In this entry, I’ll summarise the technical choices done for the project Genes2Behave at Norsvin. It might be technical in some parts, and not so technical in others. Just feel free to contact me LinkedIn if you would like to discuss this further.

Research project: Genes2Behave

G2B

Figure 1. Schematic of the Genes2Behave project


Phenotype extraction

Figure 2. Phenotpe extraction - two parallel routes


Video acquisition

In total 18 cameras were installed, with a range of brand (Hikvision, Lorex), geometry (top/down, angle from the wall) and resolution (1920x1080)

Pig detection

Annotated Dataset

Model architecture: YOLOv8

Training metrics

Inference

Data model

Tracking animals

Algorithm choice: StrongSORT

Changes to base StrongSORT

Tracking metrics

Identifying the animals

Using feed data to identify animals

Using eartags to identify animals

Using both

Conclusion