AI-based high-altitude operation engineering entity quality defect identification method and system

Abstract

The invention provides an AI-based aerial work engineering entity quality defect identification method and system, and the method comprises the steps: constructing a quality defect sample library, and building a standard database; associating information in the standard database with the standard image data, extracting target features from the standard database, and merging the target features with the standard image data; using supervised learning to train an image recognition model according to the data label; data are collected from the standard database and the real-time image data and provided for the image recognition model for prediction and inference; and carrying out cross validation on an output result of the image recognition model and information in a standard database, wherein the output result contains corresponding database information. The system comprises a data processing module, a model construction module and a result output module. According to the method, the quality defect of the aerial work engineering entity can be efficiently identified, the working efficiency and quality are improved, and important decision support is provided for quality control.

Description

Technical Field

[0001] The present invention relates to the field of image recognition technology, and in particular to an AI-based method and system for identifying quality defects of aerial work engineering entities. Background Art

[0002] High-altitude line operations refer to high-altitude operations performed on transmission lines or other power facilities, usually involving inspections, maintenance, installation and replacement of high-altitude poles, insulators, conductors and other components. This type of operation requires operators to perform in a high-altitude environment, which requires special operating skills and safety measures. In order to ensure the safety of high-altitude line operations, it is necessary to monitor the quality of the operation, thereby reducing the rework rate and ensuring safety and quality. General high-altitude operation monitoring includes: safety monitoring, including ensuring that operators comply with safety operating procedures, wear safety protection equipment correctly, and avoid accidents caused by human factors. Equipment monitoring, including monitoring the compliance and safety of operating equipment and tools to ensure that the equipment is intact and operating normally. Operation process monitoring, monitoring the operation process of each link in the operation process to ensure that it is carried out in accordance with standard operating procedures to prevent operational errors or illegal operations. Quality control, monitoring the quality of the operation results, including inspection and acceptance of the quality of the completed operation project to ensure that the operation meets the requirements of the specifications. Environmental protection, monitoring the impact of the operation on the surrounding environment to prevent environmental pollution or safety hazards caused by the operation. Data recording and analysis, record and analyze the data of monitoring operations, establish archive records, and analyze key data in the operation process in order to evaluate the operation effect and improve the operation process in the future. By monitoring the quality of high-altitude operations on the lines, it can be ensured that the operation process is safe and reliable, the quality meets the standard requirements, and the safety of the operators and the integrity of the equipment are protected to the maximum extent, and the impact on the surrounding environment is reduced. It is an important link to ensure the normal operation of high-altitude operations on the lines and maintain the stable operation of power facilities. However, the existing monitoring equipment relies too much on cameras and other equipment, resulting in high monitoring costs, complex layout, and high professional requirements.

[0003] Prior art 1, application number: CN2202210336847.1 discloses a method and system for quality detection of power transmission projects based on image recognition, the method includes autonomous flight of a drone after take-off, obtaining point clouds by scanning power transmission line projects; comparing the first recognition model trained based on point cloud data, obtaining the construction progress corresponding to the current point cloud data, calling the corresponding preset route under the current construction progress, and obtaining the image of the target point; comparing the second recognition model of the key points in the image, obtaining the quality of the current power transmission project. Although drones are used as mobile platforms to automatically collect visible light images and point cloud data, efficient information acquisition at the construction site is achieved; the difference between the engineering entity and the recognition model is automatically judged by the front-end computing unit through the integration of AI technology, thereby realizing the intelligent management of the entire process of the power transmission project; in addition, based on the analysis and induction of existing data, the subsequent construction progress is intelligently predicted. However, the lack of the function of identifying quality defects leads to its single function and the inability to realize intelligent monitoring of construction quality, which increases the difficulty and cost of the operation to a certain extent.

[0004] Prior art 2, application number: CN202211092372.2 discloses an automatic measurement method for construction parameters of a power transmission line project based on laser point cloud, including the following steps: using a drone equipped with a laser radar to obtain point cloud data of a power transmission line tower; selecting the point cloud data to evaluate the inclination angle of the power transmission line tower; planning the route measurement area and route according to the path of the power transmission line, using a drone to obtain a holographic image in the corridor of the power transmission line, and calculating the tower height and the height space of the hanging point; by obtaining the parameters of the power transmission line tower, calibrating the coordinates of the power transmission line tower; by obtaining the real-time video of the drone, identifying the defects of the power transmission line tower. Although the point cloud data of the power transmission line tower is obtained by a drone equipped with a laser radar, it is convenient to calculate the inclination angle of the tower, the tower height and the height space of the hanging point, the tower coordinates and other key parameters, so as to facilitate the judgment of whether it meets the design and specification acceptance requirements. However, only the detection of quality defects of overhead lines cannot monitor the structures such as towers, vertical arcs and bolts of the line high-altitude operations, resulting in inaccurate results.

[0005] Prior art three, application number: CN 202210987319.2 discloses a transmission line engineering defect detection method based on image semantic segmentation, and the transmission line engineering defect detection method based on image semantic segmentation includes obtaining a transmission line image, performing semantic segmentation processing on the transmission line image, obtaining image shooting parameters, determining a target defect image from a preset transmission line defect image set according to the image shooting parameters, performing model training processing on the target defect image, generating a defect recognition model, and performing defect detection on the transmission line image according to the defect recognition model to obtain a transmission line detection result, wherein, although the image shooting parameters are obtained by semantic segmentation processing, the target defect image is screened out by the image shooting parameters, while ensuring the accuracy of defect detection, the amount of data used to generate the defect recognition model is reduced, the training speed of the defect detection model is improved, and defect detection is performed more efficiently to ensure the safety of the transmission line project. However, the operation is complicated and has extremely high professional requirements, which is not conducive to improving the efficiency of high-altitude operations. At present, the existing technologies one, two and three have the problems of low intelligence in quality monitoring of high-altitude operations of lines, low monitoring efficiency, and failure to ensure that the quality of operations meets the standard requirements. Therefore, the present invention provides an AI-based method and system for identifying quality defects of aerial work engineering entities. By monitoring the quality of aerial line operations, it can ensure that the operation process is safe and reliable, and the quality meets the standard requirements, maximize the protection of the safety of operators and the integrity of equipment, and reduce the impact on the surrounding environment; it is an important link to ensure the normal progress of aerial line operations and maintain the stable operation of power facilities. Summary of the invention

[0006] In order to solve the above technical problems, the present invention provides an AI-based method for identifying quality defects of aerial work engineering entities, comprising the following steps:

[0007] Collect sample data, build a sample library of bolt missing and wire damage quality defects, and establish a standard database based on the modular design of the line three-dimensional set; associate the information in the standard database with the standard image data, and establish a connection between the standard image data and the standard database;

[0008] Extract target features from the standard database and merge the target features with the standard image data; use supervised learning to train the image recognition model based on the data labels, and identify and infer relevant information from the image data and the standard database information;

[0009] Data is collected from standard databases and real-time image data and provided to the image recognition model for prediction and inference; the output of the image recognition model is cross-validated with the information in the standard database, and the corresponding database information is included in the output.

[0010] Optionally, the process of establishing a connection between the standard image data and the standard database comprises the following steps:

[0011] Determine the type of data that needs to be collected, obtain photographic equipment to inspect and photograph the engineering entity, and obtain sample data containing quality defects;

[0012] Label the collected data, mark the location of missing bolts and the area of ​​wire damage in each sample; convert and normalize the data, create a database table, and import the processed data into it;

[0013] Design and establish indexes within the standard database, and integrate standard image data and database information into corresponding fields in the database to form a complete standardized data set.

