Joyson Estibeiro | FETR-Treetop-Detection

Treetop Detection from Raster CHM - Testing Established Methods and a New Adaptation of the GTR Algorithm (Diószegi et al., 2024) for Mixed Linear Vegetation

An R-based project documenting my experiments with individual treetop detection on 1 m resolution CHMs for heterogeneous vegetation along railway corridors, aimed at tree risk assessment and spatial distribution mapping. After testing popular methods like LM+VWF and GTR, I developed FETR-Only by adapting a discarded part of the GTR algorithm to capture subtle peaks both methods missed. The script outputs both GTR and FETR detections, so users can compare and choose what fits their data.

GitHub Repo

Treetop Detection from Raster CHM

An R-based project documenting my experiments with individual treetop detection on 1 m resolution CHMs for heterogeneous vegetation along railway corridors, aimed at tree risk assessment and spatial distribution mapping. After testing popular methods like LM+VWF and GTR, I developed FETR-Only by adapting a discarded part of the GTR algorithm to capture subtle peaks both methods missed. The script outputs both GTR and FETR detections, so users can compare and choose what fits their data.

GitHub Repo

Why This Project:

Automating tree digitisation along railway lines saves significant time for risk assessment, even if results need some manual refinement. The goal is identifying every tree and shrub above 3 m that could fall across the tracks.

Most CHM-based detection methods are designed for structured forests or well-spaced urban trees. Railway vegetation is a mix of tall broadleaves, conifers, and smaller side shrubs with overlapping canopies and irregular spacing.

Using a 1 m resolution CHM with median smoothing, both LM + VWF and GTR missed smaller side trees. That gap led me to build FETR-Only, isolating a part of GTR's logic that the original discards, to better capture these smaller trees.

Tech Stack:

R (terra, sf, lidR) — Core scripting language for the detection pipeline
CloudCompare — Point cloud processing and CHM generation (DSM - DTM)
QGIS — Visual validation and comparison of detection results

Team Members:

Joyson Estibeiro

Affiliation: