Seminar: Using synthetic tree data for supervised deep learning from forest LiDAR point clouds, 15th Feb, 1:00pm
When: Thursday 15th of February, 1:00pm AEST
Where: This seminar will be partially presented at the Rose Street Seminar area (J04) and partially online via Zoom. RSVP
Speaker: Dr Mitch Bryson
Title: Using synthetic tree data for supervised deep learning from forest LiDAR point clouds
Abstract:
The Australian Centre for Robotics participates in several collaborative research projects with partners in the Australian and New Zealand forestry sector, focused on quantifying and monitoring forest resources using airborne and terrestrial LiDAR and photogrammetry. LiDAR point clouds are used for measuring forest composition, tree species identification and mapping tree structural characteristics on a tree-by-tree level that can be important for tree phenotyping, measuring fire risk and for habitat assessment. Supervised deep learning methods are now routinely used for example to segment individual trees and measure tree structural characteristics such as stem shape, volume and branching from forest LiDAR point clouds.
In real applications, manual annotation of 3D point clouds needed for supervised learning is difficult and time-consuming. In this talk I will discuss a recent research project on using synthetic/simulated models of forest trees and forest stands for learning models which can analyse real forest point clouds. Use of synthetic data for training provides one avenue for reducing the need for manual-labelling effort when working in a new forest environment. The research explores what aspects of the synthetic data generation and tree simulation benefit learning and how to transfer models learnt on synthetic source data to real target environments.
Bio:
Mitch is currently a researcher at the ACFR and lecturer in the School of Aerospace, Mechanical and Mechatronic Engineering at the University of Sydney. He leads the ACFR Forestry research group and his research interest include aerial and marine robotic navigation, mapping and sensor fusion, working in applications in forestry using computer vision, laser scanning and hyperspectral imaging.