How does learning work for PICVISA equipment?

When sorting equipment is described as using artificial intelligence, it's natural to wonder what that means in practice – how does a machine "learn" to sort materials, and what does that learning process actually look like? The answer involves a combination of sensor data, training examples, and models that improve through exposure to real material.

Evolution of technologies

Early sorting technology relied on fixed rules – a sensor would detect a specific signal, and the system would respond in a predetermined way