Animal ecology has moved into the era of big data and the Internet of Things. Extraordinary amounts of data are currently being gathered on wildlife populations, owing to advanced technology such as drones, satellites, and terrestrial devices like automatic cameras and sensors positioned on animals or in their environs.
Image Credit: Piyawat Nandeenopparit/shutterstock.com
These data have become so easy to obtain and share that they have reduced distances and time requirements for scientists while decreasing the unsettling presence of humans in natural habitats.
Nowadays, a range of artificial intelligence (AI) programs are available to examine large datasets, but they are mostly general in nature and unsuitable for monitoring the precise behavior and appearance of wild animals.
A team of researchers from EPFL and other universities has drawn a pioneering method to resolve that issue and create more accurate models by merging advances in computer vision with the know-how of ecologists. Their findings, which have been published in the journal Nature Communications, pave the way toward new perspectives on the application of AI to help safeguard wildlife species.
Building up Cross-Disciplinary Know-How
Wildlife studies have expanded from local to global. Advanced technology currently offers innovative new methods to generate more accurate estimates of wildlife populations, fight poaching, better understand animal behavior, and stop the deterioration in biodiversity.
Ecologists can utilize AI, and more precisely computer vision, to derive crucial features from images, videos, and other visual versions of data in order to swiftly categorize wildlife species, count individual animals, and extract specific information, using large datasets.
The generic programs presently used to process such data mostly function like black boxes and do not exploit the total scope of current knowledge about the animal kingdom.
They are challenging to customize, suffer from poor quality control from time to time, and are possibly subject to ethical issues associated with the use of sensitive data. They also comprise numerous biases, particularly regional ones; for example, if all the data used to train a particular program were gathered in Europe, the program might not be appropriate for other regions of the world.
We wanted to get more researchers interested in this topic and pool their efforts so as to move forward in this emerging field. AI can serve as a key catalyst in wildlife research and environmental protection more broadly.
Prof. Devis Tuia, Study Lead Author and Head of Environmental Computational Science and Earth Observation Laboratory, EPFL
If computer scientists want to decrease the margin of error of an AI program that has been taught to identify a particular species, for example, they need to be able to make use of the expertise of animal ecologists.
These professionals can stipulate which characteristics should be added into the program, such as whether a species can endure at a particular latitude, whether it is critical for the existence of another species (such as through a predator-prey relationship), or whether the physiology of the species varies over its lifetime.
For example, new machine learning algorithms can be used to automatically identify an animal. such as using a zebra's unique stripe pattern, or in video their movement dynamics can be a signature of identity.
Prof. MackenzieMathis, the head of EPFL's Bertarelli Foundation Chair of Integrative Neuroscience and co-author of the study
Prof. MackenzieMathis continued: "Here is where the merger of ecology and machine learning is key: the field biologist has immense domain knowledge about animal being studied, and us as machine learning researchers job is to work with them to build tools to find a solution."
Getting the Word Out About Existing Initiatives
The notion of building stronger ties between computer vision and ecology evolved as Tuia, Mathis, and others presented their research challenges at many conferences over the last two years. They realized that such an association could be very beneficial in stopping some wildlife species from becoming extinct.
A few initiatives have already been executed in this direction; some of them are mentioned in the Nature Communications article. For instance, Tuia and his team at EPFL have come up with a program that can detect animal species based on drone images. This was verified recently on a seal population.
Meanwhile, Mathis and her colleagues have launched an open-source software package termed as DeepLabCut that allows researchers to approximate and track animal poses with extraordinary accuracy. It has been downloaded 300,000 times until now. DeepLabCut was developed for lab animals but can be utilized for other species as well.
Scientists at other universities have created programs too, but it is tough for them to share their discoveries since no physical community has yet been set up in this area. Other researchers often do not know these programs are available or which one would be suitable for their particular research.
Nevertheless, preliminary steps toward such a community are happening through different online forums. The Nature Communications article aims for a wider audience, however, involving researchers from various parts of the world.
A community is steadily taking shape. So far we’ve used word of mouth to build up an initial network. We first started two years ago with the people who are now the article’s other lead authors: Benjamin Kellenberger, also at EPFL; Sara Beery at Caltech in the US; and Blair Costelloe at the Max Planck Institute in Germany.
Prof. Devis Tuia, Study Lead Author and Head of Environmental Computational Science and Earth Observation Laboratory, EPFL
Journal Reference:
Tuia, D., et al. (2022) Perspectives in machine learning for wildlife conservation. Nature Communications. doi.org/10.1038/s41467-022-27980-y.
Source: https://www.epfl.ch/en/