geometric features into basic shape features. After the basic shape features are processed by different membership functions, the comprehensive membership values ​​used for reasoning are output;

[0056] S53, the second layer processes color feature reasoning, inputs the color features of the labels into the fuzzy rule base, combines the environmental data, performs fuzzy reasoning on the color features, maps the color features into language variables, and compensates and corrects the color changes under different lighting conditions;

[0057] S54. In the second-level reasoning, a multimodal data fusion strategy is applied to jointly process color features and environmental data, and the influence of illumination and color temperature is compensated by dynamically adjusting the shape of the membership function. The fusion weights of color and environmental features are adjusted using an adaptive fuzzy operator to generate an adaptive reasoning result.

[0058] S55, the third layer processes the reasoning of material characteristics of items, inputs the material characteristics of the labels into the fuzzy rule library, and performs fuzzy reasoning on the material characteristics in combination with environmental data; when reasoning about materials, a multidimensional membership function is used to hierarchically process different material characteristics, mapping complex material characteristics into fuzzy language variables.

[0059] Optionally, the S6 specifically includes:

[0060] S61. The three-layer reasoning results are integrated through fuzzy operators to perform comprehensive reasoning on shape features, color features, and material features:

[0061] ;

[0062] in, represents the membership value of the comprehensive reasoning result, represents the membership value of the shape feature, Represents the membership value of the color feature, Indicates the membership value of the material feature, represents the uncertainty or error measure of the shape feature, represents the uncertainty or error measure of the color feature, represents the uncertainty or error measure of the material characteristics, , and represents the weight coefficient of the feature, , , and represents the control parameter, , , , , , , and represents the adjustment parameter;

[0063] S62: Calculate the confidence of the final inference result. If the confidence is lower than a preset threshold, the inference result is considered unreliable.

[0064] ;

[0065] in, represents the confidence of inference, represents the final inference result, represents the exponential function, and represents the adjustment parameter, Indicates a preset reference value or target value. represents the adjustment coefficient, Indicates the threshold value;

[0066] S63, adjusting the initial population of the genetic algorithm according to the error of the current reasoning result and the system feedback information, and generating population individuals containing a diverse feature combination;

[0067] S64. Use an improved genetic algorithm to perform multiple iterations of population optimization, combine adaptive mutation and crossover operations to enhance the evolution effect, and use adaptive mutation rate and crossover rate to dynamically adjust the evolution process;

[0068] S65, after multiple iterations of optimization by the genetic algorithm, an optimized fuzzy rule base is generated, and the inference rules and membership functions are reconfigured and adjusted;

[0069] S66. Use the updated fuzzy rule base to re-reason and recalculate the final reasoning result and confidence. If the new reasoning confidence is still lower than the threshold, increase the number of iterations of the genetic algorithm and further optimize the population feature combination and the rule base until the reasoning confidence is higher than the threshold.

[0070] The beneficial effects of the present invention are:

[0071] First, the present invention realizes efficient recognition of item labels in complex environments by introducing a multi-level reasoning system based on fuzzy logic control, significantly improving the accuracy and robustness of recognition. Compared with existing methods based on feature matching and deep learning, the present invention performs better in processing unclear, blurred edge or labels affected by lighting and materials. This is mainly due to the design of the multi-level fuzzy system, which reasons the shape, color and material features of the label in layers, processes them separately and then fuses the reasoning, so that the complex correlation information between different features is fully utilized, thereby enhancing the recognition's ability to adapt to environmental changes. At the same time, through fuzzy operators and adaptive reasoning mechanisms, real-time adjustment of multiple feature weights is achieved, so that a high reasoning accuracy can still be maintained when the lighting changes significantly or the material characteristics are complex.

[0072] Secondly, the improved clustering algorithm and genetic algorithm proposed in the present invention further optimize the construction and update process of the fuzzy rule base. The improved clustering algorithm adopts the Gaussian process regression kernel function to dynamically update the feature weights during the labeled sample clustering process, making the initial fuzzy rule base more representative. The improved genetic algorithm adopts a multi-level evolutionary strategy, combined with adaptive mutation, crossover operation and elite retention mechanism, to perform multi-objective optimization on the fuzzy rule base, so that the rule base achieves a better balance between label recognition accuracy and rule simplicity. In the case of low reasoning confidence, the present invention can optimize the rule base again by re-executing the evolution process of the genetic algorithm, and perform iterative reasoning, so that the system can self-adjust and adapt to environmental changes, thereby improving the reliability of label recognition.

