skip to content

Cambridge Centre for Carbon Credits (4C)

An overhead image of a forest with tree crowns circled to show individual trees and areas without trees.

This news article was written by Andrew Benton and was first published by the Cambridge Conservation Research Institute.

A new study by James Ball and colleagues in the Cambridge Conservation Research Institute has developed a computer vision method for delineating tree crowns in tropical forests from aerial RGB imagery, with the new approach available as an open-source Python package for broader uptake and further development.

Monitoring the growth and mortality of upper-canopy trees in tropical forests is essential for understanding forest carbon dynamics and the resilience of forests to climate change. Historically, monitoring has been conducted through labour-intensive ground-level field surveys in which large trees are relatively uncommon. Accordingly, estimates of their growth and mortality are poorly constrained. While remote sensing has promised to radically expand sample sizes, lidar-based approaches have thus far struggled to isolate individual trees in complex tropical forest canopies.  

Detectree2 is a new, deep convolutional neural network tool that identifies tropical forest trees in RGB images and is applicable for many aspects of forest ecology and conservation, from estimating carbon stocks to monitoring forest phenology and restoration. It builds on the Mask R-CNN computer vision framework to recognize the irregular edges of individual tree crowns from airborne RGB imagery.

"Detectree2 can be rapidly trained to perform well in new types of forest and other sparser ecosystems. By making it available as an open-source Python package, other groups can test and develop it further for their own unique circumstances," said Ball.

"This tool can help to significantly expand the area of forests that can be monitored and help us understand how they are responding to global change", added Professor Coomes.

Detectree2 can harness the spectral and textural information available in RGB aerial imagery to isolate individual crowns in complex, dense canopies. This can support the scaling up of effective forest monitoring with UAV, plane and satellite sensors.

Download the Detectree2 software

Read the paper:

James G. C. Ball, Sebastian H. M. Hickman, Tobias D. Jackson, Xian Jing Koay, James Hirst, William Jay, Matthew Archer, Mélaine Aubry-Kientz, Grégoire Vincent, David A. Coomes (2023). Accurate delineation of individual tree crowns in tropical forests from aerial RGB imagery using Mask R-CNN. Remote Sensing in Ecology and Conservation.