[0014] Optional, data type, including the engineering entity parts and risk factors for quality defect detection, risk factors including bolt missing and wire damage.

[0015] Optionally, the process of forming a complete standardized dataset includes the following steps:

[0016] Design the database table structure, with a table containing quality defect information and a table containing image data and paths; use the database management tool to execute the SQL statement to create the table;

[0017] Import the marked, format-converted and normalized data into the corresponding database table, and map the data to the correct fields; select the fields used for query as index fields; use the database's create index statement to create indexes on the required fields of the database table;

[0018] Integrate standard image data and database information into corresponding fields in the database, and use primary key and foreign key related fields to establish the connection between image data and quality defect information.

[0019] Optional, database table structure, including field names, data types, and constraints.

[0020] Optionally, the process of extracting features from the combined training dataset includes the following steps:

[0021] Preprocess the image data in the merged training dataset, including resizing the image, normalizing pixel values, and removing noise. Use the pre-trained convolutional neural network to extract image features by extracting the feature vector of the convolutional layer.

[0022] For the features extracted from each image, different types of features are concatenated together to form a feature vector, and the feature vectors extracted from different images are filled to convert them to a fixed length;

[0023] Traverse the entire feature vector, find the maximum and minimum values ​​of each feature dimension, and map each feature value to the preset range.

[0024] Optionally, the process of evaluating the performance of the image recognition model and whether it can generalize to new data includes the following steps:

[0025] Use the trained image recognition model to predict the image data in the test set;

[0026] Compare the predictions of the image recognition model with the actual labels and calculate the accuracy, precision, recall and F1 score;

[0027] Analyze the values ​​of various indicators to determine the performance and generalization ability of the image recognition model.

[0028] Optionally, the cross-validation process includes the following steps:

[0029] Use the trained image recognition model to predict and infer the real-time image data and obtain the model output results, that is, the identified quality defect information;

[0030] Obtain standard information corresponding to the real-time image from the standard database, including correct quality defect labels or descriptions;

[0031] Compare the output of the image recognition model with the information in the standard database for cross-validation.

[0032] Optionally, the quality defect information output by the image recognition model is compared with the actual situation in the standard database to ensure whether the model correctly identifies the quality defects;

[0033] Count the matches between the output results of the image recognition model and the standard database information, and calculate the matching rate or accuracy;

[0034] Analyze situations where image recognition model outputs do not match standard databases to understand the types and causes of errors.

[0035] The present invention provides an AI-based high-altitude work engineering entity quality defect identification system, comprising:

[0036] The data processing module is responsible for collecting sample data, building a sample library of bolt missing and wire damage quality defects, and establishing a standard database based on the modular design of the line three-dimensional set; associating the information in the standard database with the standard image data, and establishing a connection between the standard image data and the standard database;

[0037] Among them, standard image data is photos or video data of line facilities that meet line quality requirements obtained by cameras or drone equipment, showing specific details of power facilities, including the appearance and status of poles, insulators and conductor components;

[0038] The model building module is responsible for extracting target features from the standard database and merging the target features with the standard image data; using supervised learning based on the data labels, the merged data is used to train the image recognition model to identify and infer relevant information from the image data and the standard database information;

[0039] The result output module is responsible for collecting data from the standard database and real-time image data, and providing it to the image recognition model for prediction and inference; cross-validating the output results of the image recognition model with the information in the standard database, and including the corresponding database information in the output results.

[0040] The present invention establishes a standard library of quality defects of power facilities by collecting quality defect sample data of line entities and associating it with a standard database of modular design of line three-dimensional sets; the establishment of a standard database provides a reliable data source for training and verifying quality defect recognition models; the construction of the sample library enables the model to learn various forms of quality defects and improve the generalization ability and accuracy of the model. Target features are extracted from the standard database, the features are merged with standard image data, and the image recognition model is trained using a supervised learning method; by training the model, it is enabled to learn quality defect features from images, and accurately identify and classify quality defects, thereby realizing automatic identification and judgment of quality defects of aerial work engineering entities. Data is collected from the standard database and real-time image data, and provided to the image recognition model for prediction and inference; the output of the model is compared with the information in the standard database; by cross-validating the output results of the model, the recognition results are ensured to be accurate and reliable, and the effectiveness and credibility of the model in the actual engineering environment are verified, providing a basis for further quality control and improvement.

[0041] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0042] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] 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:

[0044] Figure 1 This is a flow chart of the method for identifying quality defects of aerial work engineering entities based on AI in Example 1 of the present invention;

[0045] Figure 2 A process diagram for establishing a connection between standard image data and a standard database in Embodiment 2 of the present invention;

[0046] Figure 3 A process diagram for forming a complete standardized data set in Example 3 of the present invention;

[0047] Figure 4 This is a process diagram of using the merged data for training an image recognition model in Embodiment 4 of the present invention;

[0048] Figure 5 This is a process diagram of extracting features from a combined training data set in Embodiment 5 of the present invention;

[0049] Figure 6 A process diagram of extracting feature vectors from different images by filling in Embodiment 6 of the present invention;

[0050] Figure 7 A diagram showing a process of using a training data set to train an image recognition model in Example 7 of the present invention;

[0051] Figure 8 A process diagram for evaluating the performance of an image recognition model and whether it can be generalized to new data in Example 8 of the present invention;

[0052] Fig. 9 This is a diagram of the process of cross-validation in Example 9 of the present invention;

[0053] Fig.10 This is a block diagram of the AI-based high-altitude work engineering entity quality defect identification system in Example 10 of the present invention. DETAILED DESCRIPTION

[0054] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0055] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the embodiments of the present application. The singular forms of "a", "said" and "the" used in the embodiments of the present application are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in this article refers to and includes any or all possible combinations of one or more associated listed items.

[0056] When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are only examples of devices and methods consistent with some aspects of the present application described in detail. In the description of the present application, it should be understood that the terms "first", "second", "third", etc. are only used to distinguish similar objects, and do not have to be used to describe a specific order or sequence, nor can they be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to the specific circumstances.

[0057] Example 1: Figure 1 As shown, an embodiment of the present invention provides an AI-based method for identifying quality defects of aerial work engineering entities, comprising the following steps:

[0058] S100: Collect sample data, build a sample library of quality defects such as missing bolts and damaged conductors, and establish a standard database based on the modular design of the line three-dimensional set; associate the information in the standard database with the standard image data, and establish a connection between the standard image data and the standard database;

[0059] Among them, standard image data is photos or video data of line facilities that meet line quality requirements obtained through equipment such as cameras or drones, showing specific details of power facilities, including the appearance and status of components such as towers, insulators and conductors;

[0060] S200: extracting target features from a standard database and merging the target features with standard image data; using supervised learning to train an image recognition model based on data labels, and identifying and inferring relevant information from the image data and standard database information;

[0061] S300: Collect data from the standard database and real-time image data, and provide it to the image recognition model for prediction and inference; cross-validate the output result of the image recognition model with the information in the standard database, and include the corresponding database information in the output result.

