solo2307/urban-tree-genera

Urban Tree Detection and Genera Mapping with YOLOv11-L

Urban Tree Detection and Genera Mapping with YOLOv11-L

This repository provides YOLOv11-L object detection models trained for urban tree mapping using multichannel very-high-resolution remote sensing imagery, including RGB, near-infrared (NIR), and height information derived from airborne LiDAR.

Two model variants are released:

1. Tree detection model for robust localization of individual urban trees
2. Tree genera detection model that assigns genus labels to detected trees

Both models are designed for large-scale urban environments and optimized for heterogeneous backgrounds, occlusions, and complex canopy structures.

Overview

Urban tree mapping is challenging due to heterogeneous backgrounds, occlusions, and varying tree structures. To address these challenges, the models in this repository leverage both spectral and structural information by combining:

• RGB orthophotos (visual structure and context)

• Near-infrared (NIR) imagery (vegetation discrimination)

• Height information derived from LiDAR (vertical structure of objects)

• The height channel represents absolute object height, not only canopy height, and is used to distinguish trees from other vertical structures and to improve robustness in dense urban environments.

Tree Detection Model

0. Tree

Tree Genera Detection Model

The genera model predicts bounding boxes with genus labels for common urban trees in Baden-Württemberg, Germany.

Included genera:

0. Acer
1. Aesculus
2. Carpinus
3. Coniferous
4. Fagus
5. Other Deciduous
6. Platanus
7. Prunus
8. Quercus
9. Tilia