[0073] In addition, the present invention introduces a multimodal data fusion strategy in the reasoning process, jointly processes image features and environmental data, and compensates for different environmental influences by dynamically adjusting the shape of the membership function. This fusion reasoning strategy significantly enhances the robustness of the system in complex scenes, especially when the label features are fuzzy, the lighting is uneven, or the background is complex, it can still maintain a high recognition success rate. At the same time, by introducing multi-scale perturbations and adaptive repair mechanisms, the evolution process of the genetic algorithm is made more efficient and flexible, and the local optimal trap is effectively avoided in the process of optimizing the fuzzy rule base, thereby improving the diversity and adaptability of the rule base. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0075] Figure 1 This is a flow chart of the article label recognition method based on fuzzy logic control proposed by the present invention;

[0076] Figure 2 Schematic diagram of the three-layer fuzzy system architecture of the object label recognition method based on fuzzy logic control proposed in the present invention. DETAILED DESCRIPTION

[0077] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0078] refer to Figure 1 and Figure 2 , an object label recognition method based on fuzzy logic control, comprising the following steps:

[0079] S1, collect image data with item labels and record environmental information to form a multi-dimensional data set;

[0080] S2, grayscale, denoise, edge detect and image enhance the image data of the multi-dimensional data set, and standardize the environmental data;

[0081] S3, classifying the labeled sample data according to the improved clustering algorithm to generate an initial fuzzy rule base, wherein the initial fuzzy rule base includes fuzzy sets and membership functions corresponding to the input features;

[0082] S4. Use the improved genetic algorithm to optimize the initial fuzzy rule base, adopt adaptive mutation and crossover operations, dynamically adjust the mutation rate and crossover rate according to the evolution stage, introduce the elite retention mechanism, combine the label recognition accuracy and rule simplicity to design the multi-objective optimization fitness function, and generate the optimized fuzzy rule base;

[0083] S5. Construct a three-layer fuzzy system, wherein the first layer processes shape feature reasoning, and infers the shape features of the label through the optimized fuzzy rule base; the second layer processes color feature reasoning, and performs fuzzy reasoning on the color features in combination with the environmental data; the third layer processes object material reasoning, and performs fuzzy reasoning on the object material in combination with the environmental data;

[0084] S6. The three-layer reasoning results are integrated through the fuzzy operator to obtain the final label recognition result. If the reasoning confidence is low, the genetic algorithm evolution process is re-executed and reasoning is performed again;

[0085] S7. The label content, reasoning confidence and label recognition result are stored in the database. If the user corrects the recognition result, the improved genetic algorithm is automatically triggered to update the fuzzy rule base and reasoning parameters.

[0086] In this implementation, S3 specifically includes:

[0087] S31, performing feature extraction on the labeled sample data of the preprocessed multidimensional data set, extracting shape features, color features and texture features of the image data, and normalizing the feature data;

[0088] S32. Use the improved K-Means clustering algorithm based on Gaussian process regression kernel function to cluster and classify the sample data after feature extraction, and use the Gaussian process kernel function to calculate the similarity between data points:

[0089] ;

[0090] in, Representation sample To cluster center The distance represents the total number of features, Indicates The weight of the feature, Representation sample No. eigenvalues, Represents the cluster center No. eigenvalues, Represents the Gaussian process kernel function:

[0091] ;

[0092] in, represents the exponential function, represents the kernel function width, and represents the adjustment parameter;

[0093] S33. In the clustering process, an optimization strategy based on GPR kernel entropy is used to adaptively update feature weights. ;

[0094] S34. Construct an initial fuzzy rule base based on the clustering results, set the fuzzy set and membership function, and the membership function adopts the dual Gaussian-Laplace function fused with the kernel function:

[0095] ;

[0096] in, Indicates the input value Compared with fuzzy sets The degree of affiliation, Indicates The center value of the peak, Indicates The width of the peak, represents the number of local peaks in the membership function, represents the parameter that controls the shape of the membership function, represents the weight of the Laplace term that controls the membership function;

[0097] S35. Redundancy detection and optimization are performed on the rule entries in the fuzzy rule base, and the correlation between the rules is calculated using the kernel entropy interaction formula:

[0098] ;

[0099] in, Representation Rules and rules The amount of fuzzy mutual information between Representation Rules In the The membership function value on the features is Representation Rules In the The membership function value on the feature.

