When conservation officer Erin Nash returned to the office with 73 field cameras from the Otways earlier this year, she was faced with a monumental task: sorting and analysing a staggering 39,000 images. These cameras had been deployed to monitor the impact of the summer bushfires on native animals in the region, including crucial populations of potoroos, bandicoots, and swamp antechinus.
"Our primary objective is to identify pockets of threatened species that have managed to survive the fires," explained Dr. Nash. "This allows us to strategically deploy our feral animal control programs, giving these vulnerable creatures the best possible chance of survival."
Traditionally, manually sifting through tens of thousands of images would consume weeks of valuable time. However, thanks to a groundbreaking new artificial intelligence tool, Dr. Nash was able to identify a significant issue within a matter of hours.
"We've observed that areas previously dense with vegetation acted as natural barriers, preventing the ingress of deer and pigs into some of our national parks," she stated. "This AI allowed me to rapidly detect the threat of red deer encroaching into burnt areas. We've already passed this crucial intelligence to deer control contractors."
This rapid insight is attributed to an innovative AI tool developed collaboratively by Dr. Nash and her Parks Victoria colleague, Mary Thorpe, alongside Dutch data scientist and world-leading expert in automated wildlife recognition, Peter van Lunteren.
AI Revolutionises Wildlife Monitoring
The species recognition model employs high-speed artificial intelligence, capable of processing an impressive 20 images per second. Crucially, it can identify animals with an accuracy rate exceeding 95 per cent.
"This is an absolute game-changer for us," Dr. Nash enthused. "It translates directly into significant savings in both time and money, resources that can now be reinvested directly into vital conservation efforts."
Getting Rangers Back to the Field
Parks Victoria rangers deploy thousands of cameras across the state annually. These devices are essential for monitoring vulnerable and endangered species, as well as detecting and responding to threats posed by feral animals. However, the motion-detection technology isn't always precise, often triggering on environmental factors like falling leaves.
Dr. Nash highlighted that these cameras can generate millions of images each year, all of which require careful examination to determine if they contain any identifiable animals. This process has historically led to substantial backlogs. While other AI recognition programs exist, they have not been specifically trained on Australian fauna.
Peter van Lunteren began developing the new AI system for Parks Victoria during a year-long backpacking trip around Australia, famously living out of a van. "When an AI model can handle the bulk sorting, ecologists are freed up to focus on their core strengths: interpreting results, making informed decisions, and conducting essential fieldwork," he commented. "My aspiration is for every region to eventually have a model trained on its unique local species and environmental conditions."
To train this sophisticated system, five million images were contributed by 18 Victorian research and conservation organisations. These institutions, including Zoos Victoria, Phillip Island Nature Parks, Dja Dja Wurrung Clans Aboriginal Corporation, the University of Melbourne, and Deakin University, all utilise remote wildlife cameras, commonly known as "camera traps." The system is now proficient in identifying over 200 native and feral species.
"The field of camera trap AI is remarkably collaborative," Mr. van Lunteren observed. "It's a genuinely positive environment to work in, characterised by widespread friendliness and a willingness to share models and data. It's truly inspiring."
Faster Responses, Better Protection
Dr. Nash confirmed that the tool, accessible via the Parks Victoria website, significantly accelerates the organisation's ability to respond to threats and other critical issues. "Previously, images of feral pigs might have sat on hard drives for months, awaiting analysis," she explained. "By the time you locate a pig amidst hundreds of thousands of images of, say, swamp wallabies, it's often too late to take effective action."
Mark Norman, Chief Conservation Scientist at Parks Victoria, noted that the system is particularly adept at capturing images of nocturnal species, animals that are notoriously difficult to locate due to their evasive nature, and wildlife that is expertly camouflaged in the bush.
"It's superior to the human eye," Dr. Norman stated. He then referred to a specific camera trap image, explaining, "There's an animal in this photograph that none of us would have spotted without this program."
The "trap" itself is a pole featuring three spoons containing peanut butter, oats, and honey, designed to entice animals towards the camera.
"This is what the computer program identified as an animal," Dr. Norman said, indicating a detection frame in the second image. "Based on those subtle markings, it allowed us to confidently identify it as an eastern ground parrot, one of our critically endangered parrot species."
"When you consider the effort involved for someone to manually review 10,000 slides over three weeks just to find one image, this advancement is monumental."
Dr. Norman anticipates that rangers will now have substantially more time to dedicate to implementing threatened species programs and undertaking feral animal control measures. "People's hair will likely turn from grey back to its natural colour, considering the immense strain of the previous workload," he quipped.
Looking ahead, Dr. Norman expressed hope for further development, particularly the ability to provide instant notifications to conservation teams, especially when predators are detected.
Real-World Application: Zoos Victoria's Success
Zoos Victoria has already integrated the AI model into its operations, actively putting Dr. Norman's vision into practice within its Lowland Leadbeater's possum safe haven program at Haining Farm in the Yarra Valley.
Arabella Eyre, acting general manager of the zoo's Threatened Species team, explained how attaching a mini-computer to field cameras allows for immediate text message alerts if a cat or fox enters the protected area. "This enables us to respond swiftly and remove the predator from the fenced areas," Ms. Eyre said. "It will significantly expedite our response times, enhancing our protection of threatened species that would otherwise be at high risk of predation."
Ms. Eyre described the model as "top of the line" for recognising species endemic to south-east Australia. However, she also sees potential for further improvement with advancements in camera technology. "It's nearly perfect with our current technology," she admitted. "The next logical step is to explore how we can develop cameras that capture higher-quality images. Better imagery will, in turn, allow us to refine these AI models even further."
Open Source: A Boon for Conservation
In a significant move, Parks Victoria has officially made the AI recognition model available to the broader community as an open-source program.
Dr. Nash pointed out the limitations of pre-existing AI systems, which are often proprietary and operated by large tech corporations. "They then retain ownership of your data," she noted. "Many researchers and land managers prefer to maintain control over their underlying data."
Ms. Eyre underscored the importance of open-source for Zoos Victoria, stating it mitigates the risk of sensitive conservation information falling into the wrong hands. "[The Parks Victoria model] is not only open source but also runs locally on your computer rather than in the cloud, which significantly enhances data security. This is particularly crucial for our threatened species or those at risk of poaching," she explained.
Peter van Lunteren confirmed that the model is offered completely free of charge. "It's open source with no licensing costs, your data remains private, and you run it on your own computer," he confirmed. "You retain complete control."
No comments:
Post a Comment