[0062] The working principle and beneficial effects of the above technical solution are as follows: In this embodiment, sample data is first collected to construct a sample library of quality defects such as missing bolts and damaged conductors, and a standard database is established according to the modular design of the line three-dimensional set; the information in the standard database is associated with the standard image data to establish a connection between the standard image data and the standard database; wherein the standard image data is a photograph or video data of line facilities that meet the line quality requirements obtained through equipment such as cameras or drones, showing the specific details of the power facilities, including the appearance and status of components such as poles, insulators and conductors; secondly, target features are extracted from the standard database and merged with the standard image data; supervised learning is used to train the image recognition model based on the data labels after the merged data is used to identify and infer relevant information from the image data and the standard database information; finally, data is obtained from the standard database and real-time image data, and provided to the image recognition model for prediction and inference; the output result of the image recognition model is cross-validated with the information in the standard database, and the corresponding database information is included in the output result. Step S100 of the above scheme establishes a standard library of quality defects of power facilities by collecting quality defect sample data of line entities and associating it with the standard database of modular design of line three-dimensional sets; the establishment of the standard database provides a reliable data source for training and verifying the quality defect recognition model; the construction of the sample library enables the model to learn various forms of quality defects and improve the generalization ability and accuracy of the model. Step S200 extracts target features from the standard database, merges the features with standard image data, and trains the image recognition model using supervised learning methods; by training the model, it can learn quality defect features from images, accurately identify and classify quality defects, and realize automatic identification and judgment of quality defects of aerial work engineering entities. Step S300 collects data from the standard database and real-time image data, and provides it to the image recognition model for prediction and inference; compares the output of the model with the information in the standard database; by cross-validating the output results of the model, ensures that the recognition results are accurate and reliable, verifies the effectiveness and credibility of the model in the actual engineering environment, and provides a basis for further quality control and improvement.

[0063] In summary, this embodiment can establish a complete and reliable quality defect identification system through the above steps. By efficiently collecting, labeling and processing sample data, the trained model can efficiently identify the quality defects of aerial work engineering entities, improve work efficiency and quality, and provide important decision support for quality control. At the same time, through the association of standard databases and image data, the recognition algorithm can be made more accurate and universal, and effective monitoring and management of line facility quality can be achieved.

[0064] Example 2: Figure 2As shown, based on Example 1, the process of establishing the connection between the standard image data and the standard database provided by the embodiment of the present invention includes the following steps:

[0065] S101: Determine the data types to be collected, including the engineering entity parts and risk factors for quality defect detection, and the risk factors include missing bolts and wire damage, etc.; obtain a shooting device to inspect and shoot the engineering entity to obtain sample data containing quality defects;

[0066] S102: labeling the collected data, marking the location of the missing bolts and the area of ​​the damaged wire in each sample; performing format conversion and normalization processing on the data, creating a database table, and importing the processed data into the database;

[0067] S103: Design and establish an index within the standard database, and integrate the standard image data and database information into corresponding fields in the database to form a complete standardized data set.

[0068] The working principle and beneficial effects of the above technical solution are as follows: This embodiment first determines the type of data to be collected, including the engineering entity parts and risk factors for quality defect detection, and the risk factors include missing bolts and damaged wires, etc.; obtains a shooting device to inspect and shoot the engineering entity, and obtains sample data containing quality defects; secondly, the collected data is annotated, marking the location of the missing bolts and the area of ​​wire damage in each sample; the data is formatted and normalized, a database table is created, and the processed data is imported into it; finally, an index in the standard database is designed and established, and the standard image data and database information are integrated into the corresponding fields in the database to form a complete standardized data set. Step S101 of the above scheme determines the type of quality defect detection data to be collected, such as missing bolts, damaged wires, etc., and obtains a shooting device to inspect and shoot the engineering entity, and collects sample data containing quality defects; through this step, a sample library containing various quality defects is established, laying the foundation for subsequent model training and database construction. Step S102 labels the collected data, marks the bolt missing position and wire damage area in each sample; converts and normalizes the data, creates a database table, and imports the processed data into it; data labeling helps the model learn quality defect characteristics, format conversion and normalization help improve data consistency and trainability, and the creation of database tables facilitates data integration and management. Step S103 designs and establishes indexes in the standard database, integrates standard image data and database information into corresponding fields in the database, and forms a complete standardized data set; by establishing database indexes and integrating data, a connection between image data and database information is established, providing support for model training and data query, and improving data retrieval speed and accuracy.

[0069] In summary, through these steps, this embodiment can effectively link standard image data and standard database information, providing strong data support for the identification and management of quality defects. Integrating and standardizing data makes model training more accurate and efficient, and provides a reliable data basis for subsequent quality defect detection and decision-making.

[0070] Example 3: Figure 3 As shown, based on Example 2, the process of forming a complete standardized data set provided in this embodiment of the present invention includes the following steps:

[0071] S1031: Design the database table structure, including field names, data types and constraints, with a table containing quality defect information and a table containing image data and paths; use the database management tool to execute the SQL statement to create the table;

[0072] S1032: Import the marked, format-converted and normalized data into the corresponding database table, and correspond the data to the correct field; select the field used for query as the index field; use the database's create index statement to create an index on the required field of the database table;

[0073] S1033: Integrate the standard image data and database information into corresponding fields in the database, and use related fields such as primary keys and foreign keys to establish a connection between the image data and the quality defect information.

[0074] The working principle and beneficial effects of the above technical solution are as follows: In this embodiment, the database table structure is first designed, including field names, data types and constraints, etc., and there is a table containing quality defect information and a table containing image data and paths; the SQL statement for creating a table is executed using a database management tool; secondly, the data after being marked, format converted and normalized is imported into the corresponding database table, and the data corresponds to the correct field; the field used for query is selected as the index field; the index creation statement of the database is used to create an index on the required field of the database table; finally, the standard image data and database information are integrated into the corresponding fields in the database, and the connection between the image data and the quality defect information is established using related fields such as primary keys and foreign keys. Step S1031 of the above scheme designs the database table structure, including field names, data types and constraints, and creates a table containing quality defect information and a table containing image data and paths respectively; the SQL statement for creating a table is executed using a database management tool to ensure that the table structure meets the design requirements; determining the appropriate table structure can standardize data storage and management, and ensure the structuring and consistency of data; creating a table and executing SQL statements enable the database to have the basic functions of storing and querying data, which facilitates subsequent data operations. Step S1032 imports the data that has been marked, format converted and normalized into the corresponding database table to ensure that the data corresponds to the correct field; selects the field used for query as the index field, and uses the database's create index statement to create an index on the required field; the import of data ensures the accuracy and completeness of the data, which is convenient for subsequent data query and analysis; creating an index can improve the efficiency of data query, speed up the data retrieval speed and performance, and make the database more efficient. Step S1033 integrates the standard image data and database information into the corresponding fields of the database to ensure the accuracy and completeness of the data; uses related fields such as primary keys and foreign keys to establish the connection between image data and quality defect information to build the connection and association between data; data integration ensures the consistency of image data and database information, and provides a basis for data query and association analysis; establishing the connection between image data and quality defect information can realize association query and multidimensional analysis between data, and help to conduct comprehensive data mining and analysis.

[0075] In summary, this embodiment establishes a complete standardized data set through the above steps, so that a close connection is established between the image data and the database information, providing reliable data support for subsequent quality defect identification and analysis, which is beneficial to data management, query and analysis, and plays an important role in quality management.

