Since then through the evolution of computing power which has grown exponentially, a new generation of researcher is using huge and complex datasets and a combination of off-the-shelf AI products in an agricultural context, and in keeping with our pioneering past, creating our own machine learning algorithms to make sense and order of data. All in a farming context.
AgResearch harnesses AI for advanced agricultural research
AgResearch was an early adopter of Artificial Intelligence (AI). Our scientists twenty years ago were programming computers, and writing bespoke code, to harness what were then the cutting edge digital technologies of the day. They were computing what were then considered large datasets.
Digital Agriculture research Associate Kevan Cote is part of the new breed.
Kevan and colleagues are currently working on a collaboration with the AI Institute at Waikato that has propelled our practical applications of AI with our expertise in animal health and welfare. Alongside primary collaborators Peter Reutemann and Dale Fletcher, this partnership has delved into the realms of Explainable AI, that aims to demystify the inner workings of neural networks, paving the way for more transparent and understandable AI systems. This research has far-reaching implications, particularly in fields like agriculture, where AI can revolutionize traditional methodologies.
One such application is in animal behaviour research, an area historically reliant on manual observation and labour-intensive data collection methods. Thanks to AI, we're streamlining these processes, automating tasks that were once time-consuming and prone to human error.
Take, for example, our work with goat animal welfare. Traditionally, researchers would spend months recording video footage and manually coding specific behaviours. Now, with the help of AI, we're automating these tasks, freeing up valuable time and resources for more in-depth analysis.
Kevan said he began with a focus on identifying goat kids in large penned areas, using surveillance cameras, lighting equipment and obstacles that act as stimulus for the goats. Previously researchers would monitor and manually code behaviours, however, with the use of AI products such as DeepLabCut, which takes data points from the footage (see image) and detects a skeleton over the animals the researchers are able to automative the analysis of specific behaviours, such as walking or eating, based on joint and head positions.
By taking a small percentage of the work and feeding it into an algorithm, a large portion of the work can be automated, reducing the overall manual effort. The process involves more upfront work but the result is easier data handling and better flow, even though the data itself is becoming more challenging as researchers increase the complexity of their inquiries.
The expressive nature of goats presented unique challenges, leading us to expand our projects to include sheep and pigs. “Through the application of AI techniques, we're tracking behaviours and analysing data with unprecedented accuracy and efficiency. But AI isn't without its challenges. We've encountered the "Blueberry Muffin Conundrum," where AI struggles to distinguish between similar patterns, such as a goat hoof and a point on a feeder. However, through collaboration with the AI Institute and continuous refinement of our algorithms, we're making significant strides in overcoming these obstacles.”
“By involving data scientists from the outset of our projects, we're able to add value and streamline future analysis. Our recent internal seminars on AI have been met with enthusiasm, showcasing a growing appreciation for the potential of AI in enhancing research capabilities.”
Moving forward, we're committed to pushing the boundaries of what's possible with AI in agriculture. With the continued support of our partners at the AI Institute and our dedicated team of researchers, we're confident that AI will play a pivotal role in shaping the future of agriculture.”
Kevan added that working with collaborators that often collect and store data in their own unique way often creates issues of compatibility. However, AI has an answer for that too. Many scientists are using software development practices and tools (Docker) that essentially homogenise different data, making apples and oranges the same thing.
“Docker helps create shareable and consistent environments, which is crucial when collaborating with external partners. This tool addresses compatibility issues by allowing projects to be packaged into containers that can be easily shared and run across different systems.
“Docker also ensures long-term usability of projects. We can revisit projects from years ago without worrying about updates or changes that might disrupt functionality. This has helped shift our focus towards leveraging AI to enhance scientific endeavours rather than treating it as a separate entity.
“Many scientists now perceive the potential disruption of AI as an opportunity to further develop their research rather than as a hindrance. “