[0100] In this implementation, S4 specifically includes:

[0101] S41. An improved genetic algorithm is used to optimize the initial fuzzy rule base, and the population is divided into a global exploration layer and a local fine optimization layer. The global exploration layer focuses on a wide search of the solution space, and the local fine optimization layer focuses on fine-tuning and optimizing high-fitness individuals, focusing on the refinement of the rule structure and local optimization operations.

[0102] S42: In the adaptive mutation operation, a multi-level dynamic mutation strategy based on the rule association matrix is ​​adopted to adaptively adjust the mutation rate according to the correlation between rules and the interactivity between features:

[0103] ;

[0104] in, represents the current mutation rate, represents the number of rule sets, represents the total number of features, Representation Rules and rules The similarity matrix elements between Representation Rules and rules The weight adjustment factor between Represents an individual The fitness value of Represents the maximum fitness in the population;

[0105] S43. In the multi-level crossover operation, the overall structure of the fuzzy rules and the feature combination within the rules are cross-reorganized, and the crossover probability is adjusted in combination with the multi-objective optimization strategy. The crossover operation combines the traditional global crossover and the high-order asymmetry and correlation between the features for dynamic reorganization:

[0106] ;

[0107] in, represents the current crossover rate, represents the number of high-order feature combinations, Indicates The weight of the high-order feature combination, represents the exponential function, and represents the adjustment parameter, Indicates The distance metric for feature combinations is represents an asymmetric correlation measure for high-order feature combinations, represents the number of structural complexity indicators, Indicates The weight of the complexity index, represents the inverse tangent function, Indicates The value of a complexity indicator;

[0108] S44, introduce elite retention and structured perturbation mechanisms, give priority to retaining individuals with higher fitness in each generation, and perform structured perturbations on elite individuals based on rule complexity and population diversity. The perturbation strategy adopts a dual perturbation mechanism;

[0109] S45, using multi-objective fitness function for comprehensive evaluation, combined with the label recognition accuracy, simplicity of fuzzy rules and complex interactions between rules for optimization:

[0110] ;

[0111] in, represents the fitness value of an individual, represents the number of error indicators considered during the optimization process, Indicates error terms, represents the total number of features, Indicates The weight of the feature interaction, Indicates The interaction measure between features is represents the adjustment parameter, represents the number of overlap indicators considered during the optimization process, Indicates The degree of overlap of rules;

[0112] S46. After multi-level evolution, adaptive operation and multi-objective optimization strategy, an optimized fuzzy rule base is finally generated.

[0113] In this implementation manner, the S44 specifically includes:

[0114] S441. For the disturbance amplitude of elite individuals, a dual disturbance strategy is used for adaptive adjustment, combining the fitness differences between individuals, the nonlinear correlation between characteristics and the diversity of the population:

[0115] ;

[0116] in, Represents an individual The disturbance amplitude, represents the basic disturbance intensity, Represents an individual The difference between the fitness and the maximum fitness, represents the maximum fitness in the population, and represents the adjustment parameter that controls the disturbance intensity, represents the number of individuals in the population, Represents an individual and individuals The characteristic differences between represents the nonlinear adjustment parameter, Represents an individual Diversity measures;

[0117] S442. After the disturbance operation, the elite individuals are adaptively repaired, and the characteristic values ​​are adjusted by combining the fitness gradient, the nonlinear interaction between the characteristics and the multi-objective optimization strategy:

[0118] ;

[0119] in, represents the repaired eigenvalue, represents the eigenvalue before perturbation, represents the learning rate, represents the fitness gradient, represents the weight adjustment coefficient, Indicates The interaction measure between features is represents the adjustment parameter, Indicates The regular overlap of features, Indicates A measure of change in a feature;

[0120] S443. Introduce a multi-scale perturbation control strategy for elite individuals. In the early stage of evolution, increase the perturbation intensity to expand the search range; in the middle stage of evolution, gradually reduce the perturbation intensity, increase the perturbation frequency, and use more frequent small-amplitude perturbations to optimize the performance of elite individuals; in the late stage of evolution, further reduce the perturbation intensity and gradually make fine adjustments to the elite individuals;

[0121] S444, elite selection is based on nonlinear correlation and feature coverage, and uses a nonlinear correlation measurement formula to calculate the distribution of individuals in the feature space.

