A method and system for detecting and identifying images of electric power materials

Abstract

This application discloses a method and system for detecting and recognizing images of power materials, relating to the field of power material detection technology. The method includes: S1, acquiring images of power materials during their movement, sorting the images according to the acquisition process and location to obtain a set of material images related to the process location; S2, performing image detection on the material image set to obtain a feature set corresponding to the material image set, performing linear regression analysis on the feature set, clustering the feature set according to the linear regression analysis results to obtain feature clustering results, and setting a power material label for the current power material based on the feature clustering results; S3, identifying the relative position of the feature vectors corresponding to the feature clustering results on the power material images, and obtaining the feature coverage and overlap ratio of the feature clustering results on each face of adjacent power material images; this method can improve the accuracy and efficiency of power material recognition.

Description

Technical Field

[0001] This invention relates to the field of power material inspection technology, and in particular to a method and system for detecting and recognizing images of power materials. Background Art

[0002] Power material warehousing management is an important component of power enterprise management. It involves aspects such as warehousing planning, inbound and outbound management, and warehousing environment regulation. Among these, the management of power material inbound and outbound operations is essential for ensuring the normal operation of power production and dispatch. Optimizing this management can improve the operational efficiency and profitability of power production enterprises.

[0003] For example, Chinese Patent Publication No. CN114971501A discloses a monitoring and guidance analysis system for the entry and exit of power materials based on feature analysis. Before the handling robot performs the entry operation on each power material, it collects images of the stacking status of the materials and identifies the stacking height of each material. At the same time, it identifies the entry distance of each material based on the designated sub-warehouse area of ​​each power material. Based on the entry distance and stacking height of each material, it comprehensively plans the entry sequence, which takes into account the principle of entry to the nearest warehouse and avoids damage to the power materials before use, thus achieving dual protection of entry efficiency and entry safety. Meanwhile, when the handling robot transports the power materials, it realizes the integrated operation of monitoring, early warning and handling of falling power materials during the entry and transportation process.

[0004] For example, Chinese Patent Publication No. CN116308047A discloses a power material inbound and outbound management system based on RFID technology, which relates to the field of material management technology. It solves the technical problem that when a large number of power materials are stored at the same time, some power material packages may be disassembled for unknown reasons, resulting in the replacement of internal materials and thus causing logistics losses. Based on the overall image of the surface where the RFID tag is located, the area ratio parameter of the corresponding area is obtained. The same method is used to process the data during the outbound process. Under different turning conditions, the obtained area ratio parameter is always consistent and there will be no numerical error due to different overall image orientations. At the same time, by processing in this way, it is possible to quickly identify whether there are missing or disassembled packages in the power material outbound process.

[0005] In existing technologies, power materials are identified by parameters such as their order of entry into the warehouse and their length. However, when multiple power materials with similar characteristics are piled up in one location, it is easy to overlook the differences between these power materials, leading to identification errors during power material tracking and identification, and reducing the efficiency of power material management. Summary of the Invention

[0006] This application provides a method and system for detecting and recognizing images of power materials, which solves the technical problem of accuracy in tracking power materials in the prior art and improves the accuracy of power material recognition.

[0007] This application provides a method for detecting and recognizing images of power materials, including:

[0008] S1: Acquire images of power materials during material movement, sort the images of power materials according to the acquired process and location, and obtain a set of material images related to the process location.

[0009] S2, perform image detection on the material image set, obtain the feature set corresponding to the material image set, perform linear regression analysis on the feature set, cluster the feature set according to the linear regression analysis results, obtain the feature clustering results, and set the power material label for the current power material based on the feature clustering results.

[0010] S3. Identify the relative position of the feature vector corresponding to the feature clustering result on the power material image, and obtain the feature coverage and overlap area ratio of the feature clustering result on each face of the adjacent power material image. Match the feature coverage and overlap area ratio with the power material label to obtain the comprehensive processing label of the current power material under different priorities.

[0011] S4, based on comprehensive tag processing, tracks the image of power materials, determines the current location and probability of allocation of power materials, and obtains the allocation feedback results of power materials.

[0012] This invention provides a system for detecting and recognizing images of power materials, comprising:

[0013] The image acquisition module is used to acquire images of power materials during their movement and sort the images according to the acquired process and location to form a set of material images related to the process location.

[0014] The image detection module is used to perform image detection on the material image set and extract the feature set; construct a linear regression model to perform linear regression analysis on the feature set; cluster the feature set according to the linear regression analysis results to obtain the feature clustering results; and set the power material label for the current power material based on the feature clustering results.

