Biodiversity is declining across the globe at a catastrophic rate, as threats from human settlement expansion, illegal wildlife killing, and climate change place enormous pressure on wildlife populations. Conservation biologists are faced with the daunting – but urgent – task of surveying wildlife populations and making policy recommendations to governments and industry. What species need legal protection from hunting? A road needs to connect two cities; which route will have the least detrimental impact on wildlife habitat? Where will it be most effective to build underpasses as wildlife migration corridors? Where should we deploy anti-poaching resources? Informed policy decisions on questions like these start with data: just as your doctor can only prescribe treatment after running trusted diagnostic tests, policy-makers and protected area managers can only act to protect biodiversity if robust, up-to-date data is available when they need it.
Data, in this case, is wildlife population estimates, for which remote sensing has become the most powerful tool in the conservation toolbox. But currently, collecting data about how animals use their habitats – or even how many animals live in a given area – is dependent on tremendous amounts of manual data annotation: it often takes years for a small NGO to annotate millions of images or audio recordings for a single project. This bottleneck precludes real-time applications, and often delays critical answers to conservation questions so long that by the time they’re available, they’re no longer relevant. Machine learning is poised to break this annotation logjam, and to greatly accelerate conservation decision-making.
We apply machine learning tools to a variety of image sources – including motion-triggered camera traps, aerial cameras, and microphones – to accelerate ecologists’ workflows. Our team spans Microsoft Research, AI for Earth, and the AI for Good Research Lab.
- Accelerating the processing of images from motion-triggered camera traps
- Species classification from handheld photos
- Active learning for wildlife detection in aerial images
- Detecting seals in aerial imagery (w/NOAA Fisheries)
- Using machine learning to detect beluga whale calls in hydrophone recordings (w/NOAA Fisheries)
- Multi-species bioacoustic classification (w/Sieve Analytics)
- In collaboration with many partners, we maintain a repository of labeled conservation images, the Labeled Information Library of Alexandria: Biology and Conservation, aka “LILA”, pronounced “lie-la”.
Publications and publication-like things
- Zhong M, Taylor R, Bates N, Christey D, Basnet H, Flippin J, Palkovitz S, Dodhia R, Ferres JL. Acoustic detection of regionally rare bird species through deep convolutional neural networks. Ecological Informatics. 2021 May 28:101333.
- Zhong M, Torterotot M, Branch TA, Stafford KM, Royer JY, Dodhia R, Lavista Ferres J. Detecting, classifying, and counting blue whale calls with Siamese neural networks. The Journal of the Acoustical Society of America. 2021 May 6;149(5):3086-94.
- Gupta G, Kshirsagar M, Zhong M, Gholami S, Ferres JL. Comparing recurrent convolutional neural networks for large scale bird species classification. Scientific reports. 2021 Aug 24;11(1):1-2.
- Robinson C, Ortiz A, Hughey L, Stabach JA, Ferres JM. Detecting Cattle and Elk in the Wild from Space. arXiv preprint arXiv:2106.15448. 2021 Jun 29.
- Kellenberger B, Tuia D, Morris D. AIDE: Accelerating Image‐Based Ecological Surveys with Interactive Machine Learning. Methods in Ecology and Evolution, September 2020.
- Norouzzadeh M, Morris D, Beery S, Joshi N, Jojic N, Clune J. A deep active learning system for species identification and counting in camera trap images. Methods in Ecology and Evolution, October 2020.
- LeBien J, Zhong M, Campos-Cerqueira M, Velev JP, Dodhia R, Ferres JL, Aide TM. A pipeline for identification of bird and frog species in tropical soundscape recordings using a convolutional neural network. Ecological Informatics. 2020 Jun 8:101113.
- Zhong M, LeBien J, Campos-Cerqueira M, Dodhia R, Ferres JL, Velev JP, Aide TM. Multispecies bioacoustic classification using transfer learning of deep convolutional neural networks with pseudo-labeling. Applied Acoustics. 2020 Sep 1;166:107375.
