Label-Efficient 3D Forest Mapping: Self-Supervised and Transfer Learning for Individual, Structural, and Species Analysis Journal: Remote Sensing of Environment
Aldino Rizaldy, Fabian Ewald Fassnacht, Ahmed Jamal Afifi, Hua Jiang, Richard Gloaguen, Pedram Ghamisi · 2025
Read Label-Efficient 3D Forest Mapping: Self-Supervised and Transfer Learning for Individual, Structural, and Species Analysis Journal: Remote Sensing of Environment by Aldino Rizaldy, Fabian Ewald Fassnacht, Ahmed Jamal Afifi, Hua Jiang, Richard Gloaguen, Pedram Ghamisi (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-06331.txt.
Description
Aldino Rizaldy and co-authors map individual trees in three dimensions while needing far fewer hand labels. Forest managers want per-tree structure and species data for precision forestry, biodiversity work, and carbon mapping. Laser scanning point clouds supply the raw detail, but labeling enough of them to train deep models is slow and expensive.
Self-supervised pretraining and transfer learning cut that labeling burden across three tasks: separating single trees, labeling tree parts, and naming species. Gains appeared on every task, with the largest jump in tree separation when self-supervision joined domain adaptation. A unified open framework runs the full pipeline, and pretrained models trimmed energy use and emissions by about a fifth.
Catalog record
| CREATOR | Aldino Rizaldy, Fabian Ewald Fassnacht, Ahmed Jamal Afifi, Hua Jiang, Richard Gloaguen, Pedram Ghamisi |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 65 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-06331