PCB-Defect: An Annotated Dataset for Surface Defect Detection in Printed Circuit Boards

Published: 6 October 2026| Version 2 | DOI: 10.17632/vdj74sngvn.2
Contributors:
,
,
,
,
,

Description

The PCB Defect Detection dataset comprises 230 images of uniquely manufactured single-layer printed circuit boards (PCBs) with authentic, chemically induced defects. Each PCB was fabricated using a controlled chemical etching process, ensuring that the defects closely replicate those encountered in industrial production environments. The dataset captures six key defect types: missing pad, mouse bite, open circuit, short circuit, spur, and spurious copper. Many of the defects are elongated and lie at arbitrary angles, and their orientation cannot be recovered from horizontal bounding boxes, so every image has been re-annotated with oriented annotations. Each oriented annotation follows the orientation of a visible defect instance, resulting in a total of 2,506 defect annotations across the six defect categories and an average of 10.9 annotations per image, with no missing annotations and no null examples. An axis-aligned bounding box can be derived from each oriented annotation by taking the extent of its corners. The annotations therefore support both horizontal (HBB) and oriented (OBB) object detection, and detectors of both types can be trained and evaluated against the same defect instances. The image resolutions range from 2.33 to 29.97 megapixels, with an average of 6.62 megapixels and a median image size of 2639×2538 pixels. The original axis-aligned annotations in COCO-style JSON format are also included, and an accompanying readme file describes both annotation sets.

Files

Institutions

Categories

Computer Vision, Circuit Systems, Interconnection (Integrated Circuit), Object Detection, Deep Learning, Fault Diagnosis

Licence