Data management and cleaning system for road patrol

Abstract

The invention discloses a road patrol-oriented data management and cleaning system, which comprises a main control module, a data cleaning module, a database module, a UI module and a data interface module, and is characterized in that the data interface module is used for receiving road patrol data, verifying the road patrol data, converting the verified and qualified road patrol data into a standard format, and storing the converted data into the database module; the data cleaning module is used for cleaning redundant data in the database module according to disease geographic positions and image features, cleaning low-quality images in the database module according to shooting angles and imaging quality, and cleaning marking error information in the database module; and the UI module is used for issuing instructions to the data cleaning module, the database module and the data interface module through the main control module. The road patrol-oriented data management and cleaning system provided by the invention can effectively manage the road patrol data and is used for training the neural network model, so that the recognition accuracy of the neural network on road diseases is improved.

Description

Technical Field

[0001] The present application relates to the technical field of data management, and in particular to a data management and cleaning system for road inspections. Background Art

[0002] Road inspection refers to the discovery and recording of various road diseases and abnormal conditions through regular or irregular inspections to ensure the safety and smooth flow of roads. It is an important part of road maintenance. Road inspection can detect road problems in a timely manner, prevent accidents, extend the service life of roads, and reduce maintenance costs. With the acceleration of urbanization and the increase in traffic volume, road maintenance and management have become increasingly important. Traditional road inspection methods mainly rely on manual inspections, which are not only time-consuming and labor-intensive, but also prone to missed inspections and misjudgments.

[0003] In recent years, with the development of artificial intelligence technology, road damage recognition methods based on neural networks have gradually become mainstream. Neural networks can learn and extract features from a large amount of data to achieve automatic recognition of road damage, which not only improves the accuracy and efficiency of recognition, but also shows good robustness in complex road environments.

[0004] The training of neural network models requires a large amount of high-quality data, which covers different road types, weather conditions and lighting environments. After the road data is collected, it is necessary to annotate the road data and mark the defects in the image in the form of points, lines or boxes. In this process, a huge amount of data will be generated, so a complete system is needed to realize data conversion, storage and query; in addition, the road data also needs to be cleaned to remove low-quality and redundant data, and to correct the mislabeled and missed data generated during the labeling process. When there is low-quality or erroneous data in the data set, it will seriously affect the training effect of the neural network model, resulting in a decrease in the recognition accuracy of the neural network.

[0005] In the existing technology, the links of data labeling, data cleaning, storage, model training and deployment are relatively complete. Among them, data cleaning includes two methods: rule-based and statistical-based. These solutions have been widely used in the Internet field. However, in road inspections, there are additional business needs:

[0006] 1. There are clear regulations for various diseases, and data definitions need to comply with industry standards and specifications;

[0007] 2. In actual images, the disease forms are diverse, and the physical scale span of the same type of disease is large;

[0008] 3. Depending on the inspection method and frequency, there may be missing or redundant data;

[0009] 4. Some diseases will worsen over time, and the types of diseases may change in different time periods;

[0010] 5. Some customers have their own data systems, and their data types need to be adapted.

[0011] This requires that the data management and cleaning of road inspections must support industry standards and operate at a more granular level. Existing solutions generally lack support for road inspection tasks at the standard and business levels, so there is a lack of guarantee for functional details and efficiency during the specific implementation process, resulting in low accuracy of road inspection results, and in turn, poor efficiency and effectiveness of road maintenance and management. Therefore, it is still necessary and valuable to design a data system specifically based on the application scenario characteristics of road inspections. Summary of the invention

[0012] To this end, the present application provides a data management and cleaning system for road inspections to solve the problems in the prior art that road inspection data cannot be effectively managed and has a low degree of automation, resulting in poor data interaction between different systems, low subsequent work efficiency, and the inability of trained neural networks to accurately identify road hazards.

[0013] In order to achieve the above objectives, this application provides the following technical solutions:

[0014] A data management and cleaning system for road inspection, comprising a main control module, a data cleaning module, a database module, a UI module and a data interface module, wherein the data cleaning module, the database module, the UI module and the data interface module are all electrically connected to the main control module, and the database module is also electrically connected to the data cleaning module and the data interface module;

[0015] Among them, the data interface module is used to receive road inspection data and perform verification, convert the verified road inspection data into a standard format and store it in the database module; the data cleaning module is used to clean the redundant data in the database module according to the geographical location of the disease and the image characteristics, to clean the low-quality images in the database module according to the shooting angle and imaging quality, and to clean the labeled error information in the database module; the UI module is used to issue instructions to the data cleaning module, the database module and the data interface module through the main control module.