[0122] In this implementation manner, S5 specifically includes:

[0123] S51, construct a three-layer fuzzy system, the first layer is used to process shape feature reasoning, the shape features of the labels are input into the optimized fuzzy rule base, the shape features are mapped to language variables in the fuzzy set through the membership function, and the fuzzy reasoning is performed in combination with the geometric features and boundary information of the shape;

[0124] S52, in the first-level reasoning process, the shape features are fuzzified by using a multi-layer membership function to decompose the complex geometric features into basic shape features. After the basic shape features are processed by different membership functions, the comprehensive membership values ​​used for reasoning are output;

[0125] S53, the second layer processes color feature reasoning, inputs the color features of the labels into the fuzzy rule base, combines the environmental data, performs fuzzy reasoning on the color features, maps the color features into language variables, and compensates and corrects the color changes under different lighting conditions;

[0126] S54. In the second-level reasoning, a multimodal data fusion strategy is applied to jointly process color features and environmental data, and the influence of illumination and color temperature is compensated by dynamically adjusting the shape of the membership function. The fusion weights of color and environmental features are adjusted using an adaptive fuzzy operator to generate an adaptive reasoning result.

[0127] S55, the third layer processes the reasoning of material characteristics of items, inputs the material characteristics of the labels into the fuzzy rule library, and performs fuzzy reasoning on the material characteristics in combination with environmental data; when reasoning about materials, a multidimensional membership function is used to hierarchically process different material characteristics, mapping complex material characteristics into fuzzy language variables.

[0128] In this implementation manner, S6 specifically includes:

[0129] S61. The three-layer reasoning results are integrated through fuzzy operators to perform comprehensive reasoning on shape features, color features, and material features:

[0130] ;

[0131] in, represents the membership value of the comprehensive reasoning result, represents the membership value of the shape feature, Represents the membership value of the color feature, Indicates the membership value of the material feature, represents the uncertainty or error measure of the shape feature, represents the uncertainty or error measure of the color feature, represents the uncertainty or error measure of the material characteristics, , and represents the weight coefficient of the feature, , , and represents the control parameter, , , , , , , and represents the adjustment parameter;

[0132] S62: Calculate the confidence of the final inference result. If the confidence is lower than a preset threshold, the inference result is considered unreliable.

[0133] ;

[0134] in, represents the confidence of inference, represents the final inference result, represents the exponential function, and represents the adjustment parameter, Indicates a preset reference value or target value. represents the adjustment coefficient, Indicates the threshold value;

[0135] S63, adjusting the initial population of the genetic algorithm according to the error of the current reasoning result and the system feedback information, and generating population individuals containing a diverse feature combination;

[0136] S64. Use an improved genetic algorithm to perform multiple iterations of population optimization, combine adaptive mutation and crossover operations to enhance the evolution effect, and use adaptive mutation rate and crossover rate to dynamically adjust the evolution process;

[0137] S65, after multiple iterations of optimization by the genetic algorithm, an optimized fuzzy rule base is generated, and the inference rules and membership functions are reconfigured and adjusted;

[0138] S66. Use the updated fuzzy rule base to re-reason and recalculate the final reasoning result and confidence. If the new reasoning confidence is still lower than the threshold, increase the number of iterations of the genetic algorithm and further optimize the population feature combination and the rule base until the reasoning confidence is higher than the threshold.

[0139] Embodiment 1:

[0140] In order to verify the feasibility of the present invention in implementation, the present invention is applied to a certain automobile parts manufacturing plant (hereinafter referred to as Plant A), where parts on its production line are usually attached with different forms of labels, which are used to trace the production batches, specifications and quality inspection records of the products. However, due to the frequent changes in lighting conditions in the production environment, the presence of oil stains and reflections on the surfaces of some parts, traditional image processing methods and deep learning methods are difficult to accurately identify the content of the labels, especially in the case of uneven lighting or blurred labels, the recognition accuracy is less than 85%, and the stability is poor.

