weecology/DeepForest

What it solves

DeepForest solves the challenge of identifying and delineating individual tree crowns and other ecological objects in high-resolution airborne RGB imagery. It addresses the gap where data acquisition speeds outpace the same-speed identification of organisms in remote sensing surveys, providing an open-source tool for biodiversity and forestry surveys.

How it works

DeepForest is a Python package built on the torchvision object detection module. It uses deep learning object detection networks to predict bounding boxes for individual trees and birds in RGB imagery. The package provides pre-trained models for tree crown and bird detection, which users can extend by annotating and training custom models for specific forest types, sensors, or spatial resolutions.

Who it’s for

Ecologists, forestry researchers, and environmental monitoring specialists who need to automate the same-speed identification of individual organisms in airborne imagery.

Highlights

• own pre-trained models for tree crown and bird detection.
• Built on PyTorch's torchvision for simplified training and training of custom models.
• Supports RGB imagery for ecological object detection.
• Open-source implementation for crown delineation in remote sensing remote sensing.