[0015] The feature processing module is used to identify the relative position of the feature vectors corresponding to the feature clustering results on the power material image; calculate the feature coverage and overlap area ratio, and match them with the power material labels; combine the attitude angle of the power materials to generate the comprehensive matching result of the power materials; and obtain the comprehensive processing label of the current power materials under different priorities based on the comprehensive matching result.

[0016] The location verification module is used to obtain the current allocation location of power materials, extract the tracking displacement value, verify whether the power materials are in the optimal location point, calculate the displacement evaluation coefficient, and output the allocation feedback result.

[0017] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0018] Image detection and feature clustering technologies can quickly and accurately identify the appearance, transportation requirements, and batch processing characteristics of power materials; by calculating feature coverage and overlap area ratio, combined with the attitude angle of the power materials, precise matching of feature vectors can be achieved; comprehensive processing labels are generated based on the matching results, providing a reliable basis for the subsequent processing of power materials; by calculating displacement evaluation coefficients, it can be verified whether the power materials are in the optimal position; and the allocation and management of power materials can be optimized. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a method for detecting and recognizing images of electrical materials according to the present invention.

[0020] Figure 2 This is a flowchart illustrating step S2 of the method for detecting and recognizing images of power materials according to the present invention.

[0021] Figure 3 A flowchart illustrating step S3 of the method for detecting and recognizing images of power materials according to the present invention.

[0022] Figure 4 A flowchart illustrating step S35 of the method for detecting and recognizing images of power materials according to the present invention.

[0023] Figure 5 A flowchart illustrating step S4 of the method for detecting and recognizing images of power materials according to the present invention.

[0024] Figure 6 The present invention provides a system framework diagram for a power material image detection and recognition system. Detailed Implementation

[0025] To facilitate understanding of the present invention, a more complete description of this application will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to enable a more thorough and complete understanding of the disclosure of the present invention.

[0026] It should be noted that the terms "vertical," "horizontal," "up," "down," "left," "right," and similar expressions used in this article are for illustrative purposes only and do not represent the only possible implementation.

[0027] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention; the term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0028] In this invention, when organizing images of power materials, if the power materials are identified based on their length, width, height, and area, it is easy to overlook the condition of the power materials during transportation. For example, if the power materials need to be transported smoothly, it is necessary to verify the smoothness of the power materials during transportation. If the materials need to be transported quickly, it is necessary to pay attention to the batch and priority of the materials. Therefore, when identifying power materials, it is necessary to prioritize obtaining the order and features corresponding to the materials and represent the power materials using features to complete the rapid identification of the materials.

[0029] Therefore, when processing power materials, this invention first constructs a linear regression model for the power materials, performs regression analysis on the features extracted from the power materials, sequentially determines the linear relationships between the current power materials, and identifies and processes the power materials according to the feature sets. Then, it clusters the data in the power materials according to the current processing method, and determines the additional features of the current materials based on the position of the image's acquisition face and the six-sided diagram in which the power materials are located. These features obtained from different faces are then matched with the material's own label to obtain the matching degree and priority correlation degree of the material under a specific priority condition, thereby obtaining a comprehensive processing label for the material. Based on this label, the current power materials are processed for warehousing and retrieval, thus completing the rapid identification and processing of power materials. The matching degree represents the matching situation between the power materials and their own labels, while the priority correlation degree represents the influence and priority representation of the current power material's features on the six-sided diagram relative to the current power material.

[0030] Meanwhile, these processing methods tend to focus on the posture recognition of electronic and electrical materials themselves, so as to classify electrical materials with similar shapes or sizes together, which helps to optimize the use of storage space and reduce the risk of collision and damage between electrical materials. At the same time, based on the posture of the electrical materials themselves, it is possible to locate problems related to electrical materials more quickly, and promptly identify improperly transported electrical materials, which can reduce the problems of electrical materials collapsing or being mishandled during placement.

[0031] like Figure 1 As shown, the method for detecting and recognizing images of power materials according to this application includes:

[0032] S1: Acquire images of power materials during material movement, sort the images of power materials according to the acquired process and location, and obtain a set of material images related to the process location.

[0033] S2, perform image detection on the material image set, obtain the feature set corresponding to the material image set, perform linear regression analysis on the feature set, cluster the feature set according to the linear regression analysis results, obtain the feature clustering results, and set the power material label for the current power material based on the feature clustering results.

[0034] like Figure 2 As shown, step S2 includes: S21, obtaining appearance recognition features, demand features for the transportation of power materials, and batch processing features from the feature set; constructing a linear regression model related to appearance recognition features, demand features, and batch processing features. Appearance features represent the feature information related to the appearance and texture of the current power materials. Demand features display the feature information of the current power materials when they are placed, stored, transported, and used. Batch processing features represent the priority of the current materials being processed and the logical order of the materials during processing.