[0076] Example 4: Figure 4 As shown, based on Example 1, the process of using the merged data for training the image recognition model provided by the embodiment of the present invention includes the following steps:

[0077] S201: extracting target feature data from a standard database, merging the target feature data with standard image data, and constructing a training data set, wherein the training data set includes labeled samples and corresponding features / labels; performing feature extraction on the merged training data set, and converting the image data into a feature vector recognizable by a computer;

[0078] S202: Select a supervised learning algorithm for a convolutional neural network, extract a test set from the merged data set, train an image recognition model using the training data set, and adjust parameters to minimize prediction errors by learning the relationship between features and labels;

[0079] S203: Use the test set to evaluate the trained image recognition model. By calculating indicators such as accuracy, precision, recall rate and F1 score, the performance of the image recognition model and whether it can be generalized to new data are evaluated. After the evaluation is completed, the trained image recognition model is used in actual application scenarios to identify quality defects of aerial work engineering entities.

[0080] The working principle and beneficial effects of the above technical solution are as follows: In this embodiment, the target feature data is first extracted from the standard database, the target feature data is merged with the standard image data, and a training data set is constructed, wherein the training data set contains labeled samples and corresponding features / labels; feature extraction is performed on the merged training data set, and the image data is converted into a computer-recognizable feature vector; secondly, a supervised learning algorithm of a convolutional neural network is selected, a test set is extracted from the merged data set, and an image recognition model is trained using the training data set, and the parameters are adjusted to minimize prediction errors by learning the relationship between features and labels; finally, the trained image recognition model is evaluated using the test set, and the performance of the image recognition model and whether it can be generalized to new data are evaluated by calculating indicators such as accuracy, precision, recall rate and F1 score; after the evaluation is completed, the trained image recognition model is used in actual application scenarios to identify quality defects of aerial work engineering entities. Step S201 of the above scheme extracts target feature data from the standard database and merges it with the standard image data to construct a training data set with labels / features; performs feature extraction to convert image data into computer-recognizable feature vectors to provide meaningful input data for the model; through data merging and feature extraction, the image data is effectively converted into a form that can be processed by the model, providing a data basis for training the image recognition model and laying the foundation for subsequent supervised learning modeling. Step S202 selects a supervised learning algorithm such as a convolutional neural network (CNN) and uses the training data set to train the image recognition model; by learning the relationship between features and labels, the model parameters are optimized to reduce prediction errors; the training model can effectively learn the characteristics of the data and establish an accurate mapping relationship, thereby improving the accuracy of the model in identifying quality defects; through appropriate model selection and training, the purpose of accurately identifying and classifying quality defects is achieved. Step S203 uses the test set to evaluate the performance of the trained model, calculates indicators such as accuracy, precision, recall, F1 score, and determines the generalization ability of the model; applies the trained image recognition model to actual scenarios to identify quality defects of aerial work engineering entities; model evaluation can help understand the performance of the model, provide feedback and guidance for adjusting the model, and ensure the accuracy and generalization ability of the model; applies the trained image recognition model to realize the identification of quality defects of aerial work engineering entities, thereby improving work efficiency and quality management level.

[0081] In summary, by executing the above steps, this embodiment can train an effective image recognition model to identify quality defects of aerial work engineering entities and improve work efficiency and correct judgment rate; these steps are crucial to ensuring the accuracy and generalization ability of the model.

[0082] Example 5: Figure 5As shown, based on Example 4, the process of extracting features from the merged training data set provided in the embodiment of the present invention includes the following steps:

[0083] S2011: Preprocessing the image data in the merged training data set, including operations such as adjusting the image size, normalizing pixel values, and removing noise, using a pre-trained convolutional neural network to extract image features, by extracting the feature vector of the convolutional layer;

[0084] Among them, the loss function updates the parameters of the denoising model to remove noise, including the following steps:

[0085] Create a noisy image dataset with resized and normalized pixel values ​​and a corresponding noise-free image dataset;

[0086] Input the noisy image to the denoising model f x (m,n), to generate the predicted denoised image;

[0087] Calculate the difference between the original image and the generated image using the selected loss function L1;

[0088] Update the denoising model parameters through the back propagation algorithm according to the loss function L1;

[0089] Iterative training: Repeat the above steps until the denoising model converges;

[0090] The loss function expression is:

[0091]

[0092] The expression of the denoising model is:

[0093]

[0094] In the formula, f x (m,n) represents the denoised image result at position (m,n), f s (m,n) represents the clean image at position (m,n), f v (m,n) represents the noise variance at position (m,n), which is used to describe the degree of image noise; n(m,n) represents the standard normal noise added at position (m,n); ε1 and ε2 represent noise parameters, and y represents the pixel value of the target image. represents the pixel value of the image predicted by the denoising model (the generated denoised image); α and γ represent the hyperparameters that control the loss component in the loss function, and p represents the learning probability value of the target image feature;

[0095] S2012: for the features extracted from each image, different types of features are concatenated together to form a feature vector, and the feature vectors extracted from different images are filled to convert them into a fixed length;

[0096] S2013: traverse the entire feature vector, find the maximum and minimum values ​​of each feature dimension, and map each feature value to a preset range.

[0097] The working principle and beneficial effects of the above technical solution are as follows: the present embodiment first pre-processes the image data in the merged training data set, including operations such as adjusting the image size, normalizing pixel values, and removing noise, and uses a pre-trained convolutional neural network to extract image features, by extracting the feature vector of the convolution layer; secondly, for the features extracted from each image, different types of features are spliced ​​together to form a feature vector, and the feature vectors extracted from different images are filled and converted to a fixed length; finally, the entire feature vector is traversed to find the maximum and minimum values ​​of each feature dimension, and each feature value is mapped to a preset range. Step S2011 of the above solution adjusts the image size, normalizes the pixel value, and removes noise, so that the image can better adapt to entering the neural network model for feature extraction; uses a pre-trained convolutional neural network to extract image features, and obtains a high-level abstract feature representation of the image data in the convolutional neural network by extracting the feature vector of the convolution layer; pre-processing can improve the quality and availability of image data, and help the model better learn effective features; using the pre-trained convolutional neural network, more representative and efficient features can be extracted from the image, laying the foundation for subsequent feature vector extraction and model learning. Step S2012 combines features from different feature extraction methods to enrich the expressive power of feature vectors; by filling or truncating, the length of feature vectors extracted from different images is made consistent, which is convenient for the next step; different types of features are merged into a complete feature vector, which improves the richness and comprehensiveness of feature representation; unifying the length of feature vectors helps to ensure that the features extracted from each image have the same dimension, making the processing of feature vectors more standardized and unified. Step S2013 analyzes the value range of each dimension in the feature vector to find the maximum and minimum values; normalizes each feature value according to a preset range to ensure that the feature value is within a unified range; the comparability and stability of feature vectors are improved through normalization, making the model easier to learn and generalize; mapping feature values ​​to a preset range helps to accelerate the training process, avoid problems such as gradient explosion, and improve the performance and accuracy of the model.

