Abstract

Recent technological advancements in laser scanning have enabled the capture of high-resolution three-dimensional (3D) forest scenes, marking a significant shift in forest management towards sustainability. This presentation highlights the intersection of these cutting-edge laser scanning techniques with the evolving field of deep learning, emphasizing their combined potential in revolutionizing forest mapping and analysis. The focus is on recent developments in deep learning methodologies applied to 3D forest data and the exploration of machine learning-ready datasets that facilitate detailed, tree-level analysis. Specifically, we will delve into the application of deep learning algorithms for critical 3D point cloud tasks, such as tree instance segmentation, forest scene panoptic segmentation, and tree species classification. These advancements not only enable a more granular understanding of forest ecosystems at an individual tree level but also open avenues for optimizing biomass production and stock management while preserving vital ecosystem functions. This confluence of high-resolution 3D data acquisition and sophisticated deep learning approaches represents a transformative approach to sustainable forest management, promising significant contributions to both ecological conservation and resource optimization.

Stefano is a research scientist at the Norwegian Institute of Bioeconomy research doing research on the use of proximal laser scanning and deep learning for capturing individual tree properties and applying this information to forest management.