[0035] For example, appearance recognition features refer to visual features extracted from images of electrical materials that can be directly used to identify and classify objects; these features are usually related to the physical properties of the objects, such as shape, color, and texture.

[0036] Visual identification features can take the following forms:

[0037] Color histogram: Describes the color distribution on the surface of electrical materials and can be used to distinguish different types of equipment or components.

[0038] Edge detection: Detects the outline of objects using the Canny operator or other methods to help determine the shape and structure of electrical materials.

[0039] Texture analysis: Using techniques such as LBP (Local Binary Pattern) to quantify surface texture characteristics, suitable for identifying specific materials or manufacturing processes.

[0040] Brand logo / barcode: Locating and reading the identification information on electrical equipment helps to quickly confirm its identity.

[0041] Shape descriptors, such as Hu moments and Fourier descriptors, can be used to describe complex shapes, facilitating accurate matching and classification.

[0042] For example, demand characteristics reflect the functional requirements and technical specifications of power materials in specific application scenarios; these characteristics usually involve information such as performance parameters, installation conditions, and maintenance standards.

[0043] Demand characteristics can include the following:

[0044] Power rating: This indicates the maximum electrical energy output that electrical equipment can provide, and is especially important for critical equipment such as transformers and generators.

[0045] Voltage range: This specifies the applicable operating voltage range for electrical equipment, ensuring correct selection and safe use.

[0046] Protection rating: Describes the ability of electrical equipment to protect against external environmental factors (such as dust and water), and affects the selection of its deployment location.

[0047] Size and weight: These determine whether electrical equipment is suitable for specific space constraints and how it is handled.

[0048] Connection interface type: such as bolt connection, plug-in connector, etc., affects compatibility with other systems and ease of installation.

[0049] Service life and warranty period: Provides information on the long-term reliability of power equipment to guide procurement decisions and service plans.

[0050] Batch processing characteristics refer to attributes associated with the same production batch, including but not limited to production date, supplier information, and quality inspection results. These characteristics are mainly used for supply chain management and quality traceability.

[0051] For example, batch processing characteristics can take the following form:

[0052] Production batch number: Each batch has a unique number, which makes it easy to track the flow and history of all products in that batch.

[0053] Production date: Marking the manufacturing date of electrical materials helps assess their freshness or shelf life.

[0054] Supplier information: Records the names and addresses of companies that provide raw materials or finished products, supporting transparent supply chain management.

[0055] Quality inspection report: Includes various test data and certificates of conformity, ensuring that the product quality meets the established standards.

[0056] Logistics information: Track the transportation path of power materials from the factory to the end user to ensure timely delivery and reduce the risk of loss.

[0057] Inventory status: Real-time updates on changes in the quantity of power supplies to assist in warehouse management and allocation decisions.

[0058] These features can be obtained using image recognition algorithms, and power inspection robots can be used to acquire real-time information on power materials that are currently being transported or have been stored. The identified features can then be sent to an external processing terminal for comprehensive processing of the images of these identified power materials.

[0059] S22, analyze the appearance recognition features, demand features and batch processing features in sequence, and obtain the first correlation coefficient corresponding to the appearance recognition features, the second correlation coefficient corresponding to the demand features and the third correlation coefficient corresponding to the batch processing features in sequence.

[0060] The calculation method for the first correlation coefficient is as follows: taking the appearance recognition features in the material image set as input, the cosine similarity between the appearance recognition features in adjacent power material images is judged sequentially. The cosine similarity is calculated by representing the appearance recognition features in vector form to obtain the cosine similarity. The calculated cosine similarity is used as the first correlation coefficient. The calculation methods for the second and third correlation coefficients are the same, both using the cosine similarity of the corresponding demand features and batch processing features in adjacent power material images to judge the similarity between related images in adjacent power material images.

[0061] S23, perform linear regression on the first correlation coefficient, the second correlation coefficient, and the third correlation coefficient to obtain the regression coefficients corresponding to the first correlation coefficient, the second correlation coefficient, and the third correlation coefficient.

[0062] The regression coefficients are obtained by using the second and third correlation coefficients as independent variables and the first correlation coefficient as the dependent variable, and performing linear regression analysis to determine the corresponding changes in the actual stored and transported materials within the adjacent power material images when the current priority of material transportation and transmission changes.

[0063] When performing linear regression analysis, the main focus is on determining the probability of the values ​​corresponding to the first, second, and third correlation coefficients appearing in continuous images of electrical materials, and comparing the relative positions of the existing image features with respect to continuous image recognition, in order to complete the linear regression analysis of the first, second, and third correlation coefficients.

