Researchers from Germany, in partnership with staff from a Norwegian university, trained a neural network to identify beavers. To recognize an animal, it is necessary to take into account not the shape of the tail, but the individual pattern on its upper surface, consisting of scales.
In fact, scientists have found a non-invasive (no tissue sampling or surgery) method for identifying individual animals using photographs. The beaver’s tail is the main organ for moving in water. Instead of wool, it is covered with dense scales located over its entire surface. These scales create a special pattern with a unique structure. Such patterns are never repeated, just as a person’s fingerprint is never repeated. The researchers found that this information could be used by AI to recognize individuals.
To identify the animals, the authors used the scale-invariant feature transformation (SIFT) algorithm, which makes it possible to determine the pattern on the tail even if the image is highly distorted due to rotation, blurriness, or the absence of some elements on it.
The researchers trained the neural network on 800 tail photographs of 100 individual Eurasian beavers obtained through Norway’s animal tracking program. All photos were taken with a DSLR camera using one lens.
The new algorithm fully justified itself – with its help, the neural network identified beavers in 95.7% of cases. This method opens the door to ecological research using non-invasive methods for recognizing and identifying animals. In fact, we are talking about semi-automatic evaluation of large photo databases without the need to catch and band animals.