Automatically identifying, counting, and describing wild animals in camera-trap images with deep learning
Problem
Extracting information from millions of camera-trap wildlife images is expensive, time-consuming manual work.
Why it matters
Manual image analysis limits ecology, conservation and wildlife research that depends on large-scale animal data.
What the AI did
Deep convolutional neural networks were trained to automatically identify, count and describe behaviors of 48 species in the 3.2 million-image Snapshot Serengeti dataset.
Results
>93% accuracy automatically identifying animals (abstract truncated mid-figure)
Impact
Demonstrates automatic, inexpensive extraction of wildlife data at scale, enabling big-data ecology.
Record
| Venue | Proceedings of the National Academy of Sciences |
|---|---|
| Published | 2018 |
| Work type | article |
| Authors | Mohammad Sadegh Norouzzadeh, Anh Nguyen, Margaret Kosmala, Alexandra Swanson, Meredith S. Palmer, Craig Packer, Jeff Clune |
| Citations | 1,211 (OpenAlex, at import time) |
| References | 80 |
| Model or method | deep convolutional neural networks (specific architecture not named) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1073/pnas.1719367115 |
| Open access | hybrid |
| Keywords | deep learning, camera traps, wildlife monitoring, conservation, image classification |
Importance rating
significant — thousands to millions affected
Abstract quantifies scale: 3.2M-image dataset, >17,000 h of human effort saved, enabling big-data ecology and conservation across multiple fields.