How Our Sensors Work

Our innovation makes the impossible possible.

Our system uses two pieces of technology working together: an incredible optical sensor with cutting-edge computer vision and deep-learning algorithms.

In-line optical sensor technology is at the core of our innovative platform. We use light scattering for rapid, on-site determinations of compounds in raw milk. Our monitoring equipment fits into existing milking lines, making it accessible and cost-effective for dairy producers. Our system does not require the use of any chemicals, consumables, or reagents meaning that the milk used to take measurements is unaltered and can be returned to the milking line after measurement. This reduces operational costs and the use of chemicals inside the dairy barn.

What is light scattering?

Our product relies on the principle of light scattering: a beam of directed light passes through a medium, in our case raw milk, and hits small particles that cause the beam to change direction and scatter. During this process, a portion of the light may be absorbed or reflected, altering the intensity of the scattered beam. Particles of different sizes, in various concentrations, will have unique scattering patterns that can be observed. 

Until our technology, the application of forward-angle light scattering in fluids has been limited to specific applications and too costly for non-laboratory uses. With our sensor, it is now possible to detect several compounds across a wide range of particulate sizes, using a cost-effective sensor system.

Predicting trends with deep learning and neural nets.

Artificial neural networks use algorithms, big data, and the computational power of graphic processing units (GPUs) to find patterns and connections between data sets. This branch of machine learning enables computers to achieve complex calculations and learn at speed, accuracy, and scale that has not been possible before. 

We use computer vision and deep neural networks to build robust algorithms to predict the presence and concentrations of herd health indicators in raw milk. We have collected thousands of milk samples to train a predictive model. This data set has yielded algorithms that act as the baseline of prediction, while additional data sets will allow us to improve the predictiveness, accuracy, and consistency across a wide range of use cases.

This approach has enabled us to develop a series of models to detect somatic cell count and reproduction status, with future implications for fat and protein content. This makes it possible to diagnose dairy diseases, including sub-clinical and clinical mastitis, and improve reproductive performance.