Power business image video library construction method based on fine management

Abstract

The invention relates to a power business image video library construction method based on fine management. The method comprises the following steps: constructing a data transmission link; processing the accessed data by using an image desensitization algorithm and an image quality evaluation algorithm to obtain treated data, exporting unqualified information and sending the unqualified information to power transformation operation and maintenance personnel; performing image annotation on the treated data by using an automatic intelligent annotation tool; then intensive management is carried out, various samples are quickly retrieved and combined, and meanwhile authority and version management is carried out. The sample data collected by the terminal is directly converged to the provincial platform side, the sample collection efficiency is improved by more than 300% compared with a traditional mode, and rich and efficient data support is provided for training and optimization of an artificial intelligence algorithm; through triple guarantee of a quality evaluation algorithm, automatic labeling and auditing, a standard and unified sample library is formed, the accuracy and reliability of data used by artificial intelligence algorithm training are ensured, and the performance and stability of the algorithm are further improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent operation and maintenance of electric power systems, and in particular to a method for constructing an electric power business image and video library based on refined management. Background Art

[0002] With the rapid development of power systems and the improvement of their intelligence, substation operation and maintenance management is also facing new challenges and opportunities. Traditional substation inspection methods have problems such as low efficiency and poor accuracy, and cannot meet the requirements of modern power grids for operation and maintenance efficiency and safety. Therefore, the substation remote intelligent inspection system came into being, which integrates high-definition cameras, sensors and other equipment to achieve remote real-time monitoring and intelligent analysis of substations.

[0003] However, in the actual application of the remote intelligent inspection system for substations, the processing and management of sample data has become an urgent problem to be solved. On the one hand, the inspection data of substations is large in volume and diverse in types, including images, videos, sensor data, etc. How to effectively collect, organize and store this data has become a technical problem. On the other hand, due to the wide variety of substation equipment and complex defect types, high requirements are also placed on the labeling and classification of sample data. Traditional labeling methods are not only inefficient, but also prone to errors, and cannot meet the processing requirements of large-scale data sets.

[0004] In addition, in the process of sample data management, data quality control and security are also crucial. If there are low-quality samples such as blur, distorted screen, black screen, etc. in the data, it will seriously affect the subsequent intelligent analysis and model training results. At the same time, the transmission and use of data must also comply with relevant laws, regulations and privacy policies to ensure the legality and security of the data.

[0005] Therefore, in response to the above problems, the present invention proposes a power business image and video library construction platform based on refined management, aiming to achieve comprehensive, efficient and secure management of sample data through automated technical means, and provide strong support for the intelligent operation and maintenance of substations. Summary of the invention

[0006] In order to solve the problems of cumbersome and inefficient uploading, governance, labeling and management of sample data in the current substation remote intelligent patrol system and artificial intelligence application support system, the purpose of the present invention is to provide a method for constructing a power business image video library based on refined management, which greatly improves the efficiency of sample data collection and significantly reduces the rate of unqualified samples.

[0007] To achieve the above object, the present invention adopts the following technical solution: a method for constructing an image and video library of electric power business based on refined management, the method comprising the following steps in order:

[0008] (1) Data access: Build a data transmission link to receive image and video sample data related to various power services from the substation remote intelligent patrol system and automatically transmit it to the artificial intelligence application support system;

[0009] (2) Data governance: Use image desensitization algorithms and image quality assessment algorithms to process the data received in step (1) to obtain governed data, and export unqualified information and send it to substation operation and maintenance personnel;

[0010] (3) Perform data labeling and sample management: first define the labeling specifications, then use automatic intelligent labeling tools to label the images of the managed data; then perform intensive management to quickly retrieve and combine various types of samples to generate a second sample set that can be used for training or evaluation, while also performing permission and version management.

[0011] The step (1) specifically comprises the following steps in order:

[0012] (1a) Build a data transmission link, use the substation remote intelligent patrol system as the data source, and automatically transmit various sample data to the artificial intelligence application support system;

[0013] (1b) Capturing and refining core information: While the data is being uploaded, the artificial intelligence application support system extracts the equipment type, component type, shooting time point, and defect type information in the sample; the core information includes the equipment type, component type, shooting time point, and defect type information in the sample;

[0014] (1c) Classify and archive core information.

[0015] The step (2) specifically comprises the following steps in order:

[0016] (2a) Image desensitization and governance screening: The AI ​​application support system uses an image desensitization algorithm to automatically desensitize the sample data received in step (1) to protect sensitive information. At the same time, it uses an image quality assessment algorithm to identify and remove low-quality sample data such as blurry, distorted, and black screens.

[0017] (2b) The artificial intelligence application support system exports unqualified information and sends it to the substation operation and maintenance personnel for subsequent processing; the unqualified information refers to the point equipment information with unqualified image quality.

