Ecology by UAV, can we use images from drone surveys to understand the traits of tree species in forest canopy and in regenerating species post disturbance?

Hello everyone,

I am interested in automatic extraction of plant traits and functional groups from drone surveys.

I have data from a few drone surveys over forest which I will bring from two areas, forests in Arizona and the tropical cloud forest of Peru.

I was interested to test out a particular project which would likely be a simpler problem with these data: to classify trait types of species in a regenerating fire scar in Arizona, where there are a mix of young pioneer and climax species regenerating. I collected some GPS point labels for these species in the field, however, I realized that my points and my imagery are not currently well aligned.

These data may still be utilized but there currently aren't existing labels for training and testing the trait types.

Another interesting option would be identifying trait types such as leaf form and habit in the tropical forest canopy, which does include a diversity of leaf types, and I believe would be a fascinating problem test case. However, as mentioned, these are also at present unlabeled.

If my were to be utilized, the process would likely include 1) segmentation of forest crowns in the imagery, 2) perhaps some manual labeling, then 3) using the outputs from processing images as predictors for the computer vision step, specifically: the imagery (RGB orthomosaic maps) and structure from motion (sfm) point cloud maps which also add some extra 3D information for the structure of the canopy in the scene.

Excited to hear any thoughts and feedback!

Thanks all.

No repository files, tests, or entry points are identified. Start by clarifying whether the Arizona regeneration or Peru canopy study is in scope, then assess the GPS and imagery alignment, available labels, RGB orthomosaics, and SfM point clouds. Done is not defined because the issue remains an exploratory proposal.

Written by the indexing model from the issue text.