A polar bear with two cubs in the Wrangel Island Nature Reserve.
Monitoring in Russian nature reserves
MIPT scientists, together with the Russian Ministry of Natural Resources and Environment, are actively implementing AI (artificial intelligence) technologies to monitor biodiversity in specially protected natural territories of the Russian Federation.
The Russian Ministry of Natural Resources and Environment manages 107 nature reserves, 66 national parks, and 63 wildlife sanctuaries, and each of these territories has from several dozen to hundreds of camera traps. Analyzing the data obtained requires viewing thousands of photos, which is difficult for humans, but can be easily solved using vision software based on trained neural networks.
“Each reserve has its own tasks: in the Land of the Leopard National Park and the Sikhote-Alinsky Reserve in Primorsky Krai, we teach AI to distinguish between felines, on Matveev Island in the Barents Sea — to keep records of walruses, in the Central Forest Reserve – to monitor the bear population, and to catch poachers — in Sebezhsky and Sayano-Shushensky the park. There are nuances everywhere,” said Andrey Leus, Associate Professor at MIPT, about the project.
Automatic processing of photographic materials allows you to track the population of wild animals, keep records of them, and find intruders in hard-to-reach places. At first, scientists taught the AI to simply sort the photos. After organizing the photo archive, a new task was born — to determine which animal is in the photo. As a result, the program already “knows” more than 30 species of wild fauna.
“First, we taught the AI to filter objects and then classify them. The next task is identification. This is a more difficult level, as not all animal species have strong individual differences. For example, it is quite difficult for AI to distinguish brown bears.
It is much easier to teach how to distinguish between animals that have their own unique pattern that the neural network “remembers”. As a result, we record the vital activity of each individual without disturbing its habitat. And we are collecting a large database for AI training.”
Identification of walruses
Walruses in the aerial image (above) and walruses marked with a neural network (below).
Walruses in an aerial image processed by a neural network. Neural networks are increasingly used every year in the research of various species of marine mammals. Atlantic walruses are excellent for testing technology in this regard: they are a small and largely isolated group.
The task of identifying walruses was also faced by developers on Matveev Island in the Nenets Autonomous Okrug. What is the difference between walruses? It turned out that they have a very specific set of scars and chips on their fangs.
“A solution to the walrus accounting problem is being tested on Matveev Island. Here, data is collected from aerial photographs. Ecologists need to recalculate walruses quite often — the population size is directly related to the environmental situation. We trained AI in the nature reserve on the island, there are not very many walruses there: 1000-1500 in the rookery.
These are Atlantic walruses, and they are listed in the Red Book. If we study the Pacific ones, there can be up to 15,000 of them at a time. Accurate daily calculation of such a quantity is a difficult task for a person, but it is quite feasible for a neural network,” said Andrey Leus.
Neural networks are increasingly used every year in the research of various species of marine mammals. Atlantic walruses are excellent for testing technology in this regard: they are a small and largely isolated group.
The Central Forest Reserve is studying the bear population, and although the neural network cannot yet distinguish between individuals, it is able to track their movement. Thanks to the counting methods developed by biologists, the program not only studies their migration, but also predicts the development of the population: growth or decline.
Poachers are a big problem
Do you see the poacher in the left picture? And the neural network (on the right) sees!
In the Sebezhsky and Sayano-Shushensky nature reserves, in addition to accounting and studying animals, there is a big problem — poachers. In this case, the neural network is able to find people among thousands of photos. With the help of a drone, you can record the fact of illegal fishing or hunting, and the neural network selects photos of violations, while the AI easily recognizes the protective camouflage.
But new tasks are already forming ahead, and one of them is the search for Arctic bears. Finding a polar bear in the white snow is not an easy task, and it is even more complicated by the fact that the territory of bear migration is quite extensive, and there are few pictures with real predators. MIPT scientists have proposed using synthetic bears to train a neural network.
Of course, such large-scale tasks require a lot of resources, and this can be easily solved by attracting young specialists. To this end, the Russia — Land of Opportunity platform, together with MIPT, has been holding large-scale Digital Breakthrough hackathons for several times, where participants solve practical problems.
The Ministry of Natural Resources and Environment of Russia attracts both employees of nature reserves and representatives of scientific and public organizations engaged in environmental protection activities as industry experts. Hackathon participants often offer interesting alternative solutions, and the winners, in addition to monetary rewards, can find a real application for their ideas.
In May, the Ministry of Natural Resources and Environment jointly with MIPT held a hackathon in Khabarovsk on the topic “AI on guard of the Nenets walrus population.” By the end of the year, the Russian Ministry of Natural Resources and Environment is planning two more hackathons, where they will consider the topic of optimal methods for searching for polar bears in the Arctic and monitoring economic activities based on remote sensing materials.
MIPT Press Service
