UruDendro, a public dataset of cross-section images of Pinus taeda

• DOI: 10.48550/arxiv.2404.10856
• Publication Date: 2024-04-16
• Authors: Henry Marichal; Diego Passarella; Christine Lucas; Ludmila Profumo; Verónica Casaravilla; M. Galli; Serrana Ambite; Grégory Randall
• Keywords: Section (typography)

The automatic detection of tree-ring boundaries and other anatomical features using image analysis has progressed substantially over the past decade with advances in machine learning imagery technology, as well increasing demands from dendrochronology community. This paper presents a publicly available database 64 scanned images transverse sections commercially grown Pinus taeda trees northern Uruguay, ranging 17 to 24 years old. collection contains several challenging for ring detection, including illumination surface preparation variation, fungal infection (blue stains), knot formation, missing cortex or interruptions outer rings, radial cracking. dataset can be used develop test tree algorithms. community one such method, Cross-Section Tree-Ring Detection (CS-TRD), which identifies marks complete annual rings cross-sections species presenting clear definition between early latewood. We compare CS-TRD performance against ground truth manual delineation all UruDendro dataset. software identified an average F-score 89% RMSE error 5.27px entire less than 20 seconds per image. Finally, we propose robust measure growth \emph{equivalent radius} circle having same area enclosed by detected ring. Overall, this study contributes dendrochronologist's toolbox fast low-cost methods automatically detect conifer species, particularly measuring diameter rates stem cross-sections.