AI powered tree mapping in urban, rural and forest environments

Mapping individual trees is useful for a wide range of applications including forest management, urban planning, carbon monitoring and ecological modelling. Previously, various remote sensing techniques were employed in New Zealand to detect individual trees from 3D LiDAR point clouds and aerial imagery over large areas. A reliable approach for region-wide studies involved a canopy height-based method, which first identifies treetops using local maxima and then segments crowns using region-growing algorithms. However, recent advancements in artificial intelligence (AI) and high-performance computing enable more flexible and accurate tree mapping on a larger scale. In particular, Mask R-CNN, a region-based Convolutional Neural Network, is used globally to detect tree crowns from remote sensing imagery.
Jan Schindler leads a data science programme that applies these methods in diverse environments in New Zealand, including urban, silvopastoral and native forests using multi-resolution remote sensing data. In this presentation we showcase three key studies (i) detecting 1.8 million urban and forest trees in Wellington City, (ii) mapping different tree species on pastoral hill-country farms to inform landslide susceptibility modelling and (iii) delineating individual tree crowns in a complex native forest plot using high-resolution drone imagery. We will discuss the strengths and practical limitations of these novel methods in the New Zealand context.

[@andrewpkiror1982]

you're doing a good job. Is it possible that I can get your codes and datasets, just for practice ? thank you

[@love_dose-yt4dm]

Hey, can you please the model which is identifying the trees. It will be very helpful to me. Also it's great work. I really like it. Thank you

[@wakandavernon1412]

As he said, it's very difficult to identify the type of plantation from the top view. It should be region bounded.

[@geospatial90]

Dear sir,
Actually I want to classify tree species from high resolution image (12.5 cm). First of all I tried to experiment for individual tree crown segmentation using YOLO-8 model but in my case I have found multiple polygon for each tree instead of a single polygon. Can you please advise me why it happened?