New AI Model Targets the Species Scientists Have Been Missing
Researchers have developed an artificial intelligence system that identifies tree-dwelling animals in tropical forests-a gap that existing camera-trap technology has largely ignored. TropiCam-AI recognizes 84 taxa, including 63 species, with 95% accuracy for the majority of them.
Camera traps generate millions of images annually, and AI has become standard for sorting through them. But most models focus on ground-dwelling animals. Arboreal species-primates, birds, small mammals living in forest canopies-have received minimal attention from developers.
Andrea Zampetti, a Ph.D. candidate at Sapienza University of Rome, led the project with the TROPECOLNET group at Spain's National Museum of Natural Sciences. The team built TropiCam-AI specifically for neotropical canopy surveys.
Why This Matters for Conservation
Tree-dwelling species are critical to tropical forest ecosystems. Primates, small mammals, and birds consume up to 90% of plant seeds in rainforests, making them essential seed dispersers. Deforestation threatens these species directly, creating an urgent need to monitor and study them.
Yet arboreal camera trapping remains severely underrepresented in AI training data compared to terrestrial imaging, according to research published earlier this year by Zampetti and colleagues.
How the Model Was Built
Zampetti spent three months in Brazil collecting training data, working with local communities and the NGO Instituto Juruá. He expanded the dataset by obtaining camera-trap images from researchers in Peru, Costa Rica, and French Guiana, and tapped the citizen science platform iNaturalist for additional images.
The team manually annotated each image-identifying species by hand-to create training material for the algorithm. This labor-intensive process ensures the AI learns to recognize what it's actually looking at.
Scientists can now upload camera-trap images to the model for species identification. If the system cannot confidently identify a species from an image, it moves up the taxonomic hierarchy and provides a genus-level classification instead of forcing an incorrect prediction.
Current Performance and Next Steps
TropiCam-AI achieves 95% accuracy, with 50 of its 84 recognized taxa exceeding 90% precision and recall. Zampetti said the team will continue refining the tool with additional training data from collaborators eager to contribute.
The model's effectiveness depends entirely on the quality and breadth of its training data. Expanding the sample size and adjusting parameters will improve performance across different applications and regions.
For researchers processing large datasets, the tool reduces manual analysis time significantly. Ecologists can now process millions of images automatically rather than reviewing each one by hand.
Learn more about AI for Science & Research and how machine learning is applied to conservation challenges.
Your membership also unlocks:
