Task:
The client needed annotated data to train and validate models for automated tree detection and species recognition based on aerial photos. Each tree had to be identified with coordinates, height, crown shape, and species class. The task involved handling a large volume of objects with subtle visual distinctions and structuring data for digital “tree passports.”
Key challenges included:
- Filtering out objects shorter than 2.5 meters to avoid misclassification
- Managing tens of thousands of individual trees with diverse canopy shapes and colors
- Accurately matching tree species based on visual crown characteristics requiring manual expert annotation
Solution:
-
- 01
-
Preparation and guidelines
- Developed annotation guidelines defining species-specific crown shapes, color patterns, and height thresholds
- Combined automated HD-Forest recognition with manual validation for a balanced approach
- Created a standardized format for “Digital Tree Passports” including geolocation, species, height, and canopy data
-
- 02
-
Annotation process
- Automated detection and initial classification of trees using aerial imagery processing software
- Manual review and correction of ambiguous cases and species assignments based on visual traits
- Structured metadata tagging for height, vertical layering, and canopy complexity
-
- 03
-
Quality control
- Performed selective validation checks on random samples for species and parameter accuracy
- Provided annotators with regular feedback based on error analysis and consistency reviews
- Ensured dataset quality met standards for forestry monitoring applications and model training
Results:
Accurately annotated 200.000 trees across 10 species classes
Created detailed digital passports for each tree, supporting scalable forest monitoring
Established a robust dataset enabling improved species recognition and ecosystem management