Open new site version
What we do
Премия рунета 2017

Neural networks will learn to recognize walruses

16 august 2021
Nenetsky State Nature Reserve team and scientists from the Moscow Institute of Physics and Technology, with the support of WWF-Russia, will train artificial intelligence to analyze data from camera traps installed on walrus haulouts, and, moreover to recognize pinnipeds from photographs. To collect educational photographic material for the neural network, specialists spent almost three weeks on an island in the Barents Sea.
Matveyev Island was chosen as a testing ground, where the largest Atlantic walrus rookery in the region - up to a thousand individuals - is formed every summer under the protection of the Nenetsky reserve. Monitoring of this territory provides scientists with important information about the state of the subspecies listed in the Red Book of Russia. To create a safe working environment, WWF helped to equip an observation and monitoring post on Matveyev in 2017.
Walruses on Matveyev Island
Yulia Bogomolova / WWF-Russia
“Every year neural networks are more and more widely used in studies of marine mammals, but no one has ever worked with walruses in this direction, at least in our country,” says Varvara Semenova, Arctic biodiversity conservation projects coordinator at WWF-Russia. - In this regard, the Pechora walruses are excellent for practicing this approach - they are a small and isolated group. This means that the probability of re-meeting the same walruses in the rookery in different years is quite high. When we learned about the initiative of the Nenetsky reserve to start such work, we immediately decided to support it. If neural networks prove to be effective, then a similar approach can be applied in other regions of the Arctic, both in specially protected natural areas and beyond."
Camera traps are one of the main tools for studying the life of a haulout. They work autonomously all year round and make it possible to understand exactly when the walruses came ashore and left it, how many animals were there. However, the problem is that camera traps take tens of thousands of pictures in just one season. Disassembling and analyzing such a huge amount of data is an extremely time-consuming task. It was decided to attract modern technologies to help people. The neural network is able to quickly "digest" the entire volume, leave only photos where there are pinnipeds and sort them, thereby saving hundreds of man-hours for the reserve team.

An even more ambitious task is to teach the computer to recognize the identity of the walrus. For this, an initial catalog is now being created - a photo base of already met animals - on the basis of which the neural network will learn. Each walrus has its own special features (experts call them “marker points”) that make it individual: “wrinkles” in the area of ​​the nose, eyes, scars, scratches, spots on the skin, chips and cracks on the fangs. As a result of their analysis, the program will be able to "recognize" each individual and report on its previous visits to the rookery.
Examples of walrus ID tables
Yulia Bogomolova / WWF-Russia
“For example, last year we saw a walrus at the rookery, which came once a week. And it turned out just during the analysis of photographs for the catalog. We have been tracking another male on Matveyev for 4 years in a row - since 2017,” says Yulia Bogomolova, Head of the Research department of the Nenetsky reserve. - I hope that such identification will also help us monitor changes in the physical state of the animal - wether the walrus has grown fat or lost weight, the appearance of new scars, the external manifestation of diseases, etc. In the future, perhaps, the duration of life too."
MIPT specialists are responsible for the technical part of the project and direct work with neural networks. They have already received more than 10 thousand images to start training the program. The head of the direction Andrey Leus, PhD in Technical Sciences even personally joined the expedition this year in order to better assess the tasks and specifics of the work.

The staff of the reserve were also able to conduct traditional monitoring of the rookery - to assess how walruses resting on the reserved beaches live this year. The conclusions are optimistic: according to a general preliminary assessment, the state of walruses at the Matveyev rookery does not cause concern.

Look at these marine giants!

“The number of animals that came ashore during the period of our observations at the rookery, in general, did not differ much from what we observed in 2017,” says Yulia Bogomolova. - On some days, there were up to 700 individuals. In addition, this summer, walruses began to go out on the spit on the Golets Island, just 11 km from Matveyev. In the last 10 years, they have appeared here rarely and in very small numbers. Thus, the number of these giants in both rookeries can be estimated at almost 1000 animals."
Now experts are analyzing the data obtained at Matveev Island. Yet manually. But perhaps already in the next field season, the routine will become much less - scientists plan to conduct the first tests and an "exam" for the neural network in a few months.
The research is done within the WWF-Russia project which is part of the International Climate Initiative (IKI). The German Federal Ministry for the Environment, Nature Conservation, and Nuclear Safety (BMU) supports this initiative on the basis of a decision adopted by the German Bundestag.