[0141] In this production process, the article label recognition method based on fuzzy logic control of the present invention is applied. The system first collects images of parts on the production line through a high-definition camera, and obtains environmental data such as light, temperature and humidity in combination with the environmental sensors of the production line. The collected images usually include labels with blurred outlines and areas with poor lighting. The method of the present invention makes the outline of the label area clearer by graying, denoising, edge detection and image enhancement, and standardizes the environmental data so that multimodal data can be fused and analyzed.

[0142] In this embodiment, the part numbers of Factory A include various forms of QR codes, bar codes and character combinations. The present invention performs multi-level feature analysis on these numbers. First, the shape features in the image (such as the border of the QR code and the lines of the bar code) are input into the first-level fuzzy inference system for processing, and mapped to language variables in the fuzzy set through the membership function, such as "approximately square", "close to linear", etc., and the membership of the shape features is inferred in combination with the edge information. Then, the second layer infers the color features. Considering that there may be color differences between parts in different batches and the lighting conditions on the production line are complex, the system performs fuzzy inference on the color features combined with the environmental data through the fuzzy rule base to compensate for the color distortion caused by the change in lighting. The third layer is for the reasoning of material features. Combined with the reflectivity and texture features of the surface of the parts, the complex material characteristics are mapped into fuzzy language variables, such as "rough", "smooth", etc., to enhance the system's adaptability to material changes.

[0143] In the process of label recognition, the present invention combines image features with environmental data through a multimodal data fusion strategy to achieve adaptation to complex environments. Test data show that in the production line environment of Factory A, after image preprocessing and multi-level fuzzy reasoning, the accuracy of label recognition reaches more than 94%, which is about 10 percentage points higher than the traditional image processing method. In addition, in areas with large changes in lighting, the recognition success rate of traditional methods is only 75%, while the method of the present invention can increase the success rate to 92%, significantly improving the stability of the system.

[0144] In order to further verify the beneficial effects of the present invention, continuous testing for three months was carried out in Factory A. The test areas include bright areas (light intensity of 1000 to 1500 lux), medium light areas (light intensity of 500 to 800 lux), and darker areas (light intensity below 300 lux). Statistical data show that in bright areas, the accuracy of the label recognition of the present invention is 96.5%, while that of the traditional method is 88.3%; in medium light areas, the accuracy of the present invention is 94.7%, while that of the traditional method is 82.1%; in darker areas, the accuracy of the present invention is 91.8%, while that of the traditional method is only 73.4%. These data show that the present invention has better adaptability in variable lighting environments.

[0145] In addition, the method of the present invention also demonstrated its ability of adaptive evolution in the quality inspection link of Factory A. When the recognition confidence is low, the system will automatically execute the improved genetic algorithm, optimize the fuzzy rule base and reason again. The test results show that after multiple iterations of evolution, the recognition confidence of the system has increased by an average of 12%, especially for parts with strong reflective materials, the recognition accuracy of the system has been significantly improved, from 89% in the initial recognition to more than 94%.

[0146] In the specific application process, the method of the present invention significantly reduces the misjudgment and missed judgment caused by environmental changes, and improves the efficiency of label recognition. In a certain quality inspection process, the number of labels that originally required manual correction was an average of 65 times per day, but after using the method of the present invention, it was reduced to an average of no more than 18 times per day, greatly reducing the frequency and time cost of manual intervention.

