Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Automated forest inventory: analysis of high-density airborne LiDAR point clouds with 3D deep learning

Binbin Xiang, Maciej Wielgosz, Theodora Kontogianni, Torben Peters, Stefano Puliti, Rasmus Astrup, Konrad Schindler

TL;DR

This work introduces ForAINet, a 3D deep learning framework for automated forest inventories from high-density LiDAR data. It jointly performs semantic segmentation, instance segmentation, and tree-component labeling, then uses geometric methods and a TreeMix-based data-augmentation pipeline to extract per-tree and stand-level biophysical attributes. On the FOR-Instance dataset, the method achieves an $F$-score of $85.1\%$ for individual-tree segmentation and $73.5\%$ mean $IoU$ across five semantic classes, outperforming several baselines, though performance is lower for understory trees and at lower point densities. The approach demonstrates strong potential for scalable, tree-level forest inventories and informs practical deployment considerations, while also highlighting limitations related to understory annotation and data density.

Abstract

Detailed forest inventories are critical for sustainable and flexible management of forest resources, to conserve various ecosystem services. Modern airborne laser scanners deliver high-density point clouds with great potential for fine-scale forest inventory and analysis, but automatically partitioning those point clouds into meaningful entities like individual trees or tree components remains a challenge. The present study aims to fill this gap and introduces a deep learning framework, termed ForAINet, that is able to perform such a segmentation across diverse forest types and geographic regions. From the segmented data, we then derive relevant biophysical parameters of individual trees as well as stands. The system has been tested on FOR-Instance, a dataset of point clouds that have been acquired in five different countries using surveying drones. The segmentation back-end achieves over 85% F-score for individual trees, respectively over 73% mean IoU across five semantic categories: ground, low vegetation, stems, live branches and dead branches. Building on the segmentation results our pipeline then densely calculates biophysical features of each individual tree (height, crown diameter, crown volume, DBH, and location) and properties per stand (digital terrain model and stand density). Especially crown-related features are in most cases retrieved with high accuracy, whereas the estimates for DBH and location are less reliable, due to the airborne scanning setup.

Automated forest inventory: analysis of high-density airborne LiDAR point clouds with 3D deep learning

TL;DR

This work introduces ForAINet, a 3D deep learning framework for automated forest inventories from high-density LiDAR data. It jointly performs semantic segmentation, instance segmentation, and tree-component labeling, then uses geometric methods and a TreeMix-based data-augmentation pipeline to extract per-tree and stand-level biophysical attributes. On the FOR-Instance dataset, the method achieves an -score of for individual-tree segmentation and mean across five semantic classes, outperforming several baselines, though performance is lower for understory trees and at lower point densities. The approach demonstrates strong potential for scalable, tree-level forest inventories and informs practical deployment considerations, while also highlighting limitations related to understory annotation and data density.

Abstract

Detailed forest inventories are critical for sustainable and flexible management of forest resources, to conserve various ecosystem services. Modern airborne laser scanners deliver high-density point clouds with great potential for fine-scale forest inventory and analysis, but automatically partitioning those point clouds into meaningful entities like individual trees or tree components remains a challenge. The present study aims to fill this gap and introduces a deep learning framework, termed ForAINet, that is able to perform such a segmentation across diverse forest types and geographic regions. From the segmented data, we then derive relevant biophysical parameters of individual trees as well as stands. The system has been tested on FOR-Instance, a dataset of point clouds that have been acquired in five different countries using surveying drones. The segmentation back-end achieves over 85% F-score for individual trees, respectively over 73% mean IoU across five semantic categories: ground, low vegetation, stems, live branches and dead branches. Building on the segmentation results our pipeline then densely calculates biophysical features of each individual tree (height, crown diameter, crown volume, DBH, and location) and properties per stand (digital terrain model and stand density). Especially crown-related features are in most cases retrieved with high accuracy, whereas the estimates for DBH and location are less reliable, due to the airborne scanning setup.
Paper Structure (21 sections, 2 equations, 10 figures, 12 tables)

This paper contains 21 sections, 2 equations, 10 figures, 12 tables.

