Yandex Cloud and students of SHAD will help scientists explore the Red Book snow leopards using neural networks

2023: Completion of the first phase of wildlife monitoring

Yandex Cloud, with the support of the Yandex School of Data Analysis (SHAD), completed the first stage of the wildlife monitoring project for the Saylyugemsky National Park in the Altai Republic. Together with students of the SHAD, experts have developed a neural network that recognizes 10 different species of animals in images and videos obtained using camera traps on the territory of the national park. Among them are irbises, or snow leopards included in the Red Book. The neural network finds animals in the images tens of times faster than a person. The technology will help scientists more effectively study irbis in their natural habitat and understand how to protect their population. Yandex Cloud announced this on October 23, 2023.

According to experts, no more than 90 irbis live in Russia as of October 2023, but their exact number is unknown. In Altai, researchers found 44 individuals of which 24 live on the territory of the Saylyugemsky National Park (118 thousand hectares).

There is a sign that only a lucky person can meet a snow leopard. This is no coincidence: irbises are still little studied, since they live in hard-to-reach places, lead a twilight lifestyle, hunt one by one. We investigate the habits, movement, number of these animals using camera traps with motion sensors: when an animal passes next to such a device, photo or video shooting is turned on. Then the data from the devices are downloaded to the computer, and the park staff watch them. Previously, it took days and even weeks to find rare animals among thousands of blank images, "said Denis Malikov, director of the Saylyugem National Park.

Students of the SHAD, together with Yandex Cloud specialists, have developed a neural network that processes such images in just a few seconds and finds animals on them. Specialists trained the model on an array of 40 thousand marked objects: photos and video fragments, with a total volume of 150 GB. The data was uploaded to the Yandex Cloud object storage, and the neural network was trained using the Yandex DataSphere service. The program highlights animals in the image, classifies them into a certain type, and then uses this information to distribute the pictures into folders. In some cases, the model indicates several options that it "sees." At the next stage, the neural network will learn to identify specific individuals of leopards in the images by their appearance. So scientists can save resources, strength and more efficiently study and preserve snow leopards and other animals.

We began work with the national park in May 2023, the project will be implemented in several stages. At the next stage, we want to train the model not only to recognize snow leopards, but to distinguish one irbis from another, defining a unique pattern on their hair. In addition, we are studying the possibility of adding advanced data analysis and visualization to the solution: scientists will be able to collect statistics about the inhabitants of the park, build a map of their favorite places, and study the routes of their movement. All this will help enrich knowledge about these beautiful, but rare animals, and think about how to protect them from extinction, "said Anna Lemyakina, director of national strategic projects at Yandex Cloud.