[0064] For example, extract the first correlation coefficient C1, the second correlation coefficient C2, and the third correlation coefficient C3 corresponding to the features from the feature set; and form a linear regression model using the first correlation coefficient, the second correlation coefficient, and the third correlation coefficient.

[0065] For the appearance recognition features, demand features, and batch processing features in the feature set, these features are mapped one by one, and the first correlation coefficient, second correlation coefficient, and third correlation coefficient related to these features are obtained. These coefficients are then calculated to obtain a linear regression model and linear regression coefficients.

[0066] C1 = β0 + β1C2 + β2C3 + ε;

[0067] Where β0 represents the intercept term, β1 and β2 represent the regression coefficients, and ε represents the error term.

[0068] The obtained regression coefficients are estimated using the least squares method to obtain the minimum residual sum of squares (RSS(β)) of the regression coefficients.

[0069]

[0070] And use a matrix to obtain the optimal regression coefficient values.

[0071]

[0072] Where X represents the design matrix, Y contains all independent variables C2 and C3; and X represents the vector of dependent variable C1. T This represents the transpose operation of the design matrix; the optimal regression coefficient value calculated here represents the optimal regression coefficient value corresponding to the data under the corresponding number of independent and dependent variables, and this optimal regression coefficient value is substituted into the intercept term of minimizing the sum of squared residuals to obtain a set of regression coefficients most relevant to the current independent and dependent variables.

[0073] Specifically X and Y contain n rows of values, each row representing a set of first, second, and third correlation coefficients.

[0074] When both the minimum sum of squared residuals and the optimal regression coefficient value are satisfied, the corresponding regression coefficients are output, thereby completing the combined analysis of the first, second, and third correlation coefficients.

[0075] When obtaining the output regression coefficients, it is also necessary to determine the first correlation coefficient, the second correlation coefficient, and the third correlation coefficient under the corresponding regression coefficient, and compare the relative positions of the first correlation coefficient, the second correlation coefficient, and the third correlation coefficient with respect to the adjacent power material images. The first correlation coefficient, the second correlation coefficient, and the third correlation coefficient with the smallest relative position distance and the highest probability of occurrence are used as the initial cluster centers for subsequent cluster calculations.

[0076] S24. Based on the range of regression coefficient values, multiple clusters are set. The initial cluster center of each cluster is represented by the weighted average of the first correlation coefficient, the second correlation coefficient, and the third correlation coefficient. The distance between each element in the feature set and the initial cluster center is calculated, i.e., the distance between the feature vector in the feature set and the initial cluster center is calculated. The first correlation coefficient, the second correlation coefficient, and the third correlation coefficient of all members in each cluster are calculated. The weighted average of these correlation coefficients is used to recalculate the new center position of the cluster. The cluster center is iterated continuously until the cluster center meets the convergence condition. The convergence condition is to reach the preset maximum number of iterations, which can be set by selecting the average value of power material image processing in historical data. The iterated clusters are output as feature clustering results.

[0077] The role of regression coefficients is to guide the setting of initial clusters and ensure that different clusters have reasonable distinguishability; by considering the influence of multiple correlation coefficients, the cluster centers can comprehensively reflect the correlation between members.

[0078] After completing the clustering process, power material labels are set for each element in the feature clustering results.

[0079] S3. Identify the relative position of the feature vector corresponding to the feature clustering result on the power material image, and obtain the feature coverage and overlap area ratio of the feature clustering result on each face of the adjacent power material image. Match the feature coverage and overlap area ratio with the power material label to obtain the comprehensive processing label of the current power material under different priorities.

[0080] When acquiring feature coverage and overlap ratio, the system calculates the feature coverage and overlap ratio at the current attitude angle of the power equipment to identify the specific placement of the power equipment. Based on these identified placement details, the system performs comprehensive identification of the power equipment to complete the identification process.

[0081] Simultaneously, it compares the faces of adjacent power material images with respect to the current placement of the power material, that is, the relative position of the position of the power material image with the current power material among the six faces of the power material, and identifies the distribution of the corresponding features in the feature clustering results on these six faces, so as to identify which face features are mainly relied upon for identification when the current power material is identified in a centralized manner.

[0082] like Figure 3 As shown, step S3 is implemented in the following ways:

[0083] S31, obtain the feature clustering results of the current power materials, and map the feature clustering results onto the six surfaces corresponding to the power materials to determine the relative position of each feature vector in the feature clustering results on the power materials.

[0084] S32, calculate the feature coverage of the feature vector on each face. The feature coverage represents the ratio of the proportion of pixels occupied by the current feature vector to the sum of the proportions of pixels occupied by all feature vectors on that face.