[0147] Table 1 Performance comparison of the method of the present invention and the traditional method under different lighting conditions

[0148] index Bright area (1000-1500 lux) Medium light areas (500-800 lux) Dark areas (<300 lux) The accuracy of the label recognition of the present invention (%) 96.5 94.7 91.8 Traditional method label recognition accuracy (%) 88.3 82.1 73.4 Confidence improvement (%) 8.2 12.6 18.4 Number of test samples 1200 1150 1000 The misjudgment rate of the present invention (%) 3.5 5.3 8.2 Misjudgment rate of traditional methods (%) 11.7 17.9 26.6 Reduced number of misjudgments 98 144 184

[0149] From the analysis of Table 1 above, it can be seen that the article label recognition system of the present invention has significant advantages over traditional methods under different lighting conditions. In bright areas (light intensity of 1000-1500 lux), the label recognition accuracy of the present invention reaches 96.5%, which is 8.2 percentage points higher than the 88.3% of the traditional method, showing higher reliability. In addition, in terms of the false positive rate, the present invention is only 3.5%, which is much lower than the 11.7% of the traditional method, reducing the number of false positives by 98, which shows that the system can effectively identify labels under strong lighting conditions.

[0150] In the medium light area (500-800 lux), the accuracy of the label recognition of the present invention is 94.7%, which is 12.6 percentage points higher than the 82.1% of the traditional method, showing strong adaptability to light changes. At the same time, the misjudgment rate is also significantly reduced to only 5.3%, while the misjudgment rate of the traditional method is 17.9%, and the number of misjudgments is reduced by 144, which reflects that the present invention can still maintain high recognition stability and accuracy in complex environments with medium light.

[0151] In the test results of darker areas (light intensity less than 300 lux), the recognition accuracy of the present invention was 91.8%, which was 18.4 percentage points higher than the 73.4% of the traditional method. Despite the poor lighting conditions, the present invention can still maintain a relatively high accuracy. At the same time, the misjudgment rate dropped to 8.2%, while the misjudgment rate of the traditional method was as high as 26.6%, and the number of misjudgments was reduced by 184, showing that the present invention is more robust to labeling in low-light environments, effectively coping with the challenges brought by insufficient light.

[0152] Overall, the article label recognition method of the present invention shows higher recognition accuracy and lower misjudgment rate under various lighting conditions, especially in scenes with large lighting changes or blurred labels, which significantly improves the adaptability and reliability of the system. These data fully demonstrate the superior performance of the present invention in practical applications, and can provide a more stable and accurate label recognition solution for industrial production and quality control in a changing environment.

[0153] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. The method for identifying item numbers based on fuzzy logic control is characterized in that: The steps include: S1, collect image data with item labels and record environmental information to form a multi-dimensional data set; S2, grayscale, denoise, edge detect and image enhance the image data of the multi-dimensional data set, and standardize the environmental data; S3, classifying the labeled sample data according to the improved clustering algorithm to generate an initial fuzzy rule base, wherein the initial fuzzy rule base includes fuzzy sets and membership functions corresponding to the input features; S4. Use the improved genetic algorithm to optimize the initial fuzzy rule base, adopt adaptive mutation and crossover operations, dynamically adjust the mutation rate and crossover rate according to the evolution stage, introduce the elite retention mechanism, combine the label recognition accuracy and rule simplicity to design the multi-objective optimization fitness function, and generate the optimized fuzzy rule base; S5, constructing a three-layer fuzzy system, wherein the first layer processes shape feature reasoning, and infers the shape features of the labels through the optimized fuzzy rule base; The second layer processes color feature reasoning, and performs fuzzy reasoning on color features in combination with environmental data; the third layer processes object material reasoning, and performs fuzzy reasoning on object material in combination with environmental data; S6. The three-layer reasoning results are integrated through the fuzzy operator to obtain the final label recognition result. If the reasoning confidence is low, the genetic algorithm evolution process is re-executed and reasoning is performed again; S7. The label content, reasoning confidence and label recognition result are stored in the database. If the user corrects the recognition result, the improved genetic algorithm is automatically triggered to update the fuzzy rule base and reasoning parameters.

