[Supplementary Material] Detecting Hurricane-Induced Fallen Pecan Trees: A Novel UAV-Based Deep Learning Approach

Description

Hurricane events strongly affect the pecan crop by uprooting and lodging trees. Additionally, current fallen-tree monitoring relies on manual field surveys, which are invasive, time-consuming, and costly, constraining timely decision-making. Therefore, in this study, we deployed a deep learning (DL) framework based on the You Only Look Once (YOLO26) model to detect fallen pecan trees using unmanned aerial vehicle (UAV) RGB images. Flights were conducted over four pecan fields, ten days after Hurricane Helene crossed the state of Georgia, USA. As a result, 546 images were acquired and individually analyzed to detect fallen trees. Initially, ground-truth data were generated through assisted image processing, resulting in 2,408 annotations labeled “Fallen”. For our analysis, three fields were considered for the model development (training and validation). Subsequently, to ensure the model accuracy and reliability, a fourth independent field was used as the test dataset. Our results showed that the fallen tree detection models achieved a precision of 80.98–88.48%, a recall of 61.25–72.08%, and a mAP@50 of 70.93–77.50%. Among the evaluated variants, YOLO26m demonstrated the best performance on the independent test dataset, achieving an R2 of 0.74 and a mean absolute error (MAE) of 0.61 trees per image. Furthermore, we designed a user-friendly platform as a proof of concept to evaluate the model’s operability. This study, therefore, presents a novel UAV-based object detection framework for detecting fallen pecan trees, empowering stakeholders with a precise, accurate, non-invasive, safe, and rapid solution. These findings also support precision agriculture practices and promote the integration of advanced technologies into tree-crop management systems.

Institutions

• University of Georgia

  Georgia, Athens
• University of Missouri

  Missouri, Columbia

Categories

Licence