A Public Image Dataset for Surface Defect Detection of Water-Based Coated Wood Products
Abstract
We developed and released an image dataset for surface defect detection on waterborne painted wood products. The dataset comprises 13400 high-resolution images capturing four defect types: scratches, cracks, bubbles, and holes. These include 3645 bubble defects, 3498 scratch defects, 3256 crack defects, and 3001 hole defects. Data were acquired in an operational production facility in Jiangshan, China, using specialized industrial cameras. Professional annotators performed rigorous labeling to ensure accuracy. This dataset provides critical data support for deploying deep learning models in real-world industrial assembly lines. Researchers may leverage this dataset to develop automated machine learning solutions for multi-class defect detection. Such techniques enable timely defect detection and remediation, ensuring finish integrity and final product quality.
Fields of Study & Subjects
Industrial Vision Systems and Defect Detection
Advanced Neural Network Applications
Wood Treatment and Properties
Publication details
- Journal / source
- Scientific Data
- Publisher
- Springer Science and Business Media LLC
- Volume
- 13
- Issue
- 1
- Pages
- 128–128
- Publication date
- 14 January 2026
- Language
- EN