Norway tests a "digital shepherd." The future of reindeer herding
Researchers in Norway are testing the use of drones and artificial intelligence to count reindeer. The project is being carried out on behalf of the Norwegian Agriculture Agency. NIBIO, NORCE, and the company Biodrone are participating in the work, in cooperation with reindeer herders. The goal is to simplify and improve the accuracy of counting, both for the administration and herd owners.
The aim is to simplify and increase the precision of herd counts.
Fot. Fotolia
Drones are already widely used in reindeer husbandry. The devices help locate scattered herds in mountainous terrain. They also allow for checking fences, enclosures, and finding animals in hard-to-reach places.
In some cases, drones are also used to move reindeer. The use of technology reduces the need for helicopters, snowmobiles, and ATVs.
In some cases, drones are also used to move reindeer. The use of technology reduces the need for helicopters, snowmobiles, and ATVs.
Training the algorithm in variable conditions
As part of the ARC – Aerial Reindeer Count – project, scientists have been collecting drone photos and videos over the past year. The material covered different seasons and types of terrain. The data was used to train an artificial intelligence model to automatically recognize and count reindeer.
The process proved challenging due to the variable appearance of the animals. Reindeer differ in coat color, the presence of antlers, and the background on which they are found.
The process proved challenging due to the variable appearance of the animals. Reindeer differ in coat color, the presence of antlers, and the background on which they are found.
Large rocks in summer are sometimes mistakenly recognized as animals.Photo: Diego Delso, CC BY-SA 4.0, via Wikimedia Commons
Model effectiveness and further development
The current version of the model works best on snow and in open areas. In such conditions, it recognizes about 90 percent of animals. Difficulties arise in dense vegetation and when distinguishing white reindeer from snow.
By comparison, experienced observers achieve an accuracy of over 99 percent, but analyzing the photos takes them much more time. Researchers are constantly improving the model, relying on manually labeled data.
By comparison, experienced observers achieve an accuracy of over 99 percent, but analyzing the photos takes them much more time. Researchers are constantly improving the model, relying on manually labeled data.
Instructions, platform, and further applications
At the same time, a manual for drone operators has been developed. It contains recommendations on flight altitude and camera angle. The best results are achieved when flying at an altitude of 40 to 120 meters and an angle of 45 to 90 degrees.
The next stage of the project will be to make the technology available to herders through a public platform planned for launch and testing in 2026. Work is also underway on the use of thermal drones, which may enable reindeer counting in winter, in low-visibility conditions.
The next stage of the project will be to make the technology available to herders through a public platform planned for launch and testing in 2026. Work is also underway on the use of thermal drones, which may enable reindeer counting in winter, in low-visibility conditions.
How do you rate this article?