[0016] Preferably, the data interface module is also used to access upstream and downstream tool chains and exchange data with the upstream and downstream tool chains.

[0017] Preferably, the upstream and downstream tool chain includes a model training system and a labeling platform.

[0018] Preferably, the data interface module verifies the road inspection data by specifically verifying whether the road inspection data is missing and whether the road inspection data is damaged.

[0019] Preferably, the data cleaning module is used to clean the redundant data in the database module according to the geographical location and image features of the disease, specifically including: if the type of the disease is the same, but the longitude and latitude information or the mileage from a specific pile number is consistent, it is determined to be duplicate data with the same geographical location; on the basis of the same geographical location, it is determined whether the image feature vectors of the disease in different images are similar, and if similar, it is determined to be duplicate data.

[0020] Preferably, the image feature vector is calculated based on a neural network.

[0021] Preferably, the image feature vector includes pixel coordinates, size, color, gradient and key points as well as pixel coordinate statistics, size statistics, color statistics, gradient statistics, key point statistics and a high-dimensional feature vector calculated by a neural network.

[0022] Preferably, when the data cleaning module is used to clean low-quality images in the database module according to shooting angle and imaging quality, the shooting angle is judged by the position of the disease in the image, and the imaging quality is judged by an unsupervised image quality evaluation index.

[0023] Preferably, the unsupervised image quality evaluation index includes frequency domain distribution, BRISQUE or DIQA.

[0024] Preferably, when the data cleaning module is used to clean the labeling error information in the database module, the labeling error information includes mislabeling and missing labeling.

[0025] Compared with the prior art, this application has at least the following beneficial effects:

[0026] The present application provides a data management and cleaning system for road inspection, including a main control module, a data cleaning module, a database module, a UI module and a data interface module, wherein the data interface module is used to receive road inspection data and perform verification, and convert the verified road inspection data into a standard format and store it in the database module; the data cleaning module is used to clean the redundant data in the database module according to the geographical location of the disease and the image characteristics, to clean the low-quality images in the database module according to the shooting angle and the imaging quality, and to clean the labeling error information in the database module; the UI module is used to issue instructions to the data cleaning module, the database module and the data interface module through the main control module. The present application provides a data management and cleaning system for road inspection, which can effectively manage road inspection data and be used to train neural network models, thereby improving the accuracy of neural network recognition of road diseases, significantly improving the automation level of road inspection, reducing labor costs, and improving the efficiency and effectiveness of road maintenance and management, providing strong support for the construction of smart cities. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more intuitively illustrate the prior art and the present application, several exemplary drawings are given below. It should be understood that the specific shapes and structures shown in the drawings should not generally be regarded as limiting conditions for implementing the present application; for example, those skilled in the art are capable of easily making conventional adjustments or further optimizations to the addition / reduction / attribution division, specific shapes, positional relationships, connection methods, dimensional ratios, etc. of certain units (components) based on the technical concepts and exemplary drawings disclosed in the present application.

[0028] Figure 1 A schematic diagram of the structure of a data management and cleaning system for road inspection provided in this application;

[0029] Figure 2 Data verification flow chart provided for this application;

[0030] Figure 3 Redundant data cleaning flow chart provided for this application;

[0031] Figure 4 A data transfer flow chart of a data management and cleaning system for road inspection provided by this application when used as a data transfer station;

[0032] Figure 5 A schematic diagram of the structure of a neural network training link when a data management and cleaning system for road inspection provided in this application is used for neural network training;

[0033] Figure 6 A model training flow chart for a data management and cleaning system for road inspection provided in this application when used for neural network training;

[0034] Figure 7 The overall working link provided for the embodiment of the present application.

[0035] Description of reference numerals:

[0036] 1. Main control module; 2. Data cleaning module; 3. Database module; 4. UI module; 5. Data interface module. DETAILED DESCRIPTION

[0037] The present application is further described below in detail through specific embodiments in conjunction with the accompanying drawings.

