Achieving this level of accuracy was not straightforward. Deep neural networks, a form of computational intelligence that mirrors how the human brains sees and understands the world, require considerable amounts of training data to work effectively. In the case of the recent trial the software was required to count, label, and describe an array of animal images.
The images used were obtained from Snapshot Serengeti, a citizen science project. Here motion-sensor cameras located in Tanzania collected millions of images of animals in their natural habitat, including lions, leopards, cheetahs and elephants. The images had been labelled by 50,000 human volunteers over the course of several years.
The artificial intelligence was able to screen 3.2 million images in a matter of weeks and determine which of 48 different species of animal was present in a given image, and also the number of each animal and also the activity being performed, such as eating, sleeping, moving and so on.
Discussing the development, lead researcher Professor Jeff Clune said: “This technology lets us accurately, unobtrusively and inexpensively collect wildlife data, which could help catalyze the transformation of many fields of ecology, wildlife biology, zoology, conservation biology and animal behavior into ‘big data’ sciences.”
In terms of future applications, he adds: “This will dramatically improve our ability to both study and conserve wildlife and precious ecosystems.”
The research findings have been published in the journal Proceedings of the National Academy of Sciences. The research paper is titled “Automatically identifying, counting, and describing wild animals in camera-trap images with deep learning.”