2. The method for identifying item numbers based on fuzzy logic control according to claim 1, characterized in that: The S3 specifically includes: S31, performing feature extraction on the labeled sample data of the preprocessed multidimensional data set, extracting shape features, color features and texture features of the image data, and normalizing the feature data; S32. Use the improved K-Means clustering algorithm based on Gaussian process regression kernel function to cluster and classify the sample data after feature extraction, and use the Gaussian process kernel function to calculate the similarity between data points: ； in, Representation sample To cluster center The distance represents the total number of features, Indicates The weight of the feature, Representation sample No. eigenvalues, Represents the cluster center No. eigenvalues, Represents the Gaussian process kernel function: ； in, represents the exponential function, represents the kernel function width, and represents the adjustment parameter; S33. In the clustering process, an optimization strategy based on GPR kernel entropy is used to adaptively update feature weights. ; S34. Construct an initial fuzzy rule base based on the clustering results, set the fuzzy set and membership function, and the membership function adopts the dual Gaussian-Laplace function fused with the kernel function: ； in, Indicates the input value Compared with fuzzy sets The degree of affiliation, Indicates The center value of the peak, Indicates The width of the peak, represents the number of local peaks in the membership function, represents the parameter that controls the shape of the membership function, represents the weight of the Laplace term that controls the membership function; S35. Redundancy detection and optimization are performed on the rule entries in the fuzzy rule base, and the correlation between the rules is calculated using the kernel entropy interaction formula: ； in, Representation Rules and rules The amount of fuzzy mutual information between Representation Rules In the The membership function value on the features is Representation Rules In the The membership function value on the feature.

3. The method for identifying article numbers based on fuzzy logic control according to claim 1, characterized in that: The S4 specifically includes: S41. An improved genetic algorithm is used to optimize the initial fuzzy rule base, and the population is divided into a global exploration layer and a local fine optimization layer. The global exploration layer focuses on a wide search of the solution space, and the local fine optimization layer focuses on fine-tuning and optimizing high-fitness individuals, focusing on the refinement of the rule structure and local optimization operations. S42: In the adaptive mutation operation, a multi-level dynamic mutation strategy based on the rule association matrix is ​​adopted to adaptively adjust the mutation rate according to the correlation between rules and the interactivity between features: ； in, represents the current mutation rate, represents the number of rule sets, represents the total number of features, Representation Rules and rules The similarity matrix elements between Representation Rules and rules The weight adjustment factor between Represents an individual The fitness value of Represents the maximum fitness in the population; S43. In the multi-level crossover operation, the overall structure of the fuzzy rules and the feature combination within the rules are cross-reorganized, and the crossover probability is adjusted in combination with the multi-objective optimization strategy. The crossover operation combines the traditional global crossover and the high-order asymmetry and correlation between the features for dynamic reorganization: ； in, represents the current crossover rate, represents the number of high-order feature combinations, Indicates The weight of the high-order feature combination, represents the exponential function, and represents the adjustment parameter, Indicates The distance metric for feature combinations is represents an asymmetric correlation measure for high-order feature combinations, represents the number of structural complexity indicators, Indicates The weight of the complexity index, represents the inverse tangent function, Indicates The value of a complexity indicator; S44, introduce elite retention and structured perturbation mechanisms, give priority to retaining individuals with higher fitness in each generation, and perform structured perturbations on elite individuals based on rule complexity and population diversity. The perturbation strategy adopts a dual perturbation mechanism; S45, using multi-objective fitness function for comprehensive evaluation, combined with the label recognition accuracy, simplicity of fuzzy rules and complex interactions between rules for optimization: ； in, represents the fitness value of an individual, represents the number of error indicators considered during the optimization process, Indicates error terms, represents the total number of features, Indicates The weight of the feature interaction, Indicates The interaction measure between features is represents the adjustment parameter, represents the number of overlap indicators considered during the optimization process, Indicates The degree of overlap of rules; S46. After multi-level evolution, adaptive operation and multi-objective optimization strategy, an optimized fuzzy rule base is finally generated.