[0038] In the description of this application: unless otherwise specified, the meaning of "plurality" is two or more. The terms "first", "second", "third", etc. in this application are intended to distinguish the objects referred to, and do not have special meanings in terms of technical connotations (for example, they should not be understood as emphasizing the importance or order, etc.). Expressions such as "including", "comprising", "having", etc. also mean "not limited to" (certain units, components, materials, steps, etc.).

[0039] The terms such as "upper", "lower", "left", "right", "middle", etc. used in this application are usually used to facilitate intuitive understanding by comparing with the accompanying drawings, and are not absolute limitations on the positional relationship in the actual product. Without departing from the technical concept disclosed in this application, changes in these relative positional relationships should also be regarded as the scope of this application.

[0040] See also Figure 1 The present application provides a data management and cleaning system for road inspection, including a main control module 1, a data cleaning module 2, a database module 3, a UI module 4 and a data interface module 5. The data cleaning module 2, the database module 3, the UI module 4 and the data interface module 5 are all electrically connected to the main control module 1 for receiving instructions and reporting operating status. The database module 3 is also electrically connected to the data cleaning module 2 and the data interface module 5.

[0041] Among them, the data interface module 5 is used to receive road inspection data and verify it, convert the verified road inspection data into a standard format and store it in the database module 3; the data cleaning module 2 is used to clean the redundant data in the database module 3 according to the geographical location of the disease and the image characteristics, to clean the low-quality images in the database module 3 according to the shooting angle and imaging quality, and to clean the annotation error information in the database module 3; the UI module 4 is used to issue instructions to the data cleaning module 2, the database module 3 and the data interface module 5 through the main control module 1.

[0042] Specifically, the road inspection data collected by the external device is transmitted through the data interface module 5. After the road inspection data is transmitted to the system through the data interface module 5, it is converted into a standard format and stored in an independent database module 3. The data cleaning module 2 can perform secondary processing on the stored road inspection data to remove errors and redundant data. The user can issue instructions to the system through the UI module 4 to complete various functions.

[0043] It should be noted that road inspection data is mainly based on disease attributes, including basic attributes of diseases, sensor data and associated attributes. Among them, basic attributes include disease type, geographical location, inspection time, physical size and severity, etc.; sensor data is mainly image data, including disease images, pixel coordinates and dimensions and annotation information, etc., which are usually used to support neural network training, and also cover other types of sensor data such as IMU, radar, etc.; associated attributes are the changing trend of diseases and the association between disease types, such as cracks can become longer and wider over time, and deep cracks may develop into potholes, etc., and the associated attributes have a strong relationship with subsequent engineering repairs.

[0044] The road inspection data input through the data interface module 5 will first be checked, for example: whether there is any missing or damaged. If there is any missing or damaged, the system will promptly warn. Subsequently, the road inspection data will be converted into a standard format and then stored in the database module 3. Among them, data missing means that one or more items in the road inspection data are not input, and data damage means that the data content is not compliant. For example, the expected input data includes disease type, date, and location, and the actual input only includes disease type and date, which means data missing; if the input data is complete, but the date contains negative numbers, the data content is damaged.

[0045] The specific process for verifying road inspection data is as follows:

[0046] Step 1: Input road inspection data;

[0047] Step 2: Check whether the road inspection data is complete. If it is missing, execute 0, otherwise execute 0;

[0048] Step 3: If the missing data is required data, execute 0, otherwise execute 0;

[0049] Step 4: Check whether the road inspection data content is damaged. If damaged, execute 0, otherwise execute 0;

[0050] Step 5: Issue an alarm and execute 0;

[0051] Step 6: Convert the data into a standard format;

[0052] Step 7: Store the converted data into the database module 3;

[0053] Step 8: End.

[0054] The raw data collected during road inspections often need to be cleaned to ensure data quality and improve analysis accuracy, because raw data often contains a lot of redundant information. For example, when an inspection vehicle is driving, it often takes repeated photos of the same target. In subsequent work, only one or a few frames of images are usually needed, which requires deduplication of disease data.

[0055] On the other hand, imaging conditions will also affect data quality. When there are problems with the imaging system, the captured images may have quality problems such as noise and blur, which may make it difficult to clearly identify the defects. At the same time, the shooting angle of the defects may continue to change during the movement of the patrol vehicle. It is necessary to select the image taken at the optimal angle to avoid target occlusion and lens distortion. In addition, due to external conditions such as weather and lighting, some data may be completely unusable and should be cleared in a timely manner.

