DefectDet: An Annotated Dataset of Printed Circuit Boards with Surface Defects

Published: 15 February 2026| Version 1 | DOI: 10.17632/t9d9zs3bmb.1
Contributors:
,
,
,
,

Description

The following dataset provides a collection of 268 annotated high-resolution images of manufactured double-layer PCBs, produced via standardised manufacturing protocols. Surface defects were intentionally introduced during the design phase through precise vector-level modifications of SVG and Gerber files and were physically realised during manufacturing. Five common surface defect categories are included: short circuit, open circuit, spurious copper, spur, and missing pad. Image resolutions range from approximately 1000 × 1000 to 2400 × 2400 pixels, providing full-board coverage for each PCB. Detailed bounding box annotations were generated for all visible defect instances using the Roboflow tool, resulting in approximately 1300 annotations with an average of 5 defect instances per image. All annotations follow the COCO (Common Objects in Context) JSON format.

Files

Institutions

Categories

Computer Vision, Integrated Circuit, Object Detection, Deep Learning, Fault Diagnosis

Licence