Case
FarrowTech
Intelligent camera that uses machine learning to monitor the pig during farrowing
FarrowTech has developed an intelligent camera that can be mounted behind sows and automatically monitors farrowings around the clock, sending alerts directly to the mobile phone in case of prolonged farrowing intervals. With FarrowCam, you get an extra pair of eyes in the farrowing stable.
- The case description
FarrowTech is a 8-year-old company that develops intelligent products for pig farmers. When they started the company, they wanted to develop products that would reduce piglet mortality. In 2013, 10% of piglets were stillborn, which was a big problem among pig farmers. FarrowTech's mission was therefore to halve the number of stillbirths with their technology.
In the beginning of 2019, FarrowTech had developed a piece of hardware consisting of an intelligent camera. The camera is to be placed on the farrowing trail so it can monitor the sow during farrowing. In this way, the camera can use machine learning to detect any issues during birth and the pig farmer will be able to take action.
Hours
Hours spent in total so far
FarrowTech
Manufacturing company
Innovative products
Founded in Denmark
How we solved it
The utilization of Machine Learning was a complex task, and initially, FarrowTech did not meet the existing benchmarks. To further enhance the product, our Machine Learning specialist Mads was a part-time consultant at FarrowTech for a period of six months.
Mads' focus was on computer vision. Computer vision involves the analysis of images and videos, meaning every image captured from the video taken in the farrowing trail is analyzed. Initially, labels are used, which is the "perfect" way to run a system, meaning a person manually notes when a pig is born. This method will eventually be taken over by machine learning. Over time, machine learning creates an algorithm based on an analysis that can recognize when a pig was born and the duration between each birth.
In addition, Mads was involved in the selection of the training material that was to be used to improve the analysis. In such a project, it is important that the training material is in order for the system to function properly. Mads collaborated with the Alexandra Institute to accomplish the task.
Machine learning
FarrowCam
Happy pigs
Our responsibilites
Onsite consultants
Machine learning
Technologies
Additionally, the core of the FarrowCam system is an algorithm that analyzes video captured by a thermal camera in real-time. As the analysis is performed on the camera itself, the algorithm was developed with a strong focus on performance and was written in the highly optimized programming language C++.
The algorithm used by the FarrowCam system is a hybrid of classic Computer Vision, based on Open CV and statistical calculations, and modern Deep Learning with a custom model trained in Tensorflow and deployed to a Docker Container using TensorFlow Serving. Python scripts are utilized to compare the algorithm's runs to the corresponding manual analyses. Based on these comparisons, interesting portions of the video are automatically selected and added as future training data.
The results
With the help of Mads, FarrowTech has optimized their product and their FarrowCam now meets the standards they have set. They have created a tool that makes the work easier for pig farmers. Mads has been involved in the customization and adjustment, making the product more accurate. This means that pig farmers are able to help even more piglets than ever before.
Having a consultant allows for assistance with specific needs without committing to a long-term contract. FarrowTech required a machine learning specialist to optimize their product and had access to one for the duration they required the help.
Do you need Machine learning for your next digital project?
Machine learning enables computers to make predictions based on data, which is effective for businesses seeking to keep up with trends and patterns that cannot be detected by humans. To learn more about machine learning, please click here.