[0056] To support the training of neural network models, the raw data will be manually or semi-automatically labeled after preliminary cleaning, and the labeled information may also be mislabeled or missing. These labeling errors can be checked according to certain rules and finally eliminated from the data set with secondary manual verification.

[0057] Therefore, data cleaning module 2 is mainly used to clean redundant data, low-quality images and annotation information:

[0058] (1) Cleaning redundant data

[0059] When cleaning redundant data, the system supports deduplication by disease location and image features. When the disease type is the same, and the longitude and latitude information, or the distance from a specific pile number is basically the same, it can be regarded as duplicate data with the same geographical location.

[0060] The use of image features for deduplication requires the analysis of disease images based on geographical location matching. When the block features of the disease on different images are similar, they are considered duplicate data. Among them, image features include pixel coordinates, size, color, gradient, key points and statistical values ​​of the above features, such as mean, standard deviation and histogram, etc., as well as image feature vectors calculated by algorithm models and high-dimensional feature vectors calculated by neural networks. In order to ensure accuracy, this application mainly uses a method based on neural networks to calculate image feature vectors, supplemented by other image features.

[0061] See also Figure 3 , the specific process of cleaning redundant data is:

[0062] Step 1: Input multiple sets of data to be deduplicated;

[0063] Step 2: For a certain set of data, if the geographical location of this set of data is different from that of other sets of data, execute 0; otherwise, execute 0;

[0064] Step 3: Extract data image feature vector;

[0065] Step 4: For a certain set of data, if the image feature vectors of this set of data are not similar to those of other sets of data, execute 0, otherwise execute 0;

[0066] Step 5: Mark the data as non-duplicate data and proceed to step 8;

[0067] Step 6: Group similar data together;

[0068] Step 7: Mark the data as duplicate data;

[0069] Step 8: End.

[0070] (2) Cleaning low-quality images

[0071] The judgment of image quality mainly includes the judgment of shooting angle and the judgment of the image quality of the image itself. Among them, the shooting angle can be roughly judged by the position of the defect in the image. When the defect is close to the center of the image, it can be approximately considered that the shooting angle meets the requirements; the image quality can be judged by conventional unsupervised image quality evaluation indicators, including frequency domain distribution, BRISQUE or DIQA. Considering the computing power consumption, the image quality evaluation can be carried out by sampling. The data collected each time is segmented according to the mileage, and a small number of images are randomly selected in each segment for evaluation as the result of the entire segment data.

[0072] (3) Cleaning and labeling information

[0073] During the annotation process, defects are usually marked on the image using geometric shapes such as points, lines, rectangles or polygons, and are supplemented with relevant attribute values, mainly including: defect category, road scene, shape, position, rotation angle, size, color, and the number of defects of this category in the same image, the number of defects of this category in the entire data set, the geometric relationship between annotation boxes (such as overlap, inclusion, etc.), and the differences between images of the same geographical location at different times.

[0074] Labeling errors include mislabeling and missing labels. Mislabeling mainly includes labeling errors in the type, size, and position of the disease, while missing labels refer to the disease not being labeled. Mislabeling can be cleaned up using rules related to the characteristics of the disease. For example, the size and position of a label do not match the true characteristics of the disease of that category, the labeling boxes of different types of diseases completely overlap, and the labeling of cement road-specific diseases appears on the image of asphalt roads.

[0075] Missing labels require iteration with neural network model training. After the initial cleaning of the data set is completed, it is first passed to the downstream model training system to train the initial version of the model. The initial version of the model is then used to perform reasoning on the current data set, and images that do not match the labeling information are screened out and returned to the upstream labeling platform for secondary manual verification.

[0076] In a data management and cleaning system for road inspection provided by the present application, the data interface module 5 can serve as an interactive interface for upstream and downstream link tools. The entire system can be connected to other systems through the data interface module 5 for data exchange, and then serve as a transfer station between various data sources, or be used to support the training of neural network models.

[0077] (1) Data Transfer Station

[0078] When the system is used as a data transfer station, the input data is parsed and converted into the format required by the downstream link for output, and the data is backed up to the database module 5. The system itself does not use the input data. After the data is stored in the system, the subsequent downstream link can directly obtain the data from the system without accessing the upstream link. Figure 4 shown.