[0018] The step (3) specifically comprises the following steps in order:

[0019] (3a) First, define the universal marking specifications for substations to provide a unified standard for subsequent marking work;

[0020] (3b) using an automatic intelligent annotation tool to annotate the managed data, providing annotation examples as references, and the annotated data forms a first sample set; the automatic intelligent annotation tool uses annotation software based on deep learning, uses a neural network model to deeply analyze image content, intelligently identifies key elements and features in the image, and automatically assigns relevant labels or descriptions to the image;

[0021] (3c) Intensively manage the labeled sample data, and support the rapid combination of sample data based on conditions such as defect type, photographed equipment, photographed parts, and acquisition time;

[0022] (3d) The data after conditional rapid combination generates a second sample set that can be used for training or evaluation, meeting the needs of model evaluation or iteration;

[0023] (3e) Rights management: uniform rights management is performed on the generated second sample set, and a data use approval mechanism is established to ensure the security of data transmission;

[0024] (3f) Version management: The second sample set is annotated or modified again to generate different versions of the third sample set to meet different application requirements.

[0025] It can be seen from the above technical scheme that the beneficial effects of the present invention are as follows: First, the present invention greatly improves the efficiency of sample data collection: the sample data collected by the terminal is directly aggregated to the provincial platform side, avoiding the cumbersome process of multi-level transmission, so that each provincial network company can easily form more than 200,000 standardized sample data each year, and the sample collection efficiency is improved by more than 300% compared with the traditional method, providing rich and efficient data support for the training and optimization of artificial intelligence algorithms; Second, the present invention significantly reduces the rate of unqualified samples: through the triple guarantee of quality assessment algorithm, automatic labeling and review, a standardized and unified sample library is formed, and the rate of unqualified samples is reduced to less than 1%, ensuring the accuracy and reliability of the data used for artificial intelligence algorithm training, and further improving the performance and stability of the algorithm; Third, after the present invention reviews the sample data uploaded by the business system, it can quickly count the actual application indicators of the artificial intelligence algorithms of each business system, effectively urge each algorithm manufacturer to optimize and iterate the model, so that the reliability of the actual application indicator statistics is increased to 100%, which provides strong data support for the continuous optimization and upgrading of the remote intelligent patrol system of the substation. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a schematic diagram of the framework of the present invention;

[0027] Figure 2 A schematic diagram of data access in the present invention;

[0028] Figure 3It is a schematic diagram of data governance in the present invention;

[0029] Figure 4 Schematic diagram of data labeling and sample management in the present invention. DETAILED DESCRIPTION

[0030] like Figure 1 As shown, a method for constructing a power business image and video library based on refined management includes the following steps in sequence:

[0031] (1) Data access: Build a data transmission link to receive image and video sample data related to various power services from the substation remote intelligent patrol system and automatically transmit it to the artificial intelligence application support system;

[0032] (2) Data governance: Use image desensitization algorithms and image quality assessment algorithms to process the data received in step (1) to obtain governed data, and export unqualified information and send it to substation operation and maintenance personnel;

[0033] (3) Perform data labeling and sample management: first define the labeling specifications, then use automatic intelligent labeling tools to label the images of the managed data; then perform intensive management to quickly retrieve and combine various types of samples to generate a second sample set that can be used for training or evaluation, while also performing permission and version management.

[0034] like Figure 2 As shown, the step (1) specifically includes the following steps in order:

[0035] (1a) Build a data transmission link, use the substation remote intelligent patrol system as the data source, and automatically transmit various sample data to the artificial intelligence application support system;

[0036] (1b) Capturing and refining core information: While the data is being uploaded, the artificial intelligence application support system extracts the equipment type, component type, shooting time point, and defect type information in the sample; the core information includes the equipment type, component type, shooting time point, and defect type information in the sample;

[0037] (1c) Classify and archive core information.

[0038] like Figure 3 As shown, the step (2) specifically includes the following steps in order:

[0039] (2a) Image desensitization and governance screening: The AI ​​application support system uses an image desensitization algorithm to automatically desensitize the sample data received in step (1) to protect sensitive information. At the same time, it uses an image quality assessment algorithm to identify and remove low-quality sample data such as blurry, distorted, and black screens.

[0040] (2b) The artificial intelligence application support system exports unqualified information and sends it to the substation operation and maintenance personnel for subsequent processing; the unqualified information refers to the point equipment information with unqualified image quality.

[0041] like Figure 4 As shown, the step (3) specifically includes the following steps in order:

[0042] (3a) First, define the universal marking specifications for substations to provide a unified standard for subsequent marking work;

[0043] (3b) Annotating the data after treatment using an automatic intelligent annotation tool, providing annotation examples as references, and the annotated data forms a first sample set; the automatic intelligent annotation tool uses annotation software based on deep learning, uses a neural network model to deeply analyze the image content, intelligently identifies key elements and features in the image, and automatically assigns relevant labels or descriptions to the image; the automatic annotation software based on deep learning combines model and data training. Specifically, first, a deep learning model is established by collecting a large number of high-quality annotated data sets, and then the model is trained using these data to learn the characteristics and laws of the data. After the training is completed, the model can automatically predict the labels of unlabeled data;

[0044] (3c) Intensively manage the labeled sample data, and support the rapid combination of sample data based on conditions such as defect type, photographed equipment, photographed parts, and acquisition time;

[0045] (3d) The data after conditional rapid combination generates a second sample set that can be used for training or evaluation, meeting the needs of model evaluation or iteration;

[0046] (3e) Rights management: uniform rights management is performed on the generated second sample set, and a data use approval mechanism is established to ensure the security of data transmission;

[0047] (3f) Version management: The second sample set is annotated or modified again to generate different versions of the third sample set to meet different application requirements.

[0048] In summary, the present invention greatly improves the efficiency of sample data collection: the sample data collected by the terminal is directly aggregated to the provincial platform side, avoiding the tedious process of multi-level transmission, so that each provincial network company can easily form more than 200,000 standardized sample data each year, and the sample collection efficiency is improved by more than 300% compared with the traditional method, providing rich and efficient data support for the training and optimization of artificial intelligence algorithms; the present invention significantly reduces the rate of unqualified samples: through the triple guarantee of quality assessment algorithm, automatic labeling and review, a standardized and unified sample library is formed, and the rate of unqualified samples is reduced to below 1%, ensuring the accuracy and reliability of the data used for artificial intelligence algorithm training, and further improving the performance and stability of the algorithm; after the present invention reviews the sample data uploaded by the business system, it can quickly count the actual application indicators of the artificial intelligence algorithms of each business system, effectively urge each algorithm manufacturer to optimize and iterate the model, so that the reliability of the actual application indicator statistics is improved to 100%, which provides strong data support for the continuous optimization and upgrading of the remote intelligent patrol system of the substation.

Claims

1. A method for constructing an image and video library of electric power business based on refined management, characterized in that: The method comprises the following steps in order: (1) Data access: Build a data transmission link to receive image and video sample data related to various power services from the substation remote intelligent patrol system and automatically transmit it to the artificial intelligence application support system; (2) Data governance: Use image desensitization algorithms and image quality assessment algorithms to process the data received in step (1) to obtain governed data, and export unqualified information and send it to substation operation and maintenance personnel; (3) Perform data labeling and sample management: first define the labeling specifications, then use automatic intelligent labeling tools to label the images of the managed data; then perform intensive management to quickly retrieve and combine various types of samples to generate a second sample set that can be used for training or evaluation, while also performing permission and version management.

2. The method for constructing a power business image and video library based on refined management according to claim 1 is characterized in that: The step (1) specifically comprises the following steps in order: (1a) Build a data transmission link, use the substation remote intelligent patrol system as the data source, and automatically transmit various sample data to the artificial intelligence application support system; (1b) Capturing and refining core information: While the data is being uploaded, the artificial intelligence application support system extracts the equipment type, component type, shooting time point, and defect type information in the sample; the core information includes the equipment type, component type, shooting time point, and defect type information in the sample; (1c) Classify and archive core information.

3. The method for constructing a power business image and video library based on refined management according to claim 1 is characterized in that: The step (2) specifically comprises the following steps in order: (2a) Image desensitization and governance screening: The artificial intelligence application support system uses an image desensitization algorithm to automatically desensitize the sample data received in step (1) to protect sensitive information; At the same time, the image quality assessment algorithm is used to identify and remove low-quality sample data such as blur, distorted screen, and black screen; (2b) The artificial intelligence application support system exports unqualified information and sends it to the substation operation and maintenance personnel for subsequent processing; the unqualified information refers to the point equipment information with unqualified image quality.

4. The method for constructing a power business image and video library based on refined management according to claim 1 is characterized in that: The step (3) specifically comprises the following steps in order: (3a) First, define the universal marking specifications for substations to provide a unified standard for subsequent marking work; (3b) annotating the managed data using an automatic intelligent annotation tool, providing annotation examples as references, and the annotated data forming a first sample set; the automatic intelligent annotation tool uses annotation software based on deep learning; (3c) Intensively manage the labeled sample data, and support the rapid combination of sample data based on conditions such as defect type, photographed equipment, photographed parts, and acquisition time; (3d) The data after conditional rapid combination generates a second sample set that can be used for training or evaluation, meeting the needs of model evaluation or iteration; (3e) Rights management: uniform rights management is performed on the generated second sample set, and a data use approval mechanism is established to ensure the security of data transmission; (3f) Version management: The second sample set is annotated or modified again to generate different versions of the third sample set to meet different application requirements.

Images