DeepForest

DeepForest delineates individual tree crowns in high-resolution RGB imagery using deep learning for remote sensing and forest ecological studies.

Key Features:

• Deep Learning Integration: Leverages deep learning algorithms to detect individual tree crowns in high-resolution RGB images.
• Pre-trained Model: Includes a model pre-trained on over 30 million algorithmically generated tree crowns from 22 diverse forests.
• Fine-Tuning Capabilities: Supports fine-tuning using 10,000 hand-labeled crowns from six forests to adapt the model to specific datasets.
• Versatile Application: Enables applying the pre-trained model to new data, fine-tuning with user-labeled crowns, training new models, and evaluating model predictions.

Scientific Applications:

• Tree Crown Delineation: Provides precise tree crown delineation for remote sensing studies of forest structure and composition.
• Cross-Sensor and Resolution Adaptation: Can be adapted to different sensors and spatial resolutions for varied ecological imaging conditions.
• Demonstrated Use Cases: Has been applied to data from the National Ecological Observatory Network (NEON), tropical forests in French Guiana, and urban street trees in Portland, Oregon.

Methodology:

Model training uses deep learning trained on over 30 million algorithmically generated tree crowns from 22 forests, followed by fine-tuning with 10,000 hand-labeled crowns from six forests.