[0079] The specific process is:

[0080] Step 1: Receive input data;

[0081] Step 2: Parse and verify data content;

[0082] Step 3: Store the data in the database as a backup;

[0083] Step 4: Format the data into the format required by the downstream link;

[0084] Step 5: Output the data.

[0085] (2) Neural network training

[0086] Another major function of the system is to manage and clean the neural network training data. At this point, the system is connected to the data input source, image annotation platform and model training system at the same time, such as Figure 5 As shown in the figure. In this way, the system can incrementally synchronize data between the annotation system and the training system without being restricted by the original interface of the external system. When the data set is large, this method can effectively reduce the amount of data transmission and improve development efficiency.

[0087] In this mode, the system works as follows: Figure 6 As shown, the specific steps are:

[0088] Step 1: Receive raw data;

[0089] Step 2: Preliminary cleaning to remove redundant data;

[0090] Step 3: The cleaned data is transferred to the annotation platform for annotation;

[0091] Step 4: Clean the labeled data;

[0092] Step 5: The labeled data is transmitted to the model training system to train the model;

[0093] Step 6: Clean the data again after getting the model;

[0094] Step 7: Repeat 0 to 0 until the data is cleaned.

[0095] Example:

[0096] Using the data system proposed in this application to transform the original data labeling and model training links can improve work efficiency.

[0097] The overall working link includes a data server, a model training server, and a R&D server. The data server is equipped with an open source annotation tool, and the training server is equipped with an open source automated training framework. In the original link, the data formats used by the annotation tool, automated training framework, and R&D server are all different. When updating data, it is necessary to export the entire data from the annotation tool and manually convert the format. The training server and R&D server each require two full data transmissions and manual conversions.

[0098] After the transformation, the data system proposed in this application is introduced into the link as an intermediate layer, and the data server is shared with the annotation tool, and connected with other modules and servers through the data interface module 5. The specific structure is as follows Figure 7 shown.

[0099] The processed road inspection data includes basic attributes of defects such as inspection time and geographic location, and defect images. When new data is generated, the data system imports and parses the original data. Among them, the basic attributes of the defects are directly stored in the database module 3, and the defect images are stored in the database module 3 after base64 encoding, and are imported into the annotation tool through the data interface module 5.

[0100] After the disease image is manually annotated, annotation information in Json format is generated, for example:

[0101] [{"id":"25d7946409","type":"rectanglelabels","value":{"x":21.4583,"y":76.574065,"width":8.125,"height":4.166 67,"rotation":0,"rectanglelabels":["Height difference of manhole cover"]},"image_rotation":0,"original_width":1920,"original_height":1080}]

[0102] It can be seen that since the annotation tool is a general tool, fixed fields cannot be used, and the annotation information contains redundant information. This will result in parsing every time the data is accessed, affecting the query efficiency and inconvenience for subsequent data cleaning. Therefore, after the annotation information is transmitted back to the data system, it will be converted to a standard format, the field name will be removed, only its value will be retained, and redundant fields such as id, type, image_rotation will be removed and stored in the database. At this time, the user can check and clean the data through the data system UI module 4.

[0103] After the data is cleaned, it is transferred to the training framework and R&D server in the form of files. Before the transfer, the data system will first scan the existing files and synchronize the new data to the server hard disk in incremental form. When the data set is large, incremental transmission will effectively reduce the amount of data transmission. During synchronization, the data can be converted into multiple formats such as VOC, COCO and YOLO. Users can configure the synchronization process through the data system UI.

[0104] After the model training is completed, it can be transmitted back to the data system through the data interface. Users can also upload model files through the system UI module 4 for subsequent data cleaning.

[0105] The present application provides a data management and cleaning system for road inspections. External data is input through a data interface module 5, which can also be used as an interactive interface for upstream and downstream link tools. After the data is input into the system, it is converted into a standard format and stored in an independent database module 3. The data cleaning module 2 can perform secondary processing on the stored data to remove errors and redundant data. Users can issue instructions to the system through the UI module 4 to complete various functions. The system supports the storage and rapid retrieval of large-scale data, and can use a variety of image processing and deep learning algorithms to clean the data, thereby improving the model training effect and effectively improving the accuracy and efficiency of road inspections.

