Friday, April 12, 2024

AI can identify up to 200 species of birds from just a photo

Artificial intelligence is already helping scientists in various fields, from medicine to material science. Now, ornithology can be added to this list. If you are professionally involved in bird recognition based on photos, know that there is already an artificial intelligence in this world.

Researchers at Duke University have created a machine-learning algorithm that can identify up to 200 species of birds from just a photo. The technology works in much the same way as to face recognition; that is, it searches for certain features in the images and, based on them, gives an answer.

However, there is one important difference – this algorithm informs researchers about the progress of its conclusions. For example, identifying the bird in the photo as a barn swallow, it highlights the distinctive features of this species – long wings, a forked tail and a red spot on the throat.

To train the algorithm, researchers fed it 11,788 photos of 200 bird species, ranging from swimming ducks to hovering hummingbirds. As a result, it learned to correctly identify these species with an accuracy of 84%, a Duke press release explained.

Note that the ability to recognize 200 species of birds is a very modest achievement, considering that at present, about 10,000 species of birds live on Earth. At the same time, the average beginner ornithologist will take a lot of time to acquire even such a limited skill, while the algorithm acquired it literally in a few hours.

The authors admit that their main task was not to create an AI ornithologist at all. Human scientists also do a good job of identifying birds. However, birds are an ideal example to try out new types of algorithms that will later be used in medicine. They will not only accurately identify diseases, but will also inform doctors and patients how they came to this or that conclusion.