[0098] In summary, this embodiment shows the complete process of image data feature extraction from aspects such as preprocessing, feature fusion and normalization, so that the extracted feature vector is more expressive and applicable, thereby improving the training effect and inference accuracy of the model. This embodiment removes noise and improves image quality, which is to remove noise from the input image and make the output image closer to the original (clean) image. Using machine learning or deep learning models, the mapping relationship between noise and clean images can be more accurately fitted; the model automatically learns the most important features in the denoising process without manually designing complex feature extraction steps, and the pre-trained convolutional neural network can capture features at different levels, thereby improving the performance of the denoising model. Through the loss function, the model back propagation stage can dynamically adjust parameters, thereby effectively reducing the prediction error of the model, ensuring that the model gradually approaches the characteristics of the real data, thereby improving the quality of the final output image; the loss function contains the variance and parameters of the noise, which can better model the noise characteristics in the model, and by providing quantitative information about the noise, the denoising process can adaptively adjust the strategy according to different noise levels; through multiple iterations, each time the model parameters are updated, the loss function value is continuously reduced during the training process, ensuring that the model converges towards better performance. This process can effectively improve the generalization ability of the model on new images. The mentioned parameters control different components of the loss, which can help the model handle different types of noise and information more carefully during feature learning, so that the model can retain the texture and structural information of the image while denoising; through the design of the loss function and the description of the noise model, the model can adaptively handle noise of different types and intensities. In this way, the model can not only handle known noise types, but also handle unknown or new noise forms to a certain extent.

[0099] Example 6: Figure 6 As shown, based on Example 5, the process of extracting feature vectors from different images by filling provided in the embodiment of the present invention includes the following steps:

[0100] S20121: For each image, the features extracted have different lengths or dimensions. During the concatenation process, different types of features are sequentially connected together to form a feature vector;

[0101] S20122: Determine the required fixed length. For feature vectors that are not long enough, use a padding method to add a zero vector or a repeated vector value at the end of the vector so that the feature vector reaches the fixed length.

[0102] S20123: For feature vectors whose length exceeds a fixed length, a truncation method is used to retain some key feature values.

[0103] The working principle and beneficial effects of the above technical solution are as follows: first, for each image, the features extracted have different lengths or dimensions. During the splicing process, different types of features are sequentially connected together to form a feature vector; secondly, the required fixed length is determined. For feature vectors that are insufficient in length, a padding method is used to add a zero vector or a repeated vector value at the end of the vector so that the feature vector reaches a fixed length; finally, for feature vectors that exceed the fixed length, a truncation method is used to retain some key eigenvalues. Step S20121 of the above solution sequentially connects different types of features together to form a feature vector, so that different types of features can be combined together to provide richer information representation; by splicing features from different feature extractors, a more comprehensive and integrated feature representation can be obtained, thereby improving the model's ability to express data and classification performance. Step S20122: The required fixed length is determined. For feature vectors with insufficient length, zero vectors or repeated vector values ​​are added to the end of the vector by padding to expand the feature vector to a fixed length; this ensures that the feature vector has a fixed dimension and length during concatenation and subsequent processing, which is convenient for model input and calculation, reduces uncertainty in data processing, and improves the operability of the feature vector. Step S20123: For feature vectors with a length exceeding the fixed length, a truncation method is used to retain some key eigenvalues; this avoids the computational burden and model overfitting problems caused by the high dimension of the feature vector, ensures that the feature vector retains key information within a fixed length, and improves the efficiency and generalization ability of the model.

[0104] In summary, this embodiment combines the methods of splicing, padding and truncation to process the feature vectors extracted from different images so that they have a fixed length and a consistent feature representation. Through this processing, the interpretability of the feature vector can be increased, the complexity of data processing can be reduced, and the model's ability to extract and analyze data features can be improved, thereby improving the performance and generalization ability of the model.

[0105] Example 7: Figure 7 As shown, based on Example 4, the process of using the training data set to train the image recognition model provided by the embodiment of the present invention includes the following steps:

[0106] S2021: Build a convolutional neural network model, including convolutional layers, pooling layers, and fully connected layers, define loss functions and optimizers; use training data sets to train convolutional neural network models; continuously adjust network parameters through back-propagation algorithms to minimize loss functions; supervised learning learns model parameters through the relationship between features and labels in training data sets;

[0107] S2022: The convolutional neural network adjusts the parameters in the convolutional neural network according to the calculation results of the loss function through the back propagation algorithm; for each training sample, the convolutional neural network model calculates the predicted output of the sample through the forward propagation algorithm; that is, the feature vector is input into the model, and the final category prediction result is obtained through the convolution layer, pooling layer and fully connected layer;

[0108] S2023: Compare the predicted output of the convolutional neural network model for the training samples with the true labels and calculate the value of the loss function; the loss function measures the degree of error between the model prediction results and the true labels; according to the value of the loss function, update all parameters in the convolutional neural network model through the back propagation algorithm so that the loss function value gradually decreases; repeat the steps and continuously optimize the model parameters by inputting all samples in the training data set into the network. The iterative process allows the model to gradually learn the relationship between features and labels, and adjust the network parameters to minimize prediction errors.

[0109] The working principle and beneficial effects of the above technical solution are as follows: In this embodiment, a convolutional neural network model is first built, including a convolutional layer, a pooling layer, and a fully connected layer, etc., and a loss function and an optimizer are defined; the convolutional neural network model is trained using a training data set; the network parameters are continuously adjusted through a back-propagation algorithm to minimize the loss function; supervised learning learns model parameters through the relationship between features and labels in the training data set; secondly, the convolutional neural network adjusts the parameters in the convolutional neural network according to the calculation results of the loss function through a back-propagation algorithm; for each training sample, the predicted output of the convolutional neural network model for the sample is calculated through a forward propagation algorithm; that is, the features are forward propagated to the convolutional neural network. The amount is input into the model, and after passing through the convolution layer, pooling layer and fully connected layer, the final category prediction result is obtained; finally, the prediction output of the convolutional neural network model for the training sample is compared with the true label, and the value of the loss function is calculated; the loss function measures the degree of error between the model prediction result and the true label; according to the value of the loss function, all parameters in the convolutional neural network model are updated through the back propagation algorithm, so that the loss function value gradually decreases; repeat the steps, by inputting all samples in the training data set into the network, the model parameters are continuously optimized, and the iterative process allows the model to gradually learn the relationship between features and labels, and adjust the network parameters to minimize prediction errors. Step S2021 of the above scheme: defines the structure of the model, including convolutional layer, pooling layer and fully connected layer, etc., determines the role and connection mode of each layer in the neural network; establishes a complex, multi-level network structure, so that the model can better learn the features and patterns in the data, and improves the expressiveness of the model; the loss function is used to measure the error between the model prediction result and the true label, and helps the network to optimize; the optimizer is used to adjust the network parameters according to the gradient of the loss function to reduce the loss value; by defining a suitable loss function and optimizer, guide the model to optimize in the right direction, accelerate the convergence speed of the model, and improve the training effect. Step S2022 uses the back propagation algorithm to adjust the parameters in the convolutional neural network according to the calculation results of the loss function, so that the model can better fit the training data; optimize the parameters of the model so that the network can better learn the features and patterns of the data and improve the generalization ability of the model; input the feature vector into the model, and obtain the final category prediction result after passing through the convolutional layer, pooling layer and fully connected layer; calculate the prediction output of the model for the sample through forward propagation, and judge the model's processing ability and classification accuracy for the input data.Step S2023 calculates the value of the loss function, and uses the back propagation algorithm to update all parameters in the model according to the loss value, so that the loss function gradually decreases; improves the prediction accuracy of the model, optimizes the network parameters, and allows the model to better adapt to the training data set and be able to generalize to new data; by continuously inputting training samples and optimizing the model parameters, gradually learns the relationship between the characteristics of the data and the label; continuous iterative training allows the model to continuously improve, improve the performance and generalization ability of the model, and enable the model to more accurately predict the labels of unknown data.

