A team of professors and students from the UC Santa Barbara Department of Computer Science has developed a system designed to aid ecological research by processing large quantities of digital photographs and identifying which, if any, include images of animals.
The testing grounds for the new system is Sedgwick Reserve Ranch, one of the 39 sites that make up the UC Natural Reserve System. The UC Natural Reserve System aims to “contribute to the understanding and wise stewardship of the Earth and its natural systems by supporting university-level teaching, research, and public service at protected natural areas throughout California,” according to the mission statement found on the reserve’s website.
Located in the Santa Ynez Valley, the Sedgwick Reserve is home to a variety of wildlife, including bears, coyotes, and deer. The project being tested there, called “Where’s the Bear?,” goes further than determining whether or not an animal is present, as the system can also identify what kind of animal is in the photograph.
In an interview with the UCSB Current, Professor Chandra Krintz, one of the two professors involved in the project, commented on its impact: “This is hugely powerful technology and we want to bring it to bear on important problems. It saves tons of time — what once took 14 days we can now do in 3 hours — and it saves money, communications, energy, carbon footprint.”
The process begins with digital cameras that are placed throughout Sedgwick Reserve Ranch. Although the pictures taken are limited to those with motion in the frame, the set of images captured would almost certainly have included many which were not relevant to the ecologists’ research.
These scientists would previously have spent countless hours classifying the images manually. In order to optimize the image sets as research material, the sets would ideally be pre-classified by a computer.
In order to automate image classification, the “Where’s the Bear?” team used machine learning, a process described by the SAS Institute as “a branch of artificial intelligence based on the idea that systems can learn from data, identify patterns, and make decisions with minimal human intervention.”
In this case, the method used to automate the image classifying process was to “teach” a computer system to recognize what the images of animals would look like. The “Where’s the Bear?” team created thousands of fake photographs, in which images of bears were placed onto “empty” images of the part of the reserve which would be photographed. These images were then used in as the “training” set for the system.
During her interview with the Current, Krintz also stated how accurate this system is in identifying particular animals. “We don’t get any coyote wrong. We don’t get any bears wrong. We get about 12 percent error on deer — there are lots of deer — and we are trying to improve on that,” Krintz said.
A presentation published by the Department of Computer Science indicates that this work will result in the conservation of resources and may assist in future research. Decreasing the amount of network use required by image transfer reportedly saves energy.
This presentation, which details the methodology and results of the “Where’s the Bear?” project, also lists several additional functions which may be developed in the future, including identifying features, counting, and distinguishing empty images from those depicting small animals.
UCSB Computer Science Professors Chandra Krintz and Rich Wolski lead the development of the system, with students Andy Rosales Elias and Nevena Golubovic also contributing as members of the team, according to project documentation available on the UCSB Computer Science website. Additional collaborators to this work include Sedgwick Reserve Resident Director Kate McCurdy, as well as volunteer research assistant Grant Canova-Parker.