Students who need it can contact them via private message. It is recommended to pay attention to the subscription account platform in the lower right corner of the above picture to download it.
Perform end-to-end target detection on images or videos collected by drones, fixed line monitoring, inspection robots in substations, fixed cameras in stations, etc., and use a combination of deep learning and traditional image processing technology to achieve intelligent understanding of inspection images, achieving Transmission line and substation equipment identification, defect detection, foreign object detection, etc. can effectively improve the safety of the power system. Therefore, the editor has compiled an ultra-complete collection of AI+ smart power detection data, including target detection data sets for 10+ subdivided scenarios in the power field, to facilitate the research and innovation of AI+ smart power detection.
1. Guangdong Power Grid Insulating Glove Wearing Detection Data Set
2. Guangdong power grid safety clothing wearing detection data set
3. Guangdong Power Grid’s high-altitude work safety belt wearing detection data set
4. China Southern Power Grid pole and tower foreign object detection data set
5. AI power line patrol helmet detection data set
6. Substation intelligent inspection data set
7. Transmission line tower bird’s nest detection data set
8. Transmission pole target detection data set
9. High-voltage line insulator defect detection data set
10. Transmission line fittings data set
01
—
Guangdong Power Grid Insulating Glove Wearing Detection Data Set
[Data background] Power grid operators need to perform power inspection and power outage operations on site every day. To ensure operational safety, power inspection personnel must wear rubber insulating gloves before performing power inspection and power outage operations. At the same time, there need to be guardians on site to supervise the action specifications of the electrical inspection personnel, and to call the police for rescue in time when unexpected situations occur. The specific labels and explanations included in the training data set provided by this scenario are as follows:
badge: guardianship armband (only red armbands are recognized)
person: all people present in the picture
glove: insulating gloves (rubber material)
wrongglove: Not wearing insulating gloves (other gloves or bare palms)
operatingbar: operating lever
powerchecker: electric test pen
Through the above information, the target detection algorithm is required to be able to automatically detect the following content in units of "people" during power grid field operations:
1.1 Identify all persons present and specifically distinguish the guardians (wearing red armbands);
1.2 Identify those who are present and wear insulating gloves (rubber material up to the forearm) in compliance with the regulations, as well as those who wear insulating gloves illegally (wearing gloves made of other materials, bare palms, semi-naked palms, exposed hand skin, etc.) are considered inappropriate. compliance) personnel;
1.3 Identify whether the on-site personnel have hand-held tools (operating poles or insulated pens);
There are several special situations that need attention:
Situation 1: Both hands can be observed, but only one glove is worn, which is illegal.
Situation 2: Only one hand can be seen from the two hands, and only one glove is worn, which is worn in compliance.
Situation 3: The gloves are not worn on the hands and the gloves are held in the hands, which is illegal.
[Algorithm Result] It is necessary to make predictions on the evaluation set and submit the evaluation results. Requirements are submitted as a single json file. Specific document requirements are as follows:
[{
"image_id": int,
"category_id": int,
"bbox": [xmin,ymin,xmax,ymax],
"score": float
}]
image_id is the serial number of the image in the test data in the csv, starting from 0, it is an int type number of 0, 1, 2, 3..., its maximum value is less than 599, if it exceeds, an error will be reported; because there are multiple copies of the same box In the case of identification results, there is no limit to the number of results submitted and can exceed 600.
category_id is the result category that needs to be detected. It is also an int. Category_id only submits one category at a time. If a box meets multiple recognition results, it needs to be written in blocks. The test results that need to be submitted are in the following categories, and the category_id of the image should remain the same as the comment below:
guarder
gloveperson (personnel who are required to wear insulating gloves)
wronggloveperson (person who does not wear insulating gloves in compliance with regulations)
operator (person with hand tools)
[Data files] The file list of the data set contains a total of 4 files, 1_testa_user.csv, 1train_rname.csv, 1_images.tar.gz and 1_test_images.tar.gz.
1_images.tar.gz is a training image set with a size of about 11.0GB;
1_test_images.tar.gz is a test image set with a size of about 2.5GB;
1train_rname.csv is the training set label;
1_testa_user.csv is the test set information.
02
—
Guangdong power grid safety clothing wearing detection data set
[Data background] According to the regulations of Guangdong Power Grid Company, workers need to wear work clothes every day to keep their mental appearance neat and unified, and also provide a safety barrier for workers operating outdoors. As for guardians, they need to wear an additional red armband on their work clothes. In the training data set provided in this scenario, the specific labels and explanations included are as follows:
badge: guardianship armband (only red armbands are recognized)
person: all people present in the picture
clothes: Compliant work clothes
wrongclothes (including the tags "wrongbottom", "wrongtop", and "wrongsuit"): Irregular work clothes (including open tops, rolled trouser legs, rolled sleeves, incomplete sets, etc.)