[0110] In summary, the steps of this embodiment combine important training processes such as model building, forward propagation, back propagation, loss calculation and parameter updating, so that the convolutional neural network can learn effective feature representations from the training data, thereby improving the performance and generalization ability of the model in image recognition tasks.

[0111] Example 8: Figure 8 As shown, based on Example 4, the process of evaluating the performance of the image recognition model and whether it can be generalized to new data provided by the embodiment of the present invention includes the following steps:

[0112] S2031: Use the trained image recognition model to predict the image data in the test set;

[0113] S2032: Compare the prediction results of the image recognition model with the actual labels and calculate the accuracy, precision, recall and F1 score;

[0114] S2033: Analyze the values ​​of various indicators to determine the performance and generalization ability of the image recognition model.

[0115] The working principle and beneficial effects of the above technical solution are as follows: This embodiment first uses the trained image recognition model to predict the image data in the test set; secondly, the prediction results of the image recognition model are compared with the actual labels, and the accuracy, precision, recall and F1 scores are calculated; finally, the values ​​of various indicators are analyzed to determine the performance and generalization ability of the image recognition model. Step S2031 of the above solution: Apply the trained image recognition model to the unknown test data set to predict the category label of each image; evaluate the generalization ability of the model for new data, and verify whether the model can accurately classify and recognize images in actual scenes. Step S2032 compares the prediction results of the model with the actual labels, calculates the accuracy, precision, recall and F1 scores, and thus evaluates the performance of the model on the test set; the accuracy measures the overall classification accuracy of the model, the precision and recall evaluate the accuracy of the model in the prediction of the positive category and the ability of the model to detect the positive category, and the F1 score comprehensively considers the precision and recall, and provides an evaluation indicator for the comprehensive performance of the model. Step S2033 comprehensively considers the values ​​of indicators such as accuracy, precision, recall rate and F1 score to judge the performance and generalization ability of the image recognition model on the test set; based on the numerical results of each indicator, the performance of the model can be determined, and the applicability and generalization ability of the model can be further evaluated, providing reference and guidance for further optimization of the model.

[0116] In summary, through the above process, this embodiment can comprehensively evaluate the performance of the image recognition model and determine its generalization ability to new data. These evaluation indicators help determine the actual performance of the model and guide subsequent improvement and optimization measures.

[0117] Example 9: Fig. 9 As shown, based on Example 1, the cross-validation process provided in the embodiment of the present invention includes the following steps:

[0118] S301: Use the trained image recognition model to predict and infer the real-time image data to obtain the model output result, that is, the identified quality defect information;

[0119] S302: Obtain standard information corresponding to the real-time image from a standard database, including a correct quality defect label or description;

[0120] S303: Compare the output result of the image recognition model with the information in the standard database to perform cross-validation;

[0121] Compare the quality defect information output by the image recognition model with the actual situation in the standard database to ensure that the model correctly identifies the quality defects;

[0122] Count the matches between the output results of the image recognition model and the standard database information, and calculate the matching rate or accuracy;

[0123] Analyze situations where image recognition model outputs do not match standard databases to understand the types and causes of errors.

[0124] The working principle and beneficial effects of the above technical solution are as follows: In this embodiment, the trained image recognition model is first used to predict and infer the real-time image data, and the model output result, i.e., the identified quality defect information, is obtained; secondly, the standard information corresponding to the real-time image is obtained from the standard database, including the correct quality defect label or description; finally, the image recognition model output result is compared with the information in the standard database for cross-validation; the quality defect information output by the image recognition model is compared with the actual situation in the standard database to ensure whether the model correctly identifies the quality defect; the image recognition model output result matches the standard database information, and the matching rate or accuracy rate is calculated; the image recognition model output does not match the standard database to understand the error type and cause. Step S301 of the above solution predicts the real-time image data through the trained model and obtains the model output result, i.e., the identified quality defect information; obtains the model's recognition result of the quality defect, provides basic data for subsequent cross-validation, and evaluates the performance of the model in actual situations. Step S302 obtains the standard information corresponding to the real-time image from the standard database, including the correct quality defect label or description; as a reference standard for the model prediction result, it is used to compare with the model output to ensure that the model's quality defect recognition is consistent with the actual situation. Step S303 compares the output results of the model with the information in the standard database for cross-validation; ensure recognition accuracy: by comparing the model output and standard information, confirm whether the model can correctly identify quality defects and improve the recognition accuracy of the model; evaluate model performance: statistical matching rate or accuracy rate can evaluate the accuracy and reliability of the model; analyze errors: by analyzing mismatches, we can understand the problems and improvement directions of the model and improve the generalization ability and performance of the model.

[0125] In summary, through the above process, this embodiment ensures that the output of the model meets the actual standard by comparing the actual situation and the model prediction results in cross-validation, improves the recognition accuracy of the model and guides the optimization and improvement of the model.

[0126] Example 10: Fig.10 As shown, based on Examples 1 to 9, the AI-based aerial work engineering entity quality defect identification system provided by the embodiment of the present invention includes:

[0127] The data processing module is responsible for collecting sample data, building a sample library of quality defects such as missing bolts and damaged conductors, and establishing a standard database based on the modular design of the line three-dimensional set; associating the information in the standard database with the standard image data, and establishing a connection between the standard image data and the standard database;

[0128] Among them, standard image data is photos or video data of line facilities that meet line quality requirements obtained through equipment such as cameras or drones, showing specific details of power facilities, including the appearance and status of components such as towers, insulators and conductors;

[0129] The model building module is responsible for extracting target features from the standard database and merging the target features with the standard image data; using supervised learning based on data labels, the merged data is used to train the image recognition model to identify and infer relevant information from the image data and standard database information;

[0130] The result output module is responsible for collecting data from the standard database and real-time image data, and providing it to the image recognition model for prediction and inference; cross-validating the output results of the image recognition model with the information in the standard database, and including the corresponding database information in the output results.

[0131] The working principle and beneficial effects of the above technical solution are as follows: the data processing module of this embodiment collects sample data, constructs a sample library of quality defects such as missing bolts and damaged wires, and establishes a standard database based on the modular design of the line three-dimensional set; the information in the standard database is associated with the standard image data, and a connection between the standard image data and the standard database is established; wherein the standard image data is a photo or video data of line facilities that meet the line quality requirements obtained through equipment such as cameras or drones, showing the specific details of the power facilities, including the appearance and status of components such as towers, insulators and wires; the model building module extracts target features from the standard database and merges the target features with the standard image data; supervised learning is used to train the image recognition model based on the data labels, and the merged data is used to identify and infer relevant information from the image data and the standard database information; the result output module obtains data from the standard database and real-time image data, and provides it to the image recognition model for prediction and inference; the output result of the image recognition model is cross-validated with the information in the standard database, and the corresponding database information is included in the output result. The data processing module of the above scheme collects sample data, builds a quality defect sample library, and provides a data basis for model training; based on the modular design of the line three-dimensional set, a standard database is established, and it is associated with the standard image data to establish the connection between the data; a comprehensive and reliable data foundation is established to provide necessary support for the subsequent work of the system; through the association of the standard database and image data, standardized training data is provided for model construction to ensure the data quality and consistency of model training. The model construction module extracts target features from the standard database and merges them with the standard image data to construct a training data set; uses a supervised learning algorithm to train the image recognition model according to the data label to realize the identification and inference of quality defects; extracts important features and trains the model to realize the accurate identification of quality defects and improve work efficiency and accuracy; through supervised learning, the model can learn and generalize new samples, so that the system has the ability to self-learn and adapt to new situations. The result output module collects real-time data, provides it to the trained image recognition model for prediction and inference, and outputs the recognition results; cross-validates the output results of the image recognition model with the information of the standard database to further verify the accuracy of the recognition results; realizes the combination of the model and the real scene, so that the model can process the image data of the aerial work engineering entity in real time and enhance the practicality of the system; verifies the accuracy of the model output results through cross-validation, ensures the reliability and stability of the system, and provides key reference information for decision-making.

