ICDSUPL4-P003

Volume: 4, 2025
4th International PhD Student’s Conference at the University of Life Sciences in Lublin, Poland:
ENVIRONMENT – PLANT – ANIMAL – PRODUCT

Abstract number: P003

DOI: https://doi.org/10.24326/ICDSUPL4.P003

Published online: 9 April 2025

ICDSUPL, 4, P003 (2025)

Detection of pear trees using convolutional neural network based on data collected by drones

Kamil Buczyński^1*, Magdalena Kapłan¹

¹ Institute of Horticulture Production, University of Life Sciences in Lublin, Głęboka 28, 20-612 Lublin, Poland

^{* Corresponding author: kamil.buczynski@up.lublin.pl}

Abstract

Unmanned Aerial Vehicles (UAVs) are playing an increasingly important role in the monitoring and management of fruit tree crops. This study evaluated the performance of the YOLOv11s model, based on a convolutional neural network architecture, in detecting pear trees of the Lukasówka variety in the leafless phase using drone-acquired images. To train two models using different image annotation approaches, the dataset was labeled in two ways: one method involved labeling individual trees, while the other involved labeling entire rows of trees. The dataset, comprising 1400 images, was collected under varying lighting conditions and flight altitudes using DJI Mini 3 Pro and Mavic 3 Multispectral drones. The images were rescaled to a resolution of 640 × 480 while preserving the original 4:3 aspect ratio. The YOLO models were trained in the Google Colab environment using an NVIDIA A100 40 GB Tensor Core GPU. Each model was trained for 300 epochs. The performance of the trained models was evaluated using recall, precision, mAP0.5, and mAP50:95 metrics. Across both annotation methods, individual tree detection and tree row identification, the trained models demonstrated high levels of precision, recall, and mAP0.5. The greatest variation was observed in mAP50:95, suggesting the necessity for further refinement under stricter IoU thresholds. Future research should focus on expanding the analysis to other fruit tree species and integrating additional deep learning techniques to improve the system’s adaptability and resilience in diverse environmental conditions.

Keywords: UAV, YOLO, cnn, object detection, machine vision

How to cite

K. Buczyński, M. Kapłan, 2025. Detection of pear trees using convolutional neural network based on data collected by drones. In: 4th International PhD Student’s Conference at the University of Life Sciences in Lublin, Poland: Environment – Plant – Animal – Product. https://doi.org/10.24326/ICDSUPL4.P003