AI tested in Primorye to identify tigers and leopards

Neural networks will help scientists get information on the rarest wild cats faster.

Photo by: Leopard Land National Park

Artificial intelligence (AI)-based software for recognising Far Eastern leopards, Amur tigers and other animals has been tested in Primorsky Territory for the first time. Experts from the Moscow Institute of Physics and Technology (MIPT) worked on the project, the Russian Ministry of Natural Resources and Environment said.

The software is able to semi-automatically analyse and sort camera images by animal species.

Photographic traps allow the precise monitoring of large areas of natural areas. However, this method of observation generates a huge amount of data, which makes it difficult to process.

A neural network is able to look through vast amounts of data in a short time and with high accuracy. The test results of the AI have been praised by experts. The program checked and sorted 7.5 thousand photos in just one hour on a normal computer. In manual mode, such a task would take an entire day.

“We have more than 400 camera traps in operation, bringing in a colossal number of images, more than a million a year. For more than 10 years now, we have been continuously systematising the data by reviewing each file manually, highlighting images of tigers, leopards and many other species, which is necessary, among other things, to establish the number of the world’s rarest cats every year.

Machine vision will simplify and speed up the routine process, allowing us to concentrate on more complex tasks,” said expert Taisia Marchenkova.

As Andrei Leus, lead programmer of the MIPT laboratory, explained, first the experts taught the AI to filter the objects, then to classify them. The next step was to identify what appeared in the images. This is a more difficult task, as some animals are difficult to distinguish from one another. In particular, the neural network does not recognise brown bears “by sight” very well.

With representatives of large felines, the algorithms are somewhat easier to work with, as each individual has its own unique colouration.

Conducting research is important for the conservation of the Far Eastern leopard and tiger. Other forest-dwelling animals, on which the state of predator populations directly depends, also need to be monitored. The condition of deer, roe deer, badgers, raccoon dogs and other animals are also monitored using photographs.

Cover photo: Leopard Land National Park