This week, I had the chance to sit down with Dr Bin Wang, our Head of Machine Learning at Herd-i, for a closer look at how our technology works, what drives the tech team, and what keeps them striving for excellence. In the interest of efficiency (and a little fun), we even used AI to help draft the interview questions and final article—a nod to the ways Herd-i harnesses the power of AI to streamline complex tasks.
So here goes: Bin, who joined us from a comprehensive background in Artificial Intelligence and computer vision, has, together with the technology team, been developing the machine learning models that power our Herd-i Lameness system and soon-to-be-released automated Body Condition Scoring (BCS) System. Together, these clever pieces of software offer farmers new ways to monitor herd health with accuracy and ease.
Bin’s Journey to Herd-i: A Passion for Practical AI
Originally from China, Bin moved to New Zealand in 2014, bringing extensive experience in AI and a drive to apply it to real-world challenges. With a PhD in deep learning for computer vision from Victoria University of Wellington, his academic journey was fuelled by his belief that AI could transform industries. “Prediction and detection are powerful together,” he explained.
“Prediction helps us anticipate potential health trends, while detection ensures nothing is missed.” His expertise in these two capabilities made Herd-i a natural fit.
Practical AI: Prediction and Detection in Action
Bin’s work centres on these critical capabilities: prediction and detection. In previous roles, he used prediction technology to forecast concrete quality, helping manufacturers avoid costly mistakes. Today, he brings that same focus on the practical application of AI to Herd-i, where prediction and detection help farmers make timely, data-driven actions. This combined approach enables farmers to act proactively, whether that means addressing early signs of lameness or responding to shifts in body condition over time.
How the Technology Works: Five Core Steps
While the Herd-i technology is powerful in its simplicity for farmers, the underlying technology operates through a series of complex steps:
- Detection and Tracking: Small, non-invasive cameras are installed in-shed on farm to capture video footage of each cow as it exits the race. Herd-i’s detection models identify each cow in every frame, and a tracking system links each cow’s track to its unique electronic ID. This step ensures that all data corresponds accurately to the right individual cow.
- Key Point Detection and Feature Extraction: Our system pinpoints critical features, or “key points,” for each cow and creates a complex feature construction based on movement (gait) or body condition. These features are essential for producing precise assessments.
- Ground Truth Collection: To ensure accuracy, Herd-i’s team of experts train the machine learning models using “ground truth” data collected by experts who do lameness scoring of cows in a vet portal or on-farm scoring of body condition in person. This expert scoring data helps inform Herd-i’s AI models, aligning them with industry standards and providing reliable scores. In excess of 30,000 videos of cows are scored in order to have a statistically valid data set for training the machine learning model.
- Predictive Scoring and Trend Analysis: Using the trained machine learning model, Herd-i generates scores for each cow every time they pass the in-shed camera and tracks trends over time. Accessible via a simple dashboard, this data allows farmers to respond quickly to changes in herd health.
- Ongoing Calibration and Machine Learning Optimisation: The power of AI is in the ability of the machine learning models to continue to learn and become more accurate over time as the data set grows. Ongoing scoring of images in the Lameness vet scoring portal continues to train the Lameness machine learning model and a programme of ongoing on-farm body condition scoring will be used to continuously enhance the performance of the automated BCS System over time.
Lameness and BCS: Tailoring the Process for Different Needs
The Lameness and BCS systems each use the same core process, which is adapted to different health indicators. Lameness identifies subtle changes in cow, gait, locomotion and posture, detecting early signs of lameness and assigning scores which are aligned with DairyNZ’s lameness scoring model. BCS, meanwhile, uses an overhead camera to capture video images which are used to assign a Body Condition Score based on the DairyNZ BCS scoring model, which allows farmers to monitor changes in physical health and act proactively based on patterns over time.
Making Data Usable for Farmers
One of Bin’s goals, shared by his team, is to make Herd-i’s data straightforward and actionable for farmers. “Our dashboard is about simplicity and usability,” he shared. “Farmers don’t need to see every bit of the tech behind it—they just need data they can trust and act on with confidence. We aim to provide that at a glance, whether it’s today’s lameness score or a BCS trend that might signal the need for intervention.” The dashboard is accessible across devices, enabling farmers to monitor herd health and make data-driven decisions from anywhere.
A Commitment to Continuous Improvement
When asked what he loves most about his role at Herd-i, Bin’s answer was simple: the people and the purpose. “We have an incredible team that’s collaborative and open,” he shared. “Working on technology that makes a real difference in animal wellness is deeply rewarding. New Zealand’s dairy farmers are the best in the world at what they do, and they are quick to embrace technology to help them farm more smartly”. Bin also appreciates the flexibility to push the boundaries of AI, with a constant focus on accuracy and improvement. “At Herd-i, we’re always looking for ways to enhance our models, to go from 95% to 99% accuracy. It’s that drive for excellence that makes this work so fulfilling for me.”
A big thank you to Bin for haring his insights and the passion that drives Herd-i forward. His work, and that of the technology team, remains essential to our mission of helping New Zealand’s dairy farmers care for their herds and remain the most productive dairy industry in the world.
Stay tuned for the next instalment from Herd-i.