Using the 3.37 million images, the computer model was used to assess some 375,000 animal images. The processing of these was rapid, at rate of about 2,000 images per minute. The program was run on a standard laptop computer. When compared with the identifications made by scientists, the artificial intelligence achieved 97.6 percent accuracy and at a pace far faster than any human could hope to match. The program therefore serves a practical use for wildlife image classification.
The artificial intelligence has been built around Program R, which is a widely used programming language and available as free software, designed for statistical computing. The R language is widely used among statisticians and data miners for developing statistical software and data analysis.
The images were captured by a camera-trap, with cameras situated across the U.S. Earlier work, from the same research group, used 3.2 million images that were collected in Africa (primarily Tanzania) by a citizen science project called Snapshot Serengeti.
With the initial research, lead scientist 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.”
He added: “This will dramatically improve our ability to both study and conserve wildlife and precious ecosystems.”
To develop the software, the researchers trained a deep neural network based on Mount Moran, a high-performance computer cluster located at the University of Wyoming. The v was additionally used to review an image data subset of 5,900 images of moose, cattle, elk and wild pigs from Canada. This delivered an accuracy rate of 81.8 percent.
The results have been published in the journal Methods in Ecology and Evolution. The research paper is titled “Machine learning to classify animal species in camera trap images: Applications in ecology.”