[0106] The data management and cleaning system for road inspection provided by this application can effectively solve the data quality problems existing in the prior art and improve the recognition accuracy of neural networks. Through the implementation of this application, road inspection data can be effectively managed and used to train neural network models, thereby improving the recognition accuracy of neural networks for road diseases, significantly improving the automation level of road inspections, reducing labor costs, and improving the efficiency and effectiveness of road maintenance and management, providing strong support for the construction of smart cities.

[0107] In summary, the data management and cleaning system for road inspection provided by this application has the following advantages:

[0108] (1) Designed to meet the needs of road inspection business, it can effectively manage and clean inspection data;

[0109] (2) Support neural network model training and automated inspection;

[0110] (3) The data structure design refers to national standards and industry specifications to facilitate the integration and standardization of data from different sources;

[0111] (4) The system interface is flexible, facilitating business system access and data import.

[0112] The technical features of the above embodiments may be arbitrarily combined (as long as there is no contradiction in the combination of these technical features). To make the description concise, not all possible combinations of the technical features in the above embodiments are described; these embodiments that are not explicitly written should also be considered to be within the scope of this specification.

[0113] The above is a relatively specific and detailed description of the present application through general description and specific embodiments. It should be understood that, based on the technical concept of the present application, several conventional adjustments or further innovations can be made to these specific embodiments; however, as long as they do not deviate from the technical concept of the present application, the technical solutions obtained by these conventional adjustments or further innovations also fall within the scope of protection of the claims of the present application.

Claims

1. A data management and cleaning system for road inspection, characterized in that: It includes a main control module, a data cleaning module, a database module, a UI module and a data interface module, wherein the data cleaning module, the database module, the UI module and the data interface module are all electrically connected to the main control module, and the database module is also electrically connected to the data cleaning module and the data interface module; Among them, the data interface module is used to receive road inspection data and perform verification, convert the verified road inspection data into a standard format and store it in the database module; the data cleaning module is used to clean the redundant data in the database module according to the geographical location of the disease and the image characteristics, to clean the low-quality images in the database module according to the shooting angle and imaging quality, and to clean the labeled error information in the database module; the UI module is used to issue instructions to the data cleaning module, the database module and the data interface module through the main control module.

2. The data management and cleaning system for road inspection according to claim 1 is characterized in that: The data interface module is also used to access the upstream and downstream tool chains and exchange data with the upstream and downstream tool chains.

3. The data management and cleaning system for road inspection according to claim 2 is characterized in that: The upstream and downstream tool chains include a model training system and an annotation platform.

4. The data management and cleaning system for road inspection according to claim 1 is characterized in that: The data interface module verifies the road inspection data by specifically verifying whether the road inspection data is missing and whether the road inspection data is damaged.

5. The data management and cleaning system for road inspection according to claim 1 is characterized in that: The data cleaning module is used to clean the redundant data in the database module according to the geographical location and image features of the disease, specifically including: if the type of the disease is the same, but the longitude and latitude information or the mileage from a specific pile number is consistent, it is determined to be duplicate data with the same geographical location; on the basis of the same geographical location, it is determined whether the image feature vectors of the disease on different images are similar, and if they are similar, it is determined to be duplicate data.

6. The data management and cleaning system for road inspection according to claim 5 is characterized in that: The image feature vector is calculated based on a neural network.

7. The data management and cleaning system for road inspection according to claim 5 is characterized in that: The image feature vector includes pixel coordinates, size, color, gradient and key points, as well as pixel coordinate statistics, size statistics, color statistics, gradient statistics, key point statistics and a high-dimensional feature vector calculated by a neural network.

8. The data management and cleaning system for road inspection according to claim 1 is characterized in that: The data cleaning module is used to clean the low-quality images in the database module according to the shooting angle and the imaging quality. The shooting angle is judged by the position of the defect in the image, and the imaging quality is judged by the unsupervised image quality evaluation index.

9. The data management and cleaning system for road inspection according to claim 8, characterized in that: The unsupervised image quality evaluation index includes frequency domain distribution, BRISQUE or DIQA.

10. The data management and cleaning system for road inspection according to claim 1, characterized in that: When the data cleaning module is used to clean the labeling error information in the database module, the labeling error information includes mislabeling and missing labeling.

Images