Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

UruDendro4: A Benchmark Dataset for Automatic Tree-Ring Detection in Cross-Section Images of Pinus taeda L

Henry Marichal, Joaquin Blanco, Diego Passarella, Gregory Randall

TL;DR

UruDendro4 addresses the scarcity of annotated cross-section tree-ring data by introducing 102 Pinus taeda samples with multi-height rings for volumetric growth analysis. The paper benchmarks three baselines—CS-TRD, DeepCS-TRD, and INBD—on this dataset, showing DeepCS-TRD with a frozen encoder and tuned alpha achieves the best performance (mAP around 0.838, mAR around 0.782, ARAND around 0.084). It also demonstrates that training with UruDendro4 improves generalization, including cross-domain gains when combining datasets and modest cross-species transfer to Douglas fir. The dataset, along with comprehensive ablations and metadata, provides a valuable resource for dendrochronology and forest-management research with practical implications for volumetric wood estimation.

Abstract

Tree-ring growth represents the annual wood increment for a tree, and quantifying it allows researchers to assess which silvicultural practices are best suited for each species. Manual measurement of this growth is time-consuming and often imprecise, as it is typically performed along 4 to 8 radial directions on a cross-sectional disc. In recent years, automated algorithms and datasets have emerged to enhance accuracy and automate the delineation of annual rings in cross-sectional images. To address the scarcity of wood cross-section data, we introduce the UruDendro4 dataset, a collection of 102 image samples of Pinus taeda L., each manually annotated with annual growth rings. Unlike existing public datasets, UruDendro4 includes samples extracted at multiple heights along the stem, allowing for the volumetric modeling of annual growth using manually delineated rings. This dataset (images and annotations) allows the development of volumetric models for annual wood estimation based on cross-sectional imagery. Additionally, we provide a performance baseline for automatic ring detection on this dataset using state-of-the-art methods. The highest performance was achieved by the DeepCS-TRD method, with a mean Average Precision of 0.838, a mean Average Recall of 0.782, and an Adapted Rand Error score of 0.084. A series of ablation experiments were conducted to empirically validate the final parameter configuration. Furthermore, we empirically demonstrate that training a learning model including this dataset improves the model's generalization in the tree-ring detection task.

UruDendro4: A Benchmark Dataset for Automatic Tree-Ring Detection in Cross-Section Images of Pinus taeda L

TL;DR

UruDendro4 addresses the scarcity of annotated cross-section tree-ring data by introducing 102 Pinus taeda samples with multi-height rings for volumetric growth analysis. The paper benchmarks three baselines—CS-TRD, DeepCS-TRD, and INBD—on this dataset, showing DeepCS-TRD with a frozen encoder and tuned alpha achieves the best performance (mAP around 0.838, mAR around 0.782, ARAND around 0.084). It also demonstrates that training with UruDendro4 improves generalization, including cross-domain gains when combining datasets and modest cross-species transfer to Douglas fir. The dataset, along with comprehensive ablations and metadata, provides a valuable resource for dendrochronology and forest-management research with practical implications for volumetric wood estimation.

Abstract

Tree-ring growth represents the annual wood increment for a tree, and quantifying it allows researchers to assess which silvicultural practices are best suited for each species. Manual measurement of this growth is time-consuming and often imprecise, as it is typically performed along 4 to 8 radial directions on a cross-sectional disc. In recent years, automated algorithms and datasets have emerged to enhance accuracy and automate the delineation of annual rings in cross-sectional images. To address the scarcity of wood cross-section data, we introduce the UruDendro4 dataset, a collection of 102 image samples of Pinus taeda L., each manually annotated with annual growth rings. Unlike existing public datasets, UruDendro4 includes samples extracted at multiple heights along the stem, allowing for the volumetric modeling of annual growth using manually delineated rings. This dataset (images and annotations) allows the development of volumetric models for annual wood estimation based on cross-sectional imagery. Additionally, we provide a performance baseline for automatic ring detection on this dataset using state-of-the-art methods. The highest performance was achieved by the DeepCS-TRD method, with a mean Average Precision of 0.838, a mean Average Recall of 0.782, and an Adapted Rand Error score of 0.084. A series of ablation experiments were conducted to empirically validate the final parameter configuration. Furthermore, we empirically demonstrate that training a learning model including this dataset improves the model's generalization in the tree-ring detection task.

Paper Structure

This paper contains 6 sections, 1 equation, 4 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. The UruDendro4 Dataset
  4. Methodology
  5. Experimental results
  6. Conclusions

Figures (4)

  • Figure 1: Cross-section samples of Pinus taeda L. (T0_B3_N23_D and T6_B3_N21_D from UruDendro4, respectively) and their tree-ring curve delineations. The fifth growth ring area is highlighted in red in both samples.
  • Figure 2: Example images from the UruDendro4 dataset.
  • Figure 3: Automatic tree-ring detection results for sample T6_B3_N21_D (see \ref{['fig:main']}). The CS-TRD method achieved a mAP of 0.256, a mAR of 0.241, and an ARAND score of 0.487. INBD obtained 0.306 (mAP), 0.288 (mAR), and 0.302 (ARAND). DeepCS-TRD outperformed both methods with scores of 0.688 (mAP), 0.688 (mAR), and 0.245 (ARAND), respectively.
  • Figure 4: Automatic tree-ring detection results for sample T0_B1_N32_ADAP (see \ref{['fig:urudendro4']}). The CS-TRD method achieved a mAP of 0.588, a mAR of 0.470, and an ARAND score of 0.200. INBD obtained 0.677 (mAP), 0.610 (mAR), and 0.135 (ARAND). DeepCS-TRD outperformed both methods with scores of 0.800 (mAP), 0.680 (mAR), and 0.118 (ARAND), respectively.