Through the above information, the target detection algorithm is required to be able to automatically detect the following content in units of "people" during power grid field operations:
2.1 Identify all persons present and specifically distinguish the guardians (wearing red armbands);
2.2 Identify workers who wear work clothes in compliance with regulations;
2.3 Identify workers who do not wear work clothes in compliance with regulations. Among them, the work clothes do not match the top and bottom, and the work clothes are worn with the placket open, trouser legs rolled up, sleeves rolled up, etc., which are all cases of irregular wearing of work clothes.
[Algorithm Result] It is necessary to make predictions on the evaluation set and submit the evaluation results. Requirements are submitted as a single json file. Specific document requirements are as follows:
[{
"image_id": int,
"category_id": int,
"bbox": [xmin,ymin,xmax,ymax],
"score": float
}]
image_id is the serial number of the image in the test data in the csv, starting from 0, it is an int type number of 0, 1, 2, 3..., its maximum value is less than 599, if it exceeds, an error will be reported; because there are multiple copies of the same box In the case of identification results, there is no limit to the number of results submitted and can exceed 600.
category_id is the result category that needs to be detected. It is also an int. Category_id only submits one category at a time. If a box meets multiple recognition results, it needs to be written in blocks. The test results that need to be submitted are in the following categories, and the category_id of the image should remain the same as the comment below:
guarder
rightdressed (personnel wearing work clothes in compliance with regulations)
wrongdressed (personnel who do not wear work clothes in compliance with regulations)
Note: All persons appearing (including power grid workers, guardians, onlookers, passers-by, etc.) need to be identified by wearing work clothes.
For example: A power grid worker wearing work clothes appears at the scene. The non-on-site workers are not supervisors. They still need to identify whether their work clothes are compliant. If a passerby (not wearing work clothes) appears, he needs to be judged as wrongdressed (personnel who does not wear work clothes in compliance with regulations).
[Data files] The file list of the data set contains a total of 4 files, 2_testa_user.csv, 2train_rname.csv, 2_images.tar.gz and 2_test_images.tar.gz.
1_images.tar.gz is a training image set with a size of about 12.8GB;
1_test_images.tar.gz is a test image set with a size of about 2.6GB;
1train_rname.csv is the training set label;
1_testa_user.csv is the test set information.
03
—
Guangdong Power Grid high-altitude work safety belt wearing detection data set
[Data background] In the daily work of Guangdong power grid operators, they often need to climb to high places on the power supply tower for inspection. In order to ensure the safety of workers, Guangdong Power Grid stipulates that people climbing off the ground must wear safety belts and have guardians on site to prevent accidents. In the training data set provided in this scenario, the specific labels and explanations included are as follows:
badge: guardianship armband (only red armbands are recognized)
Offground: A person who is off the ground
ground: A person who is on the ground
safebelt: wear a safety belt
Through the above information, the target detection algorithm is required to be able to automatically detect the following content in units of "people" during power grid field operations:
3.1 Identify all persons present and specifically distinguish the guardians (wearing red armbands);
3.2 Identify people wearing safety belts on site;
3.3 Identify persons who are off the ground;
[Algorithm Result] It is necessary to make predictions on the evaluation set and submit the evaluation results. Requirements are submitted as a single json file. Specific document requirements are as follows:
[{
"image_id": int,
"category_id": int,
"bbox": [xmin,ymin,xmax,ymax],
"score": float
}]
image_id is the serial number of the image in the test data in the csv, starting from 0, it is an int type number of 0, 1, 2, 3..., its maximum value is less than 599, if it exceeds, an error will be reported; because there are multiple copies of the same box In the case of identification results, there is no limit to the number of results submitted and can exceed 600.
category_id is the result category that needs to be detected. It is also an int. Category_id only submits one category at a time. If a box meets multiple recognition results, it needs to be written in blocks. The test results that need to be submitted are in the following categories, and the category_id of the image should remain the same as the comment below:
guarder
safebeltperson (person wearing a safety belt)
offlineperson (personnel off the ground)
Note: All appearing characters (including power grid workers, guardians, onlookers, passers-by, etc.) need to be identified with relevant labels.
For example: A power grid worker wearing work clothes appears at the scene. He is neither an on-site worker nor a supervisor. It is still necessary to identify whether he is wearing a safety belt. If a passerby appears without a matching tag description, the result is not submitted. If a passerby is off the ground, he or she can be identified as an off-ground person.
[Data files] The file list of the data set contains a total of 6 files, 3_images.tar.gz, 3train_rname.csv, 3_test_imagesa.tar.gz, 3_testa_user.csv, 3_testa_user.csv, 3_testB.zip, 3_testb_imageid.csv.
3_images.tar.gz is a training image set with a size of about 10.5GB;
3train_rname.csv labels the training set information;
3_test_imagesa.tar.gz is the A list test set, with a size of about 2.4GB;
3_testa_user.csv is the A list test set information;
3_testB.zip is the B list encrypted data set, with a size of about 8.2GB;
3_testb_imageid.csv is the encrypted data for list B.
