# PCB-defect: An annotated dataset for surface defect detection in printed circuit boards

## TL;DR

The PCB-Defect dataset provides annotated high-resolution images of printed circuit boards with various manufacturing defects to improve automated defect detection systems.

## Contribution

The dataset introduces a curated collection of annotated PCB images with diverse defect types and image qualities for robust computer vision model development.

## Key findings

- The dataset includes 230 annotated images with 1704 defect instances across six defect categories.
- Images range from 800 × 600 to 6000 × 4000 pixels, averaging 6.61 megapixels, with annotations in COCO JSON format.
- The dataset supports applications in automated optical inspection, quality control, and transfer learning.

## Abstract

The PCB-Defect dataset was developed to advance automated defect detection in Printed Circuit Boards. This dataset presents a comprehensive collection of 230 annotated high-resolution images of single-layer PCBs, manufactured through a controlled laboratory process. The curated diversity in PCB layouts, defect types, image quality, and annotation density supports the development of robust computer vision models for defect detection. Each board was fabricated using chemical etching on FR4 substrates to introduce manufacturing defects, including missing pad, mouse bite, open circuit, short circuit, spur, and spurious copper. The defect types were embedded at the design phase, physically realized during etching, and confirmed post-manufacturing. Images range from 800 × 600 to 6000 × 4000 pixels, averaging 6.61 megapixels. Detailed bounding box annotations for all visible defect instances were produced using the Roboflow tool, resulting in a total of 1704 annotated defects across the dataset with about 7.4 annotation per image. Annotations follow the COCO (Common Objects in Context) JSON format, which includes detailed metadata and precise localization of each defect by category and bounding box coordinates. This dataset offers significant reuse potential for both academic and industrial research communities focusing on automated optical inspection, quality control, and transfer learning applications.