Table of Contents

  1. Introduction
  2. Materials and methods
  3. Dataset
  4. Deep learning framework for multiple segmentation tasks
  5. authorcolorChanges@AuthorColor Data augmentation and balancing strategies Data augmentation and balancing strategies Data augmentation and balancing strategies Data augmentation and balancing strategies Data augmentation and balancing strategies Data augmentation and balancing strategies Data augmentation and balancing strategies Data augmentation and balancing strategies Data augmentation and balancing strategies Data augmentation and balancing strategies Input data Input data Input data Input data Input data Input data Input data Input data Input data Input data Input data Input data Data augmentation and balancing strategies Data augmentation and balancing strategies Data augmentation and balancing strategies Data augmentation and balancing strategies Data augmentation and balancing strategies Data augmentation and balancing strategies Data augmentation and balancing strategies Data augmentation and balancing strategies Data augmentation and balancing strategies Data augmentation and balancing strategies Input data Input data Input data Input data Input data Input data Input data Input data Input data Input data Input data Input data Data augmentation and balancing strategies Data augmentation and balancing strategies Data augmentation and balancing strategies
  6. Network architecture
  7. Automated retrieval of tree parameters and stand structure
  8. Evaluation metrics
  9. Metrics for point cloud segmentation
  10. Metrics for individual tree features and stand structure
  11. Evaluation for understory and suppressed trees
  12. Implementation details
  13. Results
  14. 3D point cloud segmentation
  15. Tree features and stand attributes
  16. ...and 6 more sections

Figures (10)

  • Figure 1: Illustration of our segmentation and retrieval framework. authorcolorChanges@AuthorColor It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}. It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}.It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}. It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}. It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}. It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}. It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}.It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}.It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}.It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}. It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}. It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}.It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}. It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}. It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}. It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}. It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}.It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}.It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}.It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}. It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}. It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}. It operates in two steps: The first step segments points into semantic categories as well as individual trees, for details see Figure \ref{['fig:pipeline']} and Section \ref{['Sec:DeepLearningBasedFrameworkForMultipleSegmentationTasks']}. The second step retrieves tree parameters and stand structure from the segmentation results, see Section \ref{['Sec:AutomatedQuantificationofImportantTreesFeaturesandStandStructure']}.
  • Figure 2: authorcolorChanges@AuthorColor Illustration of the point cloud segmentation pipeline. Illustration of the point cloud segmentation pipeline.Illustration of the point cloud segmentation pipeline.Illustration of the point cloud segmentation pipeline. Illustration of the point cloud segmentation pipeline. Illustration of the point cloud segmentation pipeline. Illustration of the point cloud segmentation pipeline.Illustration of the point cloud segmentation pipeline.Illustration of the point cloud segmentation pipeline.Illustration of the point cloud segmentation pipeline. Illustration of the point cloud segmentation pipeline. Illustration of the point cloud segmentation pipeline.Illustration of the point cloud segmentation pipeline.Illustration of the point cloud segmentation pipeline. Illustration of the point cloud segmentation pipeline. Illustration of the point cloud segmentation pipeline. Illustration of the point cloud segmentation pipeline.Illustration of the point cloud segmentation pipeline.Illustration of the point cloud segmentation pipeline.Illustration of the point cloud segmentation pipeline. Illustration of the point cloud segmentation pipeline. Illustration of the point cloud segmentation pipeline.Illustration of the point cloud segmentation pipeline.
  • Figure 3: Illustration of pipeline of TreeMix method for data augmentation.
  • Figure 4: authorcolorChanges@AuthorColor Confusion matrix between the five semantic categories, computed over all points of the test set. Confusion matrix between the five semantic categories, computed over all points of the test set.Confusion matrix between the five semantic categories, computed over all points of the test set.Confusion matrix between the five semantic categories, computed over all points of the test set. Confusion matrix between the five semantic categories, computed over all points of the test set. Confusion matrix between the five semantic categories, computed over all points of the test set. Confusion matrix between the five semantic categories, computed over all points of the test set.Confusion matrix between the five semantic categories, computed over all points of the test set.Confusion matrix between the five semantic categories, computed over all points of the test set.Confusion matrix between the five semantic categories, computed over all points of the test set. Confusion matrix between the five semantic categories, computed over all points of the test set. Confusion matrix between the five semantic categories, computed over all points of the test set.Confusion matrix between the five semantic categories, computed over all points of the test set.Confusion matrix between the five semantic categories, computed over all points of the test set. Confusion matrix between the five semantic categories, computed over all points of the test set. Confusion matrix between the five semantic categories, computed over all points of the test set. Confusion matrix between the five semantic categories, computed over all points of the test set.Confusion matrix between the five semantic categories, computed over all points of the test set.Confusion matrix between the five semantic categories, computed over all points of the test set.Confusion matrix between the five semantic categories, computed over all points of the test set. Confusion matrix between the five semantic categories, computed over all points of the test set. Confusion matrix between the five semantic categories, computed over all points of the test set.Confusion matrix between the five semantic categories, computed over all points of the test set.
  • Figure 5: Visual comparison between the proposed approach and Treeiso. The colours for the individual trees were assigned randomly. White ellipses mark noteworthy differences. Best viewed on screen.
  • ...and 5 more figures