[0132] To sum up, the data processing module of this embodiment is responsible for collecting and organizing data, the model building module is used to train the model to realize the identification of quality defects, and the result output module is responsible for combining the model output with real-time data and verifying the accuracy of the identification results. These modules work together to enable the entire system to effectively realize the identification and management of quality defects of aerial work engineering entities.

[0133] Embodiment 11: Based on Embodiment 10, the data processing module provided in the embodiment of the present invention comprises:

[0134] The data acquisition submodule is responsible for determining the data types that need to be collected, including the engineering entity parts and risk factors for quality defect detection, such as missing bolts and wire damage; obtaining shooting equipment to inspect and shoot the engineering entity to obtain sample data containing quality defects;

[0135] The table creation submodule is responsible for annotating the collected data, marking the location of missing bolts and the area of ​​wire damage in each sample; converting and normalizing the data format, creating a database table, and importing the processed data into it;

[0136] The index establishment submodule is responsible for designing and establishing the index within the standard database, and integrating the standard image data and database information into the corresponding fields in the database to form a complete standardized data set.

[0137] The working principle and beneficial effects of the above technical solution are as follows: the data acquisition submodule of this embodiment determines the data type to be collected, including the engineering entity parts and risk factors for quality defect detection, and the risk factors include missing bolts and wire damage, etc.; obtains shooting equipment to inspect and shoot the engineering entity, and obtains sample data containing quality defects; the table creation submodule annotates the collected data, marks the location of the missing bolts and the area of ​​wire damage in each sample; converts and normalizes the data format, creates a database table, and imports the processed data into it; the index establishment submodule designs and establishes the index in the standard database, and integrates the standard image data and database information into the corresponding fields in the database to form a complete standardized data set. The data acquisition submodule of the above scheme determines the engineering entity parts and risk factors for quality defect detection that need to be collected, such as missing bolts and wire damage; clarifies the data type to be collected, provides guidance for subsequent data collection, and ensures that data related to the problem is obtained; uses shooting equipment to inspect and shoot the engineering entity, and obtains sample data containing quality defects; obtains sample data in real scenes for training models and establishing databases. The table creation submodule annotates the collected data in detail, marking the locations of missing bolts and damaged wires; performs format conversion and normalization processing; the annotation provides labeled training data for the model, and the format conversion and normalization processing help to organize and prepare the data; establish a database table and import the processed data into it; create a database table with a structured format to facilitate data management and subsequent model training, data analysis and other operations. The index establishment submodule designs and establishes indexes within the standard database for fast indexing and querying of data; the establishment of indexes can improve the efficiency and speed of database queries and optimize the data retrieval process; integrate standard image data and database information into the corresponding fields in the database to build a complete standardized data set; ensure that the information in the database accurately corresponds to the image data, and provide a reliable data foundation for subsequent model training and data analysis.

[0138] In summary, through the above process, the submodules of the data processing module in this embodiment are responsible for data acquisition, processing, database construction, index design, etc., providing an orderly and standardized data foundation for subsequent data analysis and model training.

[0139] Embodiment 12: Based on Embodiment 10, the model building module provided in the embodiment of the present invention comprises:

[0140] The feature extraction submodule is responsible for extracting target feature data from the standard database, merging the target feature data with the standard image data, and constructing a training data set. The training data set contains labeled samples and corresponding features / labels; extracting features from the merged training data set, and converting the image data into a computer-recognizable feature vector;

[0141] The parameter adjustment submodule is responsible for selecting the supervised learning algorithm of the convolutional neural network, extracting the test set from the merged dataset, training the image recognition model using the training dataset, and adjusting the parameters to minimize the prediction error by learning the relationship between features and labels;

[0142] The model evaluation submodule is responsible for evaluating the trained image recognition model using the test set. It evaluates the performance of the image recognition model and whether it can be generalized to new data by calculating indicators such as accuracy, precision, recall rate and F1 score. After the evaluation is completed, the trained image recognition model is used in actual application scenarios to identify quality defects of aerial work engineering entities.

[0143] The working principle and beneficial effects of the above technical solution are as follows: the feature extraction submodule of this embodiment extracts target feature data from a standard database, merges the target feature data with standard image data, and constructs a training data set, which contains labeled samples and corresponding features / labels; performs feature extraction on the merged training data set, and converts the image data into a computer-recognizable feature vector; the parameter adjustment submodule selects a supervised learning algorithm of a convolutional neural network, extracts a test set from the merged data set, and uses the training data set to train an image recognition model, and adjusts parameters to minimize prediction errors by learning the relationship between features and labels; the model evaluation submodule uses the test set to evaluate the trained image recognition model, and evaluates the performance of the image recognition model and whether it can be generalized to new data by calculating indicators such as accuracy, precision, recall rate and F1 score; after the evaluation is completed, the trained image recognition model is used in actual application scenarios to identify quality defects of aerial work engineering entities. The feature extraction submodule of the above scheme extracts feature data related to quality defects from the standard database to provide input features for model training; ensures that the training data contains key feature information, which helps the model learn the features and patterns of quality defects; merges the target feature data with the image data to construct a training data set containing labeled samples and corresponding features / labels; provides training data for the model, converts the image data into computer-recognizable feature vectors, and prepares for model training. The parameter adjustment submodule selects a supervised learning algorithm suitable for the problem, adjusts the model parameters, minimizes prediction errors, and improves model performance; optimizes model parameters by learning the relationship between features and labels, and improves the performance of the model in quality defect identification tasks. The model evaluation submodule uses an independent test set to evaluate the performance of the trained model, and calculates indicators such as accuracy, precision, recall, and F1 score; by evaluating the performance of the model, the generalization ability of the model on unknown data is judged to guide subsequent model improvement and application deployment; the trained image recognition model is used in actual application scenarios to identify quality defects of aerial work engineering entities; the model is put into practical application to realize automatic identification and monitoring of quality defects of aerial work engineering entities, and improve work efficiency and quality.

[0144] In summary, through the above process, each sub-module in the model building module of this embodiment is responsible for aspects from feature extraction, model parameter adjustment to performance evaluation, etc., providing a standardized and systematic method for building and optimizing the model, which helps to ensure that the model can accurately identify quality defects and achieve good performance in actual scenarios.

