Agriculture & Biotech
Computer vision that classifies insects with 99.9% accuracy
VLAIO-funded project: a computer vision model that identifies the sex of flies for the animal feed industry. Over 99.9% accuracy.
Objective
Automatically classify insects for optimised breeding
For a VLAIO-funded project, we were consulted on the AI component of an 'insect separator'. The client was building a device to separate a fly population into males and females for optimised breeding. The eggs are used in the animal feed industry.
The hardware was built by an engineering firm. The AI challenge: identify the sex based on photos automatically taken in the sorting box. Speed and accuracy were critical to make the process work at scale.
Our approach
Train a computer vision model on real sorting images
Image collection and labelling
We collected images from the actual sorting hardware. Each image was labelled by the client's biologist. The dataset was built from real production conditions, not from lab settings.
Model training and optimisation
We trained a computer vision model specifically for this classification task. Multiple architectures were tested to find the best fit for the image quality and speed requirements of the production environment.
Validation and discovery
The model achieved over 99.9% accuracy. But the real surprise: it detected visual patterns (highlighted in green and yellow) that were unknown to the client's biologist. The AI found features that human experts had not identified.
The result
99.9% accuracy and patterns the biologist had not seen
The model not only met the accuracy requirements, it exceeded them. And it discovered classification patterns that the human expert had not identified. This is what happens when AI is deployed on a well-defined problem with good data: it does not just match human performance, it can extend it. This project was tackled together with the Unpyle team.
Ready to build your AI setup?
Let's discuss what an AI setup looks like for your sector.