First pilot – First algorithm

The development of the first prototype of our PigBrother pig weight estimation product has been carried out in collaboration withh Hardfarm Kft. The conclusions of the research we conducted in-house were only theoretical assumptions - we wanted to gain practical experience and market validation as well.

The location of the first cameras raised a lot of questions:

• - What type of cameras would be suitable for photographing the animals?
• What kind of hardware should we use?
• Where exactly should we place the cameras?
• - What time of the day would be most suitable for photography , and how often should we take them?

Answering these questions took a long time, significant testing and learning effort had been made until we succeeded starting taking pictures continuously.

Once the images had been recorded and kept coming in, our first task was to process the visual material and educate the AI: the system needed to learn how to analyze pictures, which part of the image is useful – in other words the system had to learn to identify pixels that belong to a single pig – and which part of the image is of no use e.g., the floor, the feeder or the wall of the pen, etc.

Our technology had successfully learnt to recognize individual porkers, however, weight measurement was still a long way to go. During the experimental teaching process, we measured the porkers several times on physical scales and painted individual signs on their back for identification – this is how we knew which weight figure belonged to which porker.

The data set we used for teaching the AI was updated by the weight measurement data which was gradually expanded with the results of further experimental measurements.

As a result of this process, we successfully developped our first weight measurement algorithm that was able to measure, with the help of machine learning, pigs at a weight of 25-120 kg.