Wood image dataset links trees to improve species identification accuracy
WoodBR-Cross: A public wood-image dataset and individual-tree-disjoint benchmark for species identification
Computer Vision and Pattern Recognition
Summary
Identifying tree species from wood images can be tricky when photos of the same tree appear in both training and testing sets. The authors created WoodBR-Cross, a collection of wood surface images linked to individual trees, ensuring that all images from a single tree stay grouped. This approach helps better estimate how well computer models can identify species from unseen trees. They tested different machine learning models and found strong performance when models never saw the same tree in training and testing.
What this means in practice
- •For forestry monitoring teams: Use tree-linked wood image data to train models that better identify species from new wood samples without overlap from known trees.
- •For environmental compliance teams: Implement wood species classifiers trained on tree-disjoint data sets to improve detection of illegal logging from wood samples.$Commercial implications: This dataset enables developing commercial wood species verification tools to support law enforcement and sustainable timber trade.
Authors
Daniel Casanova, Alexandre Bahia Gontijo, Arnaldo Candido Junior, Silvana Nisgoski, Pedro Luiz de Paula Filho
Abstract
Wood-image benchmarks can yield optimistic estimates when images from the same tree occur in both training and test sets. WoodBR-Cross contains 4,875 transverse-surface images in 38 operational taxonomic classes, linked to 303 distinct trees identified by numeric catalogue witness codes (codigos-testemunho) in the Brazilian Forest Service collection. A species-level subset comprises 3,406 images, 20 classes and 214 trees. Three tasks compare duplicate-aware image partitions for all classes (A), image partitions for the species subset (B1), and individual-tree-disjoint partitions for that subset (B2). In B2, all images from each tree are kept together, so no tree is shared across training, validation or test within a fold. ResNet-50, ConvNeXt-Tiny and ViT-B/16 were evaluated with three outer folds, internal validation and five training seeds. Handcrafted and frozen-feature baselines complement border-only and randomized-label controls. Validation-selected ConvNeXt-Tiny achieved B2 tree-level macro-F1 of 0.8621 +/- 0.0317; frozen DINOv2 features yielded 0.9000 +/- 0.0722. Values are means and standard deviations across three seed-averaged fold means. Results estimate performance on previously unseen trees from this collection.