Images of defects on an electrical conductor
Description
A set of high-resolution images showing conductor anomalies that are difficult to detect in the field. These images cover defects of varying degrees of severity. We are publishing it to support research and accelerate innovation in visual anomaly detection (VAD) for robotic power line inspection.
Read the full description of the project with Mila – Institut québécois d’intelligence artificielle.
Method
The images were taken using a video camera in a laboratory setting. Three typical cables of our transmission lines were altered by our specialists to reproduce various types and sizes of defects. A mechanism controlled the lighting as well as the position and orientation of a capture camera, while allowing it to move along the cable. The video files were converted into image sequences, of which only a fraction of the frames were retained, in order to obtain a relatively uniform sequence of movement along the cable. The data was then imported into the Computer Vision Annotation Tool (CVAT), enabling a team of specialists to create and validate the labels used to identify defects. Lastly, the result was exported as a JSON file in COCO format. Together with the source images, this file can be distributed to the community of artificial intelligence (AI) specialists and archived in a shared project folder.
Useful information for interpreting data
The data and images do not represent all possible anomalies. There is no guarantee regarding their quality, integrity, accuracy or completeness and they are subject to change without notice. The anomalies and images have been produced in a laboratory for research purposes.
Description of labels
The labels and their descriptions are provided below.
- category_id [int]: Number associated with the defect category, as defined below;
- LastExecuted: Date and time of the last available measurement (in UTC);
- date: Date and time the measurement was taken;
- grading [str]: Grading of the defect, the value of which depends on the defect category, as mentioned above;
- identification [float]: Number assigned to the last artificially created defect;
- secondary [float]: Value of 0 if the defect has been artificially created, or incremental value in sequence for any subsequent defect, up to the next defect defined as artificial;
- occluded [bool]: Indicates whether the annotation is hidden or not visible in the image;
- rotation [float]: Rotation angle, in degrees, to be applied to the coordinates of the bbox annotation (bounding box);
- track_id [int]: Tracking algorithm identification number. A defect will have the same value on all images in a sequence, while a different number will be associated with another sequence showing the same defect;
- keyframe [bool]: Data indicating whether the annotation dimensions come from the person responsible for the annotation (true) or whether they have been interpolated by the tracking algorithm (false).
The seven defect categorization values [category_id] that we define are also classified according to a grading level specific to each category as follows:
[1-Melted strands]
- Superficial: The strands are still distinct.
- Partial: The strands are fused. The lower layer is visible, but the electrical continuity of the strand remains, even if it is deformed.
- Deep: A strand is severed by melting. The strands of the lower layer are generally visible.
[2-Deposits]
- Slight: The cable is stained, but the individual strands are still discernible.
- Major: The cable is stained, and it is difficult, if not impossible, to perceive the strands.
[3-Broken strands]
- Partial: The strand is damaged, but not completely severed.
- Complete: The strand is completely severed, but still in place.
- Protruding: The strand is severed and has moved. It is sticking out of the conductor.
[4-Spaced strands]
- Slight: Presence of slight spreading, without being able to make out the strands of the lower layer.
- Major : The strands of the lower layer are discernible.
[5-Crushing]
- Slight: Presence of crushing less than the width of a strand.
- Major: Presence of crushing larger than the width of a strand.
[6-Long scratch] (scratches along the electrical conductor)
- Slight: Presence of a scratch whose width is less than one third (1/3) of a strand.
- Major : Presence of a scratch wider than one third (1/3) of a strand.
[7-Twisted strands]
- Slight: Presence of deformation less than the width of a strand.
- Major: Presence of deformation greater than the width of a strand.
Additional information
- Publisher or author: Hydro‑Québec
- Language: French
- Geographical coverage: The images were taken in the laboratory.
- Temporal coverage: The images were taken in the laboratory on April 19, 2023.
- Update frequency: Unscheduled.
- Initial distribution: 2024‑12‑06
- Licence or user rights: Creative Commons CC‑BY‑NC 4.0
- Notices and conditions of use: The data and images do not represent all possible anomalies. There is no guarantee regarding their quality, integrity, accuracy or completeness and they are subject to change without notice. The anomalies and images have been produced in a laboratory for research purposes.