Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Lidar-based Norwegian tree species detection using deep learning

Martijn Vermeer, Jacob Alexander Hay, David Völgyes, Zsófia Koma, Johannes Breidenbach, Daniele Stefano Maria Fantin

TL;DR

The paper tackles the challenge of producing up-to-date Norwegian tree species maps with minimal manual labeling by leveraging lidar-only data. It deploys a U-Net segmentation model using 1 m gridded DTM/CHM inputs and weak labels derived from SR16, augmented with a novel CowBatchMix technique and focal loss to handle label noise. The method achieves an OA of approximately $0.75$ and a macro $F_1$ of about $0.70$ on independent NFI plots, demonstrating the viability of open-access lidar data for fine-scale species mapping, albeit with some limitations relative to aerial or combined modalities. The study highlights practical benefits for high-resolution planning and forest management while outlining region-specific validity and avenues for national-scale validation and data enrichment.

Abstract

Background: The mapping of tree species within Norwegian forests is a time-consuming process, involving forest associations relying on manual labeling by experts. The process can involve both aerial imagery, personal familiarity, or on-scene references, and remote sensing data. The state-of-the-art methods usually use high resolution aerial imagery with semantic segmentation methods. Methods: We present a deep learning based tree species classification model utilizing only lidar (Light Detection And Ranging) data. The lidar images are segmented into four classes (Norway Spruce, Scots Pine, Birch, background) with a U-Net based network. The model is trained with focal loss over partial weak labels. A major benefit of the approach is that both the lidar imagery and the base map for the labels have free and open access. Results: Our tree species classification model achieves a macro-averaged F1 score of 0.70 on an independent validation with National Forest Inventory (NFI) in-situ sample plots. That is close to, but below the performance of aerial, or aerial and lidar combined models.

Lidar-based Norwegian tree species detection using deep learning

TL;DR

The paper tackles the challenge of producing up-to-date Norwegian tree species maps with minimal manual labeling by leveraging lidar-only data. It deploys a U-Net segmentation model using 1 m gridded DTM/CHM inputs and weak labels derived from SR16, augmented with a novel CowBatchMix technique and focal loss to handle label noise. The method achieves an OA of approximately and a macro of about on independent NFI plots, demonstrating the viability of open-access lidar data for fine-scale species mapping, albeit with some limitations relative to aerial or combined modalities. The study highlights practical benefits for high-resolution planning and forest management while outlining region-specific validity and avenues for national-scale validation and data enrichment.

Abstract

Background: The mapping of tree species within Norwegian forests is a time-consuming process, involving forest associations relying on manual labeling by experts. The process can involve both aerial imagery, personal familiarity, or on-scene references, and remote sensing data. The state-of-the-art methods usually use high resolution aerial imagery with semantic segmentation methods. Methods: We present a deep learning based tree species classification model utilizing only lidar (Light Detection And Ranging) data. The lidar images are segmented into four classes (Norway Spruce, Scots Pine, Birch, background) with a U-Net based network. The model is trained with focal loss over partial weak labels. A major benefit of the approach is that both the lidar imagery and the base map for the labels have free and open access. Results: Our tree species classification model achieves a macro-averaged F1 score of 0.70 on an independent validation with National Forest Inventory (NFI) in-situ sample plots. That is close to, but below the performance of aerial, or aerial and lidar combined models.
Paper Structure (20 sections, 1 equation, 3 figures, 2 tables)

This paper contains 20 sections, 1 equation, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Motivation
  3. Background and related work
  4. Materials and Methods
  5. Data sources
  6. Lidar pre-processing
  7. Label pre-processing
  8. Deep Learning model
  9. Loss function
  10. Augmentation
  11. Study area
  12. Results
  13. Discussion
  14. Limitations of the validation
  15. The benefits of the lidar source
  16. ...and 5 more sections

Figures (3)

  • Figure 1: Cowmix augmentation: mixing two images in cow patterns introduces, best visible on the elevation data (B), but also noticeable in the two other subfigures. Species are in RGB, background is white, unlabeled is black.
  • Figure 2: Training (red) and validation (green) data split in the study area (Viken, Norway).
  • Figure 3: Example prediction: the high-resolution prediction follows local edges more closely.