Automatic Tree Detection and Delineation of Norway Spruce (Picea Abies) using UAV imagery across the Czech Republic

Dipl.-Ing. Salma Bijou

Bachelor's thesis

Abstract:

Unmanned aerial vehicles (UAVs) equipped with sensors or LiDAR can be used to collect high-resolution data in forests, which can then be used to generate canopy height models, point clouds, orthomosaics and multiple derived products. This approach is becoming increasingly popular in forest health assessment at the tree level as it provides accurate and detailed information about the crowns delimitation and structure of the forest canopy. To achieve this, different ITCD algorithms were used in the literature. Each algorithm has its advantages and limitations. In this study, I investigated the use of different individual tree detection and delineation (ITCD) of spruce trees using UAV-borne imagery. I tried to implement and apply different ITCD algorithms for our selected study area in Krkonoše Mountains, to evaluate and compare the results of the techniques used and examine them. For this purpose, the processing workflow includes (a) the generation of Digital Surface Models and orthorectified mosaics using UAV-borne imagery; (b) deriving Digital Terrain Models and building Canopy Height Models; (c) applying different ITCD algorithms across the selected study area. The treetop detection and delineation results were first visually validated using reference manually tree crowns. The treetop detection highest precision was achieved using the variable window size local maxima filtering. The highest overall accuracy for the delineation was achieved using the region growing algorithm with F1 score of 98%. The other algorithms, Inverse watershed segmentation, and marker controlled watershed segmentation algorithm demonstrated also impressive accuracy rates in delineating individual trees (with F1 scores of 95.77% and 96.54%, respectively). The findings of this study offer forestry professionals useful insights in determining the most effective method for identifying individual trees.

Keywords

Forestry, UAV, spruce trees, automatic tree detection and crown delineation

Language used: English

Date on which the thesis was submitted / produced: 3. 4. 2023

Thesis defence

• Date of defence: 23. 5. 2023
• Supervisor: Ing. Jan Komárek, Ph.D.
• Reader: David Moravec, Ph.D.

Czech University of Life Sciences Prague

Faculty of Environmental Sciences

Bachelor programme:
Geographic Information Systems and Remote Rensing in Environmental Sciences