04
—
China Southern Power Grid pole and tower foreign object detection data set
[Data background] Foreign objects in transmission lines are widely distributed and of many types. There may be bird's nests, kites, ropes and other debris on the towers, and kites, balloons, plastic films and other debris may also be hung on overhead lines. Therefore, there are certain problems in detection. difficulty. my country's transmission line inspection is basically based on manual inspection. However, the traditional manual inspection method is inefficient, has many restrictions, and often consumes a lot of manpower and material resources. Later, the method of helicopter inspection along the line was introduced, but this method has flight operations. It is very dangerous and extremely expensive to train and maintain, and cannot alleviate the problems caused by manual inspection to a great extent. In recent years, GPU computing capabilities have continued to improve, and more and more researchers in the electric power field have chosen to combine machine vision technology with deep learning algorithms to conduct research and application of target detection algorithm technology.
[Application fields] AI+ power target detection
[File Directory] Three folders: South China Open preliminary train, South China Open preliminary val and South China Open semi-finals data set.zip. XML annotation files and jpg image files are saved in each folder.
[Data description] The data size of the South China Open preliminary round train is about 2.8GB (with xml annotation), the data size of the South China Open preliminary round val is about 1.1GB (without xml annotation), and the data size of the South China Open semi-finals data set is about 1.7GB (with xml annotation). xml annotation), train has about 1300+ image samples, val has about 300+ image samples, and foreign object types include 4 categories: bird's nest, kite, balloon, and garbage.
05
—
AI power line patrol helmet detection data set
[Data background] At present, most of the hard hat detection is for construction sites and large machinery. There are fewer hard hat detection data sets in the power industry. This time we selected a hard hat detection data set. Most of the pictures are from the electric power industry, including pole installation, Scenarios such as power overhead construction and power emergency repair; a small part comes from other industries.
[Application fields] AI+ power target detection
[File Directory] Two folders: Annotations and JPEGImages
[Data Description] VOC format data, the Annotations directory is the xml annotation file, the JPEGImages directory is the jpg image file, the data size is about 127.9M
06
—
Substation intelligent inspection data set
[Data background] Traditional substation inspection methods usually require a lot of manpower and time investment, and often can only conduct regular inspections, and cannot monitor equipment status in real time. The intelligent inspection system can realize full-time and all-round monitoring of equipment, reduce the consumption of human resources, and can optimize maintenance plans and reduce operating costs through data analysis and prediction.
[Application fields] AI+ power target detection
[File Directory] Two folders: Annotations and JPEGImages
[Data Description] The Annotations directory contains xml annotation files, and the JPEGImages directory contains jpg image files. The data set contains 8376 inspection images with xml annotations. The data size is approximately 9.6GB, and contains a total of 17 types of inspection labels.
07
—
Transmission line tower bird's nest detection data set
[Data background] Bird nests often appear in locations such as telegraph poles and towers in the power industry. Failure to remove them in time will cause safety hazards and power security risks. As a guarantee, it is necessary to identify whether there are bird nests in designated areas.
[Application fields] AI+ power target detection
[File Directory] Two folders: Annotations and JPEGImages
[Data description] The Annotations directory contains xml annotation files, and the JPEGImages directory contains jpg image files. The data set contains 200 images of power lines or towers with bird's nest targets, with xml annotations, and the data size is approximately 0.64GB.
08
—
Transmission pole target detection data set
[Data background] High-voltage transmission line poles and towers are an important part of the national power grid construction, and their safety is related to power security and the national economy and people's livelihood. At present, the high-voltage tower inspection work method is traditional and cannot monitor and respond to accidents such as high-voltage tower base tilt and landslides caused by sudden geological disasters in real time. Therefore, the fully automatic target detection method based on AI can monitor high-voltage poles and towers in real time and is an effective way to ensure the safe operation of transmission lines.
[Application fields] AI+ power target detection
[File directory] Annotations and JPEGImage two folders
[Data Description] The Annotations directory contains xml annotation files, and the JPEGImage directory contains jpg image files. The data set contains 400 transmission line tower images, and the data size is approximately 0.14GB.
09
—
High-voltage line insulator defect detection data set
[Data background] With the proposal of the national west-to-east power transmission strategy, ultra-high voltage power transmission and high-voltage power transmission have become particularly important. Inspection of transmission lines is an important part of ensuring normal power transmission. With the rapid development of smart power grids, inspection drones and robots are gradually replacing traditional manual inspections. Insulators are an important part of transmission lines. They are the only electrical insulators and important structural supports. The performance of insulators and the rationality of their configuration directly affect the safe and stable operation of the lines. Part of this data set comes from real drone shooting, and part comes from picture synthesis.
[Application fields] AI+ power target detection
[File Directory] Two folders: Annotations and JPEGImages
[Data description] The Annotations directory contains xml annotation files, and the JPEGImages directory contains jpg image files. The data set includes 600 pictures of defective insulators in high-voltage transmission lines. It adopts the VOC annotation format and can be divided into training sets and verification sets by itself.
10
—
Conclusion
The above is all the contents of the data sets related to transmission lines and substations. For more data set downloads, please pay attention to the platform in the lower right corner of the article picture to obtain it.