VNWoodKnot: A benchmark image dataset for wood knot detection and classification

Abstract

Timber knot detection is essential for automated grading and quality control in the wood processing industry. Knots, which arise at the intersection of branches and the tree trunk, are among the most influential defects affecting both structural integrity and aesthetics. This paper introduces VNWoodKnot, a publicly available image dataset comprising 1,515 high-resolution wood surface images, collected in a Vietnamese industrial facility. The dataset includes three categories: live knots (519 images), dead knots (496 images), and knot-free surfaces (500 images). Live knots are structurally integrated and color-consistent, while dead knots are darker, cracked, and loosely attached. VNWoodKnot enables both classification and object detection tasks and addresses a critical gap in publicly accessible datasets for AI-driven wood defect inspection. It serves as a crucial benchmark for the development of real-time, scalable, and reliable deep learning models for industrial-grade wood defect inspection.