[0085] S33, calculate the overlap area ratio between the feature vector and the adjacent power material image under the corresponding feature coverage. The overlap area ratio represents the intersection-union ratio of the contours of multiple faces in the power material image.

[0086] S34, obtain the current attitude angle of the power material, compare the feature coverage and overlap area ratio of the feature vector with the current attitude angle of the power material, and determine the matching result of the feature vector and attitude angle.

[0087] S35. Based on the matching results of feature vectors and attitude angles, the matching results on the six surfaces corresponding to the power materials are combined to generate a comprehensive matching result for the power materials; based on the comprehensive matching result of the power materials, a comprehensive processing label is obtained.

[0088] In this step, the relative position of the current power material in the six faces is identified by comparing the feature distribution in adjacent power material images; the distribution of the corresponding features in the feature clustering results on these six faces is analyzed to find the most representative face and the features on it.

[0089] The feature coverage mentioned above can be expressed as: Among them, C f Ft represents the feature coverage on the f-th face. f Ft represents the proportion of pixels occupied by the feature vector on the f-th face. f,total This represents the sum of the pixel proportions of all feature vectors on the f-th face; the feature coverage calculated here is for a single feature vector to identify which feature vector stands out in the current face, so as to quickly identify the existing power materials based on the differences of the six faces corresponding to the goods.

[0090] The overlap ratio (IRR) is the ratio of the contours of two faces. It measures the degree of overlap between two different faces or regions, particularly when evaluating feature matching between adjacent images of electrical materials. It assesses the consistency of features of the same object from different viewpoints. For example, when comparing different faces of the same electrical material in two adjacent images, identifying that these faces belong to the same object and maintain consistent relative positions, and when multiple faces possess important features, the IRR can indicate which face provides more information.

[0091] When determining the comparison result between the feature vector and the attitude angle, the absolute value of the difference between the feature coverage and overlap area ratio of the feature vector under the absence of an attitude angle and the feature coverage and overlap area ratio under the current attitude angle is taken as the matching result between the feature vector and the attitude angle at this time; then, the comprehensive matching result of the power materials is obtained based on the matching result between the feature vector and the attitude angle.

[0092] The matching result can be expressed as the matching difference value of the feature vector at the corresponding attitude angle, the matching score of the current feature vector is calculated based on the obtained matching difference value, and the matching score is output as the matching result of the feature vector and the attitude angle.

[0093] The matching difference value of feature coverage is represented as ΔC. f =|C f _C f,θ |;wherein, ΔC f The matching difference value represents the feature coverage on the f-th face, θ represents the pose angle, and C f,θ This represents the feature coverage when the attitude angle θ exists on the f-th surface.

[0094] At this point, the relevant attitude angles of the power materials are compared to identify the actual placement of these power materials and the corresponding situation of the images of the power materials captured by the current inspection equipment. Since the inspection equipment does not completely translate with the power materials when capturing images, it is necessary to identify the specific characteristics of the power materials when the placement is offset or the viewing angle is offset during the shooting, so as to quickly identify these related power materials.

[0095] The matching difference value of the overlap area ratio is expressed as ΔI. ff′ =|IoU ff′ _IoU ff′,θ |;wherein, ΔI ff′ IoU represents the matching difference value between the overlapping area ratio of the f-th face and the f′-th face. ff′ Represents the f-th face and the f-th face ′ The ratio of the overlapping areas of the faces, IoU ff′,θ Represents the f-th face and the f-th face ′ There are overlapping areas of attitude angle θ on each surface, and f′ represents the other surfaces besides surface f.

[0096] The matching score at this point is represented as follows: S f The matching score for the f-th face is represented by F, where F represents the number of faces present in the power material image, e represents the exponential constant, and α represents the adjustment coefficient, which controls the impact of differences on the matching score; W f W represents the weight of the f-th face.f′ This represents the weight of the f′-th face. The weights of different faces on power materials can be set according to the ratio of the feature vector on the corresponding face to the standard feature vector. The standard feature vector is the average value of the feature vectors set in historical data. The values ​​of f and f′ are both from 1 to F.

[0097] The comprehensive matching result of power materials can be expressed as follows: the comprehensive matching score is obtained by comprehensively calculating the matching results of feature vectors and attitude angles, and the comprehensive matching score is output as the comprehensive matching result.

[0098]

[0099] Among them, S total This represents the comprehensive matching score. The comprehensive matching score obtained at this time can reflect the comprehensive situation of the power materials identified in the current power material image on multiple surfaces. These identified comprehensive situations are output to represent the corresponding comprehensive features of the current power materials. According to the value corresponding to this comprehensive matching result, a comprehensive processing label is set for the currently identified power materials to represent the corresponding features of the current power materials.