4. The method for identifying article numbers based on fuzzy logic control according to claim 3, characterized in that: The S44 specifically includes: S441. For the disturbance amplitude of elite individuals, a dual disturbance strategy is used for adaptive adjustment, combining the fitness differences between individuals, the nonlinear correlation between characteristics and the diversity of the population: ； in, Represents an individual The disturbance amplitude, represents the basic disturbance intensity, Represents an individual The difference between the fitness and the maximum fitness, represents the maximum fitness in the population, and represents the adjustment parameter that controls the disturbance intensity, represents the number of individuals in the population, Represents an individual and individuals The characteristic differences between represents the nonlinear adjustment parameter, Represents an individual Diversity measures; S442. After the disturbance operation, the elite individuals are adaptively repaired, and the characteristic values ​​are adjusted by combining the fitness gradient, the nonlinear interaction between the characteristics and the multi-objective optimization strategy: ； in, represents the repaired eigenvalue, represents the eigenvalue before perturbation, represents the learning rate, represents the fitness gradient, represents the weight adjustment coefficient, Indicates The interaction measure between features is represents the adjustment parameter, Indicates The regular overlap of features, Indicates A measure of change in a feature; S443. Introduce a multi-scale perturbation control strategy for elite individuals. In the early stage of evolution, increase the perturbation intensity to expand the search range; in the middle stage of evolution, gradually reduce the perturbation intensity, increase the perturbation frequency, and use more frequent small-amplitude perturbations to optimize the performance of elite individuals; in the late stage of evolution, further reduce the perturbation intensity and gradually make fine adjustments to the elite individuals; S444, elite selection is based on nonlinear correlation and feature coverage, and uses a nonlinear correlation measurement formula to calculate the distribution of individuals in the feature space.

5. The method for identifying article numbers based on fuzzy logic control according to claim 1, characterized in that: The S5 specifically includes: S51, construct a three-layer fuzzy system, the first layer is used to process shape feature reasoning, the shape features of the labels are input into the optimized fuzzy rule base, the shape features are mapped to language variables in the fuzzy set through the membership function, and the fuzzy reasoning is performed in combination with the geometric features and boundary information of the shape; S52, in the first-level reasoning process, the shape features are fuzzified by using a multi-layer membership function to decompose the complex geometric features into basic shape features. After the basic shape features are processed by different membership functions, the comprehensive membership values ​​used for reasoning are output; S53, the second layer processes color feature reasoning, inputs the color features of the labels into the fuzzy rule base, combines the environmental data, performs fuzzy reasoning on the color features, maps the color features into language variables, and compensates and corrects the color changes under different lighting conditions; S54. In the second-level reasoning, a multimodal data fusion strategy is applied to jointly process color features and environmental data, and the influence of illumination and color temperature is compensated by dynamically adjusting the shape of the membership function. The fusion weights of color and environmental features are adjusted using an adaptive fuzzy operator to generate an adaptive reasoning result. S55, the third layer processes the reasoning of material characteristics of items, inputs the material characteristics of the labels into the fuzzy rule library, and performs fuzzy reasoning on the material characteristics in combination with environmental data; when reasoning about materials, a multidimensional membership function is used to hierarchically process different material characteristics, mapping complex material characteristics into fuzzy language variables.

6. The method for identifying article numbers based on fuzzy logic control according to claim 1, characterized in that: The S6 specifically includes: S61. The three-layer reasoning results are integrated through fuzzy operators to perform comprehensive reasoning on shape features, color features, and material features: ； in, represents the membership value of the comprehensive reasoning result, represents the membership value of the shape feature, Represents the membership value of the color feature, Indicates the membership value of the material feature, represents the uncertainty or error measure of the shape feature, represents the uncertainty or error measure of the color feature, represents the uncertainty or error measure of the material characteristics, , and represents the weight coefficient of the feature, , , and represents the control parameter, , , , , , , and represents the adjustment parameter; S62: Calculate the confidence of the final inference result. If the confidence is lower than a preset threshold, the inference result is considered unreliable. ； in, represents the confidence of inference, represents the final inference result, represents the exponential function, and represents the adjustment parameter, Indicates a preset reference value or target value. represents the adjustment coefficient, Indicates the threshold value; S63, adjusting the initial population of the genetic algorithm according to the error of the current reasoning result and the system feedback information, and generating population individuals containing a diverse feature combination; S64. Use an improved genetic algorithm to perform multiple iterations of population optimization, combine adaptive mutation and crossover operations to enhance the evolution effect, and use adaptive mutation rate and crossover rate to dynamically adjust the evolution process; S65, after multiple iterations of optimization by the genetic algorithm, an optimized fuzzy rule base is generated, and the inference rules and membership functions are reconfigured and adjusted; S66. Use the updated fuzzy rule base to re-reason and recalculate the final reasoning result and confidence. If the new reasoning confidence is still lower than the threshold, increase the number of iterations of the genetic algorithm and further optimize the population feature combination and the rule base until the reasoning confidence is higher than the threshold.

Images