A review of three machine learning models using UAV derived photogrammetry for detecting and measuring tree stumps

Aug 5, 2026

Upload by Oliver

Accurate biophysical measurements from tree stumps left on post-harvest sites are essential for quantifying volume loss caused by inefficient harvesting and for estimating biomass-related gains. Advances in UAV-based digital aerial photogrammetry, combined with machine learning and object detection methods, enable remote sensing of forestry targets. This thesis trains three detection model families—Faster R-CNN, SSD, and YOLO—then evaluates detection performance and derives stump diameter and height from model bounding boxes and height estimation procedures.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages109

Reading Time4 h 35 min

AccessFree to download

Summary

Study approach: three machine learning models

Evaluation of detection and measurement accuracy

Results and implications

Future development

A review of three machine learning models using UAV derived photogrammetry for detecting and measuring tree stumps by Alan John Hubbard Thesis presented in partial fulfilment of the requirements for the degree of Master of Forest and Wood Science at Stellenbosch University Department of Forestry, Faculty of AgriSciences The financial assistance of the Department of Science and Technology (DST) , Forestry Sector Innovation fund (FSIF) (administered by Forestry South Africa) towards this research is hereby acknowledged. Opinions expressed and conclusions arrived at are those of the author and are not necessarily to be attributed to the DST-FSIF. Supervisor: Prof. Bruce Talbot Co-supervisor: Dr. Simon Ackerman Declaration By submitting this thesis electronically, I declare that the entirety of the work contained therein is my own, original work, that I am the sole author thereof (save to the extent explicitly otherwise stated), that reproduction and publication thereof by Stellenbosch University will not infringe any third-party rights and that I have not previously in its entirety or in part submitted it for obtaining any qualification. March 2023 Copyright © 2023 Stellenbosch University All rights reserved Summary Accurate biophysical data from stumps left on post-harvested sites is required to ascertain volume loss from inefficient harvesting techniques or volume gain from biomass utilisation. Recently advances in digital aerial photogrammetry data from unmanned aerial vehicles (UAVs) and machine learning, and object detection algorithms, have led to the increased use of this technology in forestry for remote sensing. Stumps, being mostly uniform in distribution and shape, are ideal objects for machine detection on digital orthomosaics. Resultant data from machine learning algorithms enables the possibility of estimating stump diameter and heights from virtual sources. In this study we trained three different machine learning model types, namely, Faster Regionbased Convolutional Neural Network (R-CNN), Single Shot Multibox Detector (SSD) and YouOnly-Look-Once (YOLO) . We assessed the detection rates of each model and compared metrics by using similarly annotated images. The resultant bounding boxes that encapsulated detected stumps were used to calculate diameters and compared to actual. Stump heights were determined using multiple methods which were also compared to actual height values. We found that visible stumps in post-harvested sites could be detected with high rates of accuracy, with almost perfect precision from some object detection models, albeit at low levels of recall. Overall, all three model types had an F1-score of above 73% with the best model attaining an F1-score of 89% . Diameters , although successfully calculated, produced an overestimation from actual in most cases. Similarly, calculated stump heights were underestimated in most cases. The objectives of this study were met in that insights into using machine learning algorithms for stump detection were broadened. The ability to process photogrammetry data quickly and accurately, with good estimations of diameter and height values, provides a useful tool to industry for estimating biomass volume from stumps left on post-harvested sites. Future development in technology will certainly improve accuracy and turn-around time of available data. Opsomming Akkurate data van afgesaagde bome wat op geoeste plantasies agtergelaat word, word benodig om volumeverlies van ondoeltreffende oestegnieke, asook die volume van agtergeblewe biomassa, vas te stel. Onlangse ontwikkeling in digitale lugfotogrammetrie data van onbemandelugvoertuie (hommeltuie) en masjienleer, en voorwerpopsporingsalgoritmes, het gelei tot die toenemende gebruik van hierdie tegnologie in bosbou vir afstandswaarneming. Stompe, wat meestal eenvormig in plasing en vorm is, is ideale voorwerpe vir masjienopsporing op digitale ortomosaieke. Die data

FAQs

Which machine learning model types were trained for stump detection?

Three model families were trained: Faster R-CNN, SSD, and YOLO.

How were detected stumps used to estimate dimensions?

Bounding boxes from the detected stumps were used to compute diameters, while stump heights were determined using multiple methods and compared with actual height values.

What were the main accuracy findings for detection and measurements?

Visible stumps could be detected with high accuracy, with F1-scores above 73% and a best score of 89%, though recall remained low for some models. Diameter estimates tended to overestimate, while height estimates tended to underestimate.

Aug 5, 2026

Upload by Oliver