- Shashidhara BM, Mehta D, Kale Y, Morris D, Hazen M. Sequence Information Channel Concatenation for Improving Camera Trap Image Burst Classification. arXiv preprint arXiv:2005.00116. 2020 Apr 30.
- Zhong M, Castellote M, Dodhia R, Lavista J, Keogh M, Brewer A. Beluga whale acoustic signal classification using deep learning neural network models. The Journal of the Acoustical Society of America 147, 1834 (2020).
- Beery S, Liu Y, Morris D, Piavis J, Kapoor A, Meister M, Perona P. Synthetic examples improve generalization for rare classes. Winter Conference in Applications of Computer Vision (WACV), Aspen, CO, 2020.
- Zhong M, Castellote M, Dodhia R, Lavista Ferres J, Keogh M, Brewer A. Improving passive acoustic monitoring applications to the endangered Cook inlet beluga whale. The Journal of the Acoustical Society of America. 2019 Oct;146(4).
- Norouzzadeh M, Morris D, Beery S, Joshi N, Jojic N, Clune J. A deep active learning system for species identification and counting in camera trap images. arXiv preprint arXiv:1910.09716. 2019 Oct 22.
- Beery S, Morris D, Yang S. Efficient Pipeline for Camera Trap Image Review. KDD Workshop Data Mining and AI for Conservation (DMAIC), Anchorage, AK, August 2019.
- Kaeser-Chen C, Birch T, Chou K, Gadot T, Adam H, Belongie S, Robertson T, Fegraus E, Morris D. Towards Ethical Deployment of AI for Conservation Systems. KDD Workshop Data Mining and AI for Conservation (DMAIC), Anchorage, AK, August 2019.
- Beery S, Morris D, Yang S, Simon M, Norouzzadeh M, Joshi N. Efficient Pipeline for Automating Species ID in new Camera Trap Projects. Biodiversity Information Science and Standards. 2019 Jun 19;3:e37222.
Blogs and blog-like things
- “Wildlife Protection Solutions helps protect the wildest places with Microsoft AI for Earth”. Microsoft Customer Story.
- “Using Artificial Intelligence to Identify Endangered Beluga Whales”. NOAA Fisheries News.
- “Developing Artificial Intelligence to Find Ice Seals and Polar Bears from the Sky”. NOAA Fisheries News.
- “Helping Scientists Protect Beluga Whales with Deep Learning”. Azure and AI for Earth on Medium.
- “Accelerating biodiversity surveys with Azure Machine Learning”. Azure and AI for Earth on Medium.
- “Artificial intelligence makes a splash in efforts to protect Alaska’s ice seals and beluga whales”. Microsoft News.
Other work at Microsoft
Whoever “we” are (i.e., the “we” who maintain this page), “we” are not the only ones at Microsoft working in this area. Here are some other great projects that our Microsoft colleagues have worked on in the biodiversity space:
- Counting Puffins with AI (with SSE Renewables)
- Accelerating camera trap workflows (with the Snow Leopard Trust)
- Accelerating seabird surveys with active learning (with Conservation Metrics)
- Accelerating seabird surveys with Azure ML Workbench (with Conservation Metrics)
- Accelerating poacher detection (with Peace Parks)
- Accelerating marine video surveys (with the Northern Territory)
- Accelerating giraffe surveys (with the Wild Nature Institute) (paper)
- Accelerating drone-based wildlife surveys and land cover mapping (with Kakadu National Park)
- WWF/Microsoft Hackathon to user computer vision to identify illegal pangolin products in online marketplaces (with WWF)
- WWF/Microsoft Hackathon to use computer vision to identify illegal pangolin products in online marketplaces (with WWF)
- Microsoft/Heathrow collaboration that uses computer vision to screen for illegal wildlife products at Heathrow airport (with Heathrow) (video)
- Survey of everything we know about machine learning and camera traps
- Survey of everything we know about machine learning and aerial wildlife surveys
Machine learning doing serious important things