[0145] Embodiment 13: Based on Embodiment 10, the result output module provided in the embodiment of the present invention comprises:

[0146] The prediction and inference submodule is responsible for using the trained image recognition model to predict and infer real-time image data and obtain the model output results, that is, the identified quality defect information;

[0147] The information acquisition submodule is responsible for obtaining the standard information corresponding to the real-time image from the standard database, including the correct quality defect label or description;

[0148] The cross-validation submodule is responsible for comparing the output results of the image recognition model with the information in the standard database for cross-validation;

[0149] Compare the quality defect information output by the image recognition model with the actual situation in the standard database to ensure that the model correctly identifies the quality defects;

[0150] Count the matches between the output results of the image recognition model and the standard database information, and calculate the matching rate or accuracy;

[0151] Analyze situations where image recognition model outputs do not match standard databases to understand the types and causes of errors.

[0152] The working principle and beneficial effects of the above technical solution are as follows: the prediction and inference submodule of this embodiment uses the trained image recognition model to predict and infer the real-time image data, and obtains the results of the model output, that is, the identified quality defect information; the information acquisition submodule obtains the standard information corresponding to the real-time image from the standard database, including the correct quality defect label or description; the cross-validation submodule compares the output results of the image recognition model with the information in the standard database for cross-validation; the quality defect information output by the image recognition model is compared with the actual situation in the standard database to ensure whether the model correctly identifies the quality defect; the image recognition model output results match the standard database information, and calculate the matching rate or accuracy; the image recognition model output does not match the standard database, and understand the error type and cause. The prediction and inference submodule of the above scheme predicts the real-time image data through the trained model, obtains the results of the model output, that is, the quality defect information; realizes the automatic recognition and classification of quality defects in the real-time image, and improves work efficiency and quality. The information acquisition submodule obtains the standard information related to the real-time image from the standard database, including the correct quality defect label or description; as the standard reference of the model output, it is used to compare with the quality defect information output by the model to ensure the accuracy and reliability of the model. The cross-validation submodule compares the quality defect information output by the model with the information in the standard database for cross-validation; ensures model accuracy: verifies whether the model correctly identifies quality defects by comparing the model output and standard information; evaluates the matching rate: counts the matching between the model output and the standard database, calculates the accuracy rate, and evaluates the performance of the model; analyzes the cause of the error: analyzes the mismatch, understands the error type and cause, and guides model improvement and optimization.

[0153] In summary, through the above process, each submodule of the result output module of this embodiment is responsible for the process from prediction inference, information acquisition to cross-validation, and provides a verification and evaluation method for the model output, ensuring the accuracy and quality of the model's output results, and helping to improve the reliability and application value of the model.

[0154] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention belong to the scope of equivalent technologies of the present invention, the present invention is also intended to include these modifications and variations.

Claims

1. A method for identifying quality defects of aerial work engineering entities based on AI, characterized in that: The following steps are involved: Collect sample data, build a sample library of bolt missing and wire damage quality defects, and establish a standard database based on the modular design of the line three-dimensional set; associate the information in the standard database with the standard image data, and establish a connection between the standard image data and the standard database; Extract target features from the standard database and merge the target features with the standard image data; use supervised learning to train the image recognition model based on the data labels, and identify and infer relevant information from the image data and the standard database information; Data is collected from standard databases and real-time image data and provided to the image recognition model for prediction and inference; the output of the image recognition model is cross-validated with the information in the standard database, and the corresponding database information is included in the output.

2. The AI-based method for identifying quality defects of aerial work engineering entities according to claim 1, characterized in that: The process of establishing the connection between the standard image data and the standard database includes the following steps: Determine the type of data that needs to be collected, obtain photographic equipment to inspect and photograph the engineering entity, and obtain sample data containing quality defects; Label the collected data, mark the location of missing bolts and the area of ​​wire damage in each sample; convert and normalize the data, create a database table, and import the processed data into it; Design and establish indexes within the standard database, and integrate standard image data and database information into corresponding fields in the database to form a complete standardized data set.

3. The AI-based method for identifying quality defects of aerial work engineering entities as claimed in claim 2, characterized in that: Data types include engineering entity parts and risk factors for quality defect detection. Risk factors include missing bolts and damaged wires.

4. The AI-based method for identifying quality defects of aerial work engineering entities as claimed in claim 2, characterized in that: The process of forming a complete standardized data set includes the following steps: Design the database table structure, with a table containing quality defect information and a table containing image data and paths; Use the database management tool to execute the SQL statement to create the table; Import the marked, format-converted and normalized data into the corresponding database table, and correspond the data to the correct fields; Select the field to be used for query as the index field; Use the database's create index statement to create an index on the required fields of the database table; Integrate standard image data and database information into corresponding fields in the database, and use primary key and foreign key related fields to establish the connection between image data and quality defect information.

5. The AI-based method for identifying quality defects of aerial work engineering entities as claimed in claim 2, characterized in that: Database table structure, including field names, data types, and constraints.

6. The AI-based method for identifying quality defects of aerial work engineering entities as claimed in claim 4, characterized in that: The process of feature extraction for the merged training dataset includes the following steps: Preprocess the image data in the merged training dataset, including resizing the image, normalizing pixel values, and removing noise. Use the pre-trained convolutional neural network to extract image features by extracting the feature vector of the convolutional layer. For the features extracted from each image, different types of features are concatenated together to form a feature vector, and the feature vectors extracted from different images are filled to convert them to a fixed length; Traverse the entire feature vector, find the maximum and minimum values ​​of each feature dimension, and map each feature value to the preset range.

7. The AI-based method for identifying quality defects of aerial work engineering entities according to claim 4, characterized in that: The process of evaluating the performance of an image recognition model and whether it can generalize to new data consists of the following steps: Use the trained image recognition model to predict the image data in the test set; Compare the predictions of the image recognition model with the actual labels and calculate the accuracy, precision, recall and F1 score; Analyze the values ​​of various indicators to determine the performance and generalization ability of the image recognition model.

8. The AI-based method for identifying quality defects of aerial work engineering entities according to claim 1, characterized in that: The cross-validation process includes the following steps: Use the trained image recognition model to predict and infer the real-time image data and obtain the model output results, that is, the identified quality defect information; Obtain standard information corresponding to the real-time image from the standard database, including correct quality defect labels or descriptions; Compare the output of the image recognition model with the information in the standard database for cross-validation.

9. The AI-based method for identifying quality defects of aerial work engineering entities as claimed in claim 8, characterized in that: Compare the quality defect information output by the image recognition model with the actual situation in the standard database to ensure that the model correctly identifies the quality defects; Count the matches between the output results of the image recognition model and the standard database information, and calculate the matching rate or accuracy; Analyze situations where image recognition model outputs do not match standard databases to understand the types and causes of errors.

10. An AI-based high-altitude work engineering entity quality defect identification system, characterized in that: Include: The data processing module is responsible for collecting sample data, building a sample library of bolt missing and wire damage quality defects, and establishing a standard database based on the modular design of the line three-dimensional set; associating the information in the standard database with the standard image data, and establishing a connection between the standard image data and the standard database; Among them, standard image data is photos or video data of line facilities that meet line quality requirements obtained by cameras or drone equipment, showing specific details of power facilities, including the appearance and status of poles, insulators and conductor components; The model building module is responsible for extracting target features from the standard database and merging the target features with the standard image data; using supervised learning based on the data labels, the merged data is used to train the image recognition model to identify and infer relevant information from the image data and the standard database information; The result output module is responsible for collecting data from the standard database and real-time image data, and providing it to the image recognition model for prediction and inference; cross-validating the output results of the image recognition model with the information in the standard database, and including the corresponding database information in the output results.

Images