What question did this study set out to answer?

The aim is to enhance tree instance segmentation, semantic segmentation, and species classification while minimizing reliance on large annotated datasets.

July 23, 2026Open Access

Label-efficient 3D forest mapping: Self-supervised and transfer learning for instance segmentation, semantic segmentation, and species classification

ARAldino RizaldyKarlsruhe Institute of Technology FFFabian Ewald FassnachtGeographical Institute AAAhmed J. AfifiHelmholtz Institute Freiberg for Resource Technology

Key Points

The aim is to enhance tree instance segmentation, semantic segmentation, and species classification while minimizing reliance on large annotated datasets.
Utilized self-supervised and transfer learning approaches to train models effectively.
Evaluated performance across three tasks: instance segmentation, semantic segmentation, and tree classification.
Integrated various tasks into a streamlined framework for operational use.
Self-supervised learning combined with domain adaptation improved instance segmentation AP50 by 16.98%.
For semantic segmentation, self-supervised learning alone increased mIoU by 1.79%.
Hierarchical transfer learning enhanced tree classification mean Jaccard by 6.07%.

Abstract

Detailed structural and species information at the individual tree level is increasingly important to support precision forestry, biodiversity conservation, and provide reference data for biomass and carbon mapping. Point clouds from airborne and ground-based laser scanning are currently the most suitable data source to rapidly derive such information at scale. Recent advancements in deep learning have improved the segmentation and classification of individual trees and the identification of semantic tree components. However, deep learning models typically require large amounts of annotated training data which limits further improvement. Producing dense, high-quality annotations for 3D point clouds, especially in complex forests, is labor-intensive and challenging to scale. We explore strategies to reduce dependence on large annotated datasets using self-supervised and transfer learning architectures. Our objective is to improve performance across three tasks: instance segmentation, semantic segmentation, and tree classification using realistic and operational training sets. We observe improvements across all tasks, compared to training from scratch, evaluated with their respective metrics. For instance segmentation, self-supervised learning combined with domain adaptation improves AP50 by 16.98%. For semantic segmentation, self-supervised learning alone improves mIoU by 1.79%. For tree classification, hierarchical transfer learning improves mean Jaccard by 6.07%. To simplify use and encourage uptake, we integrated the tasks into a unified framework, streamlining the process from raw point clouds to tree delineation, structural analysis, and species classification. Pretrained models reduce energy consumption and carbon emissions by 21%. This open-source contribution aims to accelerate the operational extraction of individual tree information from laser scanning point clouds to support forestry, biodiversity, and carbon mapping.

Ask AI

Helpful

Bookmark

View Full Paper