Benchmarking YOLOv11m-Seg, U-Net++ and Mask R-CNN Networks as a Tool for Assessing Forest Plantation Success

Monitoring forest plantations is a challenging task that normally requires tree counting on the ground. The task can be challenging and time-consuming, especially in areas of difficult access. The monitoring process proposed in the present research is based on several images collected via flights of a DJI Mavic 3M drone. A number of these images were collected, selected and processed to optimise the training, validation and testing of the developed algorithms. Three models were compared, including YOLOv11m-seg, U-Net++ and Mask R-CNN networks, to understand the pros and cons of using each one of these algorithms for object detection, counting and segmenting tree crowns with RGB images obtained via UAV. In order to develop a tool to assist in the evaluation of planting success, images were collected in Serra do Açor, Arganil, Coimbra, Portugal. The results were promising, as YOLOv11m-seg and Mask R-CNN achieved instance F1 scores of 86.8% and 87.8%, respectively, with the latter demonstrating greater accuracy in plantation areas. At the pixel level, IoU was 78.1% for U-Net++, 77.3% for Mask R-CNN and 76.7% for YOLOv11m-seg. YOLOv11m-seg was the fastest model on the CPU, running at 1.0 s per tile, compared to 3.3 s for Mask R-CNN and 5.8 s for U-Net++. It also predicted the fewest false tree crowns outside the plantation’s areas. This demonstrates the viability of the methodology and that model selection will rely on monitoring needs (as summarised in a decision matrix in this document). This method reduces costs and labour time, increasing the possibility of comparing aerial images more frequently, which will support decision-making in forest management.

Authors

Institutions

Publication Details

Journal
Applied Sciences
Published
2026-11-12
DOI
https://doi.org/10.3390/app162211205
Primary Topic
Advanced Neural Network Applications
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Benchmarking YOLOv11m-Seg, U-Net++ and Mask R-CNN Networks as a Tool for Assessing Forest Plantation Success

Mateus Mendes, Raúl Salas-González, Paulo Lucas Figueiredo, I. S. Brandão et al.
Applied Sciences
Advanced Neural Network Applications
article

Benchmarking YOLOv11m-Seg, U-Net++ and Mask R-CNN Networks as a Tool for Assessing Forest Plantation Success

Mateus Mendes, Raúl Salas-González, Paulo Lucas Figueiredo, I. S. Brandão, Beatriz Fidalgo
article en

Abstract

Monitoring forest plantations is a challenging task that normally requires tree counting on the ground. The task can be challenging and time-consuming, especially in areas of difficult access. The monitoring process proposed in the present research is based on several images collected via flights of a DJI Mavic 3M drone. A number of these images were collected, selected and processed to optimise the training, validation and testing of the developed algorithms. Three models were compared, including YOLOv11m-seg, U-Net++ and Mask R-CNN networks, to understand the pros and cons of using each one of these algorithms for object detection, counting and segmenting tree crowns with RGB images obtained via UAV. In order to develop a tool to assist in the evaluation of planting success, images were collected in Serra do Açor, Arganil, Coimbra, Portugal. The results were promising, as YOLOv11m-seg and Mask R-CNN achieved instance F1 scores of 86.8% and 87.8%, respectively, with the latter demonstrating greater accuracy in plantation areas. At the pixel level, IoU was 78.1% for U-Net++, 77.3% for Mask R-CNN and 76.7% for YOLOv11m-seg. YOLOv11m-seg was the fastest model on the CPU, running at 1.0 s per tile, compared to 3.3 s for Mask R-CNN and 5.8 s for U-Net++. It also predicted the fewest false tree crowns outside the plantation’s areas. This demonstrates the viability of the methodology and that model selection will rely on monitoring needs (as summarised in a decision matrix in this document). This method reduces costs and labour time, increasing the possibility of comparing aerial images more frequently, which will support decision-making in forest management.

Applied SciencesVol. 16(22)
University of Beira Interior (PT), Universidade Politécnica de Coimbra (PT)
Programa Operacional Regional do Centro
Life on land
Advanced Neural Network Applications
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon