Wildlife researchers at Oregon State University (OSU) have invented incredible success. They are increasing the artificial intelligence (AI) technology developed for wildlife species identification to automatically sort animal species photographed on trail cams. Per undergraduate student Owen Okuley, who spearheaded this pioneering research, under the mentorship of research associate Christina Aiello, he aims to improve AI performance in wildlife monitoring through species and environment-specific training. The paper, which appeared in the journal Ecological Informatics, addresses an important issue. Specifically, it looks to automate the tedious work of reviewing thousands of images from motion-activated wildlife cameras.
Through this research, scientists deployed camera traps along water holes across the Mojave and Colorado Deserts in Southern California. Okuley and Aiello connected and collaborated through the Fisheries and Wildlife Undergraduate Mentoring Program. Their study discovered that by honing an AI model’s training on just one species versus trying to identify every single species accelerated identification accuracy tremendously. By supplementing the training with images from other local environments, they were able to improve AI model performance even more.
The Journey of Research and Collaboration
Owen Okuley is looking forward to beginning a Ph.D. in ecology and environmental biology at University of Texas at El Paso.
Hands-on work
He really values the hands-on experience that came from doing this project.
“Being able to tackle a project from start to finish has allowed me to grow immensely as a scientist.” – Owen S. Okuley
Okuley developed these scientific skills during the course of this research project. He got to develop some key relationships with co-authors and mentors at every step. He engaged in various aspects of research that many undergraduates seldom encounter, such as conceptualization, grant writing, and publication processes.
“I was able to not only make strong connections with my co-authors and mentors but got to engage with the aspects of research most undergrads never see,” – Owen S. Okuley
The joint work of Okuley and Aiello is a perfect example of the rewarding results that can come from mentorship programs. Both Aiello and Bansal focused on the importance of creativity in developing training datasets to improve AI precision.
“Owen is exploring ways to curate training datasets so that we improve AI accuracy faster, with less data, which I think is a much-needed shift in how our field uses AI.” – Christina Aiello
Overcoming Challenges in Wildlife Monitoring
The project aimed to address the existing challenges that scientists face while analyzing terabytes of images from camera traps. Its goal was to streamline this process and speed it up. Traditional approaches can be cost and labor-intensive, resulting in delays on the ground where important wildlife monitoring work needs to be done. Today’s AI models tend to lack accuracy, especially for image classifications coming from new, unseen geographies.
Aiello pointed out a significant challenge in utilizing AI for wildlife research:
“One of the biggest problems in using AI in wildlife research is limited accuracy when we use the model to classify images at a novel location—one the model has never ‘seen’ before.” – Christina Aiello
To solve this dilemma, Okuley and Aiello first honed in on the training process itself. They grew wary of the data they were inputting into the models. And boy did this strategy work miracles. By narrowing their goals and training on a wide variety of images, they achieved astounding identification accuracy results of almost 90% with only 10,000 training images—a tiny fraction of what comparable AI models require.
“By narrowing objectives while still ensuring training data variety, we achieved almost 90% identification accuracy with a small fraction of the training data,” – Owen S. Okuley
Impacts on Future Research and Wildlife Conservation
The impacts of this research go beyond educational attainment. Its potential carries significant positive impacts for wildlife education and outreach efforts too. Making AI models better while requiring less training data can greatly reduce computing power requirements. This is in addition to the fact that this reduction brings about a 23 percent decrease in energy use.
Okuley pointed out that these advancements really serve researchers well. Beyond the fiscal aspect, they afford unique benefits to the flora and fauna she is studying.
“And fewer images means a model requires less computing power and less energy, both of which are beneficial to the wildlife we seek to study.” – Owen S. Okuley
Okuley’s passionate about creating AI models to identify specific characteristics of waterfowl. True battery-free operation and the need for infrared detection is more than just green marketing. This legislative focus is indicative of a broader dedication to progress innovative technology with the potential to improve understanding and management of wildlife populations.