[0100] When setting up integrated processing labels, it is also necessary to match the integrated matching results of power materials with the power material labels in order to select the corresponding integrated processing label.

[0101] like Figure 4 As shown, step S35 also includes S351, which involves matching the comprehensive matching result of the power materials with the power material tags to obtain at least one matching point.

[0102] First, the system obtains the matching points between the comprehensive matching results and the power material labels. These matching points represent a certain association or similarity between the comprehensive matching results and the power material labels. Next, the system assigns these matching points a unique identifier and records their corresponding information, such as the power material label and the comprehensive matching result.

[0103] In this step, the matching method between power material tags and comprehensive matching results is restricted according to the co-occurrence probability of power material tags and comprehensive matching results, and matching points are set according to the co-occurrence probability of power material tags and comprehensive matching results. Each matching point corresponds to a co-occurrence probability value of power material tags and comprehensive matching results.

[0104] S352, perform matching point allocation processing, sequentially identify the first matching threshold corresponding to the matching point; and filter out multiple target points from the matching points according to the first matching threshold.

[0105] For each matching point, the system identifies its corresponding first matching threshold. This threshold may be a preset value used to measure whether the matching point meets a certain condition or standard. Based on the first matching threshold, the system filters out multiple target points that meet the conditions from all matching points. These target points represent power material labels that are closest to or similar to the comprehensive matching result.

[0106] In this step, a threshold is set for the co-occurrence probability of power material tags and comprehensive matching results. This threshold is the first matching threshold. The first matching threshold can be set by selecting the average co-occurrence probability of power material tags and comprehensive matching results in historical data. Matching points that are greater than the first matching threshold are set as target points.

[0107] S353 performs a difference analysis on the target point and uses the power material label corresponding to the point with the largest difference value after the difference analysis as the output comprehensive processing label.

[0108] Difference analysis is a statistical method used to compare data differences between different samples or groups. For power supply labels, difference analysis might involve comparing differences in quantity, quality, price, supply capacity, etc., of power supplies corresponding to different labels. It calculates the difference value between each target point and other target points, and identifies the point with the largest difference value. This point represents the power supply label with the largest difference from the overall matching result.

[0109] This step involves calculating the corresponding values ​​at different target points. The value set at the target point will be selected from the feature vector value corresponding to the comprehensive processing result and the power material label, and the power material label with the largest difference value will be selected to obtain the output comprehensive processing label.

[0110] S4, based on comprehensive tag processing, tracks the image of power materials, determines the current location and probability of allocation of power materials, and obtains the allocation feedback results of power materials.

[0111] The allocation location involves verifying the specific coordinates of the power materials in the corresponding spatial location to determine whether the location of the power materials will change after changes occur or after the corresponding deployment settings are completed. These changes in displacement values ​​are regarded as tracking displacement values. The allocation probability is to determine the probability of the power materials being allocated to different locations under the current feature vector of the power materials. Combining this probability value can help to identify the location where the current power materials need to be placed in a timely manner.

[0112] like Figure 5 As shown, the implementation method of this step includes: S41, obtaining the current allocation location of power materials and extracting the tracking displacement value related to the allocation location;

[0113] S42 combines the tracking displacement value with the allocation probability of power materials to verify whether the power materials are at the optimal location point; and calculates the displacement evaluation coefficient of the power materials at the optimal location point, and outputs the data corresponding to the displacement evaluation coefficient as the allocation feedback result.

[0114] The method to verify whether the power materials are in the optimal position in this step is to obtain the current displacement deviation of the power materials, combine the displacement deviation with the allocation probability, and calculate the displacement evaluation coefficient.

[0115]

[0116] Where δ represents the displacement evaluation coefficient, D represents the tracking displacement value, ΔDI represents the displacement deviation, and ΔDI represents the standard value of the displacement deviation. In this case, the displacement deviation represents the difference between the currently acquired tracking displacement value and the predicted tracking displacement value. The standard value of the displacement deviation is the average value of the displacement deviation in historical data. P(DI) represents the allocation probability of power materials, and α1, α2 and α3 represent the weighting coefficients. The values ​​of the three weighting coefficients α1, α2 and α3 are set to 0.25, 0.35 and 0.4 in sequence.

[0117] At this stage, the actual displacement of power equipment and its distribution probability in different locations are considered. Whenever the location of power equipment changes, the new coordinates are immediately written into the database. The newly identified coordinates are tracked, the distribution probability of power equipment in different locations is evaluated, the optimal placement location is selected, and real-time feedback is generated. Finally, all location changes are recorded in the database, and any abnormal situation triggers the corresponding alarm mechanism to ensure the safety and reliability of the entire process.

