Annotated Images of Substation Equipment with 15 Object Classes

A single electrical distribution substation in Brazil provided 1,660 images collected over two years. The dataset contains 50,705 annotated objects across 15 classes of equipment, such as disconnect switches and insulators. Images were curated by electrical engineering experts to support automated inspection tasks.

Use Cases

• Training semantic segmentation models based on the provided PNG segmentation masks for 15 equipment classes.
• Developing object detection algorithms for substation equipment using the VOC-style polygonal JSON annotations.
• Evaluating model robustness under varied lighting and weather conditions based on images captured at different times and seasons.
• Automating visual inspection of electrical components like disconnect switches and transformers based on the expert-curated image angles and distances.

Strengths

• Includes 50,705 annotated objects across 15 distinct equipment classes, with instance counts provided for each class.
• Offers multiple annotation formats: VOC-style JSON polygon files and PNG segmentation masks.
• Images were captured under diverse conditions over two years, including different times of day and weather, providing variation.
• A team of electrical engineering experts curated the images to ensure suitability for automated inspection tasks.

Limitations

• All images originate from a single substation in one country, which may limit geographic generalizability.
• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

Andreas Gomes via paperswithcode.

Collection Method

Images were obtained using various cameras, including those mounted on Autonomous Guided Vehicles (AGVs), fixed cameras, and handheld devices.

Time Range

Collected over a period of two years (specific dates unknown).

Geography

A single electrical distribution substation in Brazil.