Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards general deep-learning-based tree instance segmentation models

Jonathan Henrich, Jan van Delden

TL;DR

The paper tackles the problem of generalizing deep-learning-based tree instance segmentation across diverse forest point clouds and sensing modalities. It employs TreeLearn, a 3D-UNet-based framework with semantic and offset heads, trained and evaluated on seven diverse datasets after propagating labels to complete point clouds. Results indicate feasible generalization from conifer-dominated, high-resolution UAV data to deciduous-dominated, high-density TLS data, while generalization to very low-resolution data improves when UAV data is included in training. The work underscores the need for diverse, high-quality labeled forest data and provides public access to extended labeled forests to foster robust, domain-agnostic tree segmentation.

Abstract

The segmentation of individual trees from forest point clouds is a crucial task for downstream analyses such as carbon sequestration estimation. Recently, deep-learning-based methods have been proposed which show the potential of learning to segment trees. Since these methods are trained in a supervised way, the question arises how general models can be obtained that are applicable across a wide range of settings. So far, training has been mainly conducted with data from one specific laser scanning type and for specific types of forests. In this work, we train one segmentation model under various conditions, using seven diverse datasets found in literature, to gain insights into the generalization capabilities under domain-shift. Our results suggest that a generalization from coniferous dominated sparse point clouds to deciduous dominated high-resolution point clouds is possible. Conversely, qualitative evidence suggests that generalization from high-resolution to low-resolution point clouds is challenging. This emphasizes the need for forest point clouds with diverse data characteristics for model development. To enrich the available data basis, labeled trees from two previous works were propagated to the complete forest point cloud and are made publicly available at https://doi.org/10.25625/QUTUWU.

Towards general deep-learning-based tree instance segmentation models

TL;DR

The paper tackles the problem of generalizing deep-learning-based tree instance segmentation across diverse forest point clouds and sensing modalities. It employs TreeLearn, a 3D-UNet-based framework with semantic and offset heads, trained and evaluated on seven diverse datasets after propagating labels to complete point clouds. Results indicate feasible generalization from conifer-dominated, high-resolution UAV data to deciduous-dominated, high-density TLS data, while generalization to very low-resolution data improves when UAV data is included in training. The work underscores the need for diverse, high-quality labeled forest data and provides public access to extended labeled forests to foster robust, domain-agnostic tree segmentation.

Abstract

The segmentation of individual trees from forest point clouds is a crucial task for downstream analyses such as carbon sequestration estimation. Recently, deep-learning-based methods have been proposed which show the potential of learning to segment trees. Since these methods are trained in a supervised way, the question arises how general models can be obtained that are applicable across a wide range of settings. So far, training has been mainly conducted with data from one specific laser scanning type and for specific types of forests. In this work, we train one segmentation model under various conditions, using seven diverse datasets found in literature, to gain insights into the generalization capabilities under domain-shift. Our results suggest that a generalization from coniferous dominated sparse point clouds to deciduous dominated high-resolution point clouds is possible. Conversely, qualitative evidence suggests that generalization from high-resolution to low-resolution point clouds is challenging. This emphasizes the need for forest point clouds with diverse data characteristics for model development. To enrich the available data basis, labeled trees from two previous works were propagated to the complete forest point cloud and are made publicly available at https://doi.org/10.25625/QUTUWU.
Paper Structure (7 sections, 2 figures, 2 tables)

This paper contains 7 sections, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. Labeled forest data
  4. Segmentation Method
  5. Experiments
  6. Results and discussion
  7. Conclusion

Figures (2)

  • Figure 1: Fine-grained test results on an MLS point cloud (L1W). From left to right the images show (1) the ground truth segmentation and model results obtained by fine-tuning on (2) UAV, (3) MLS+TLS and (4) all data. Results are best when in-domain data is included during training.
  • Figure 2: Qualitative test results on a low-resolution UAV-scanned point cloud (RMIT). From left to right the images show (1) the ground truth segmentation and model results obtained by fine-tuning on (2) UAV, (3) MLS+TLS and (4) all data. When only MLS and TLS data is used, segmentation results have severe mistakes. When UAV data is included during training, results are substantially improved.