[0118] The obtained displacement evaluation coefficients will reflect the relative distribution of these power materials during tracking when using the identification of the six sides of the power materials mentioned above. This will allow for real-time updates of the position coordinates of the power materials, timely verification of the accuracy of the currently identified power materials, and the output allocation feedback results will include the currently identified displacement evaluation coefficients and the feature vectors of the corresponding power materials on different sides, thereby assisting external terminals in identifying power materials.

[0119] like Figure 6 As shown, the present invention also provides a system for detecting and recognizing images of power materials, comprising:

[0120] The image acquisition module is used to acquire images of power materials during their movement and sort the images according to the acquired process and location to form a set of material images related to the process location.

[0121] The image detection module is used to perform image detection on the material image set and extract the feature set; construct a linear regression model to perform linear regression analysis on the feature set; cluster the feature set according to the linear regression analysis results to obtain the feature clustering results; and set the power material label for the current power material based on the feature clustering results.

[0122] The feature processing module is used to identify the relative position of the feature vectors corresponding to the feature clustering results on the power material image; calculate the feature coverage and overlap area ratio, and match them with the power material labels; combine the attitude angle of the power materials to generate the comprehensive matching result of the power materials; and obtain the comprehensive processing label of the current power materials under different priorities based on the comprehensive matching result.

[0123] The location verification module is used to obtain the current allocation location of power materials, extract the tracking displacement value, verify whether the power materials are in the optimal location point, calculate the displacement evaluation coefficient, and output the allocation feedback result.

[0124] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting and recognizing images of power materials, characterized in that, include: S1, acquire images of power materials during material movement, sort the power material images according to the acquired process and location, and obtain a set of material images related to the process location; S2, perform image detection on the material image set, obtain the feature set corresponding to the material image set, perform linear regression analysis on the feature set, cluster the feature set according to the linear regression analysis results, obtain the feature clustering results, and set the power material label for the current power material based on the feature clustering results; S3, identify the relative position of the feature vector corresponding to the feature clustering result on the power material image, and obtain the feature coverage and overlap area ratio of the feature clustering result on each face of the adjacent power material image. Match the feature coverage and overlap area ratio with the power material label to obtain the comprehensive processing label of the current power material under different priorities. S4, based on the comprehensive processing of tags, tracks the image of power materials, determines the current location and probability of allocation of power materials, and obtains the allocation feedback results of power materials; Step S3 can be implemented in the following ways: S31, obtain the feature clustering results of the current power materials, and map the feature clustering results onto the six faces corresponding to the power materials, and determine the relative position of each feature vector in the feature clustering results on the power materials; S32, calculate the feature coverage of the feature vector on each face. The feature coverage represents the ratio of the proportion of pixels occupied by the current feature vector to the sum of the proportions of pixels occupied by all feature vectors on that face. S33, calculate the overlap area ratio between the feature vector and the adjacent power material image under the corresponding feature coverage. The overlap area ratio represents the intersection-union ratio of the contours of multiple faces in the power material image. S34, obtain the current attitude angle of the power material, compare the feature coverage and overlap area ratio of the feature vector with the current attitude angle of the power material, and determine the matching result of the feature vector and attitude angle. S35. Based on the matching results of feature vectors and attitude angles, the matching results on the six surfaces corresponding to the power materials are combined to generate a comprehensive matching result for the power materials; based on the comprehensive matching result of the power materials, a comprehensive processing label is obtained.

2. The method for detecting and recognizing images of power materials as described in claim 1, characterized in that, Step S2 includes: S21, Obtain the appearance recognition features for the current appearance of power materials, the demand features for the transportation of power materials, and the batch processing features from the feature set; Construct a linear regression model related to the appearance recognition features, demand features, and batch processing features; S22, analyze the appearance recognition features, demand features and batch processing features in sequence, and obtain the first correlation coefficient corresponding to the appearance recognition features, the second correlation coefficient corresponding to the demand features and the third correlation coefficient corresponding to the batch processing features in sequence. S23, perform linear regression on the first correlation coefficient, the second correlation coefficient, and the third correlation coefficient to obtain the regression coefficients corresponding to the first correlation coefficient, the second correlation coefficient, and the third correlation coefficient; S24. According to the range of values ​​of the regression coefficients, multiple clusters are set. The initial cluster center of each cluster is represented by the weighted average of the first correlation coefficient, the second correlation coefficient and the third correlation coefficient. The distance between each element in the feature set and the initial cluster center is calculated. The cluster centers are iterated continuously, and the iterated clusters are output as feature clustering results.

3. The method for detecting and recognizing images of power materials as described in claim 1, characterized in that, Feature coverage can be expressed as: Among them, C f Ft represents the feature coverage on the f-th face. f Ft represents the proportion of pixels occupied by the feature vector on the f-th face. f,total This represents the sum of the pixel proportions of all feature vectors on the f-th face.

4. The method for detecting and recognizing images of power materials as described in claim 3, characterized in that, The matching result in step S34 is represented as follows: The matching difference value of the feature vector at the corresponding attitude angle is used to calculate the matching score of the current feature vector based on the obtained matching difference value. The matching score is then output as the matching result between the feature vector and the attitude angle. The feature coverage matching difference value is expressed as: ΔC f =|C f -C f,θ |; Where, ΔC f The matching difference value represents the feature coverage on the f-th face, θ represents the pose angle, and C f,θ This represents the feature coverage when an attitude angle θ exists on the f-th face; The matching difference value of the overlap area ratio is expressed as: I ff' =|IoU ff' -IoU ff',θ |; Where, ΔI ff' Represents the f-th face and the f-th face ' The matching difference value of the overlap area ratio of each face, IoU ff' Represents the f-th face and the f-th face ' The ratio of the overlapping areas of the faces, IoU ff',θ Represents the f-th face and the f-th face ' There is an overlap area ratio of attitude angle θ on each surface.

5. The method for detecting and recognizing images of power materials as described in claim 4, characterized in that, Match score is represented as: S f The matching score of the f-th face is represented by F, where F represents the number of faces present in the power material image, e represents the exponential constant, and α represents the adjustment coefficient; W f W represents the weight of the f-th face. f' Indicates the fth ' The weight of each face; The comprehensive matching result of power materials is expressed as follows: the comprehensive matching score is obtained by comprehensively calculating the matching results of feature vectors and attitude angles, and the comprehensive matching score is output as the comprehensive matching result. Among them, S total This indicates the overall matching score.

6. The method for detecting and recognizing images of power materials as described in claim 1, characterized in that, Step S35 also includes: S351, Based on the comprehensive matching results of power materials, match the comprehensive matching results with the power material tags to obtain at least one matching point; S352, perform matching point allocation processing, sequentially identify the first matching threshold corresponding to each matching point; and filter out multiple target points from the matching points according to the first matching threshold; S353 performs a difference analysis on the target point and uses the power material label corresponding to the point with the largest difference value after the difference analysis as the output comprehensive processing label.

7. The method for detecting and recognizing images of power materials as described in claim 1, characterized in that, Step S4 includes: S41, obtain the current allocation location of power materials and extract the tracking displacement value related to the allocation location; S42 combines the tracking displacement value with the allocation probability of power materials to verify whether the power materials are at the optimal location point; and calculates the displacement evaluation coefficient of the power materials at the optimal location point, and outputs the data corresponding to the displacement evaluation coefficient as the allocation feedback result.

8. The method for detecting and recognizing images of power materials as described in claim 7, characterized in that, The method to verify whether power materials are in the optimal position is to obtain the current displacement deviation of the power materials, combine the displacement deviation with the allocation probability, and calculate the displacement evaluation coefficient. Where δ represents the displacement evaluation coefficient, D represents the tracking displacement value, and ΔDI represents the displacement deviation. P(DI) represents the standard value of the displacement deviation, P(DI) represents the distribution probability of electrical materials, and α1, α2 and α3 represent the weighting coefficients, respectively.

9. A system for detecting and recognizing images of power materials, using the method for detecting and recognizing images of power materials as described in any one of claims 1-8, characterized in that, include: The image acquisition module is used to acquire images of power materials during their movement and sort the images according to the acquired process and location to form a set of material images related to the process location. The image detection module is used to perform image detection on the material image set and extract the feature set; Construct a linear regression model and perform linear regression analysis on the feature set; cluster the feature set based on the linear regression analysis results to obtain the feature clustering results. Based on the feature clustering results, set the power material label for the current power materials; The feature processing module is used to identify the relative position of the feature vectors corresponding to the feature clustering results on the power material image; Calculate feature coverage and overlap area ratio, and match them with power material tags; combine the attitude angle of the power material to generate a comprehensive matching result for the power material; based on the comprehensive matching result, obtain the comprehensive processing tag for the current power material under different priorities. The location verification module is used to obtain the current allocation location of power materials, extract the tracking displacement value, verify whether the power materials are in the optimal location point, and calculate the displacement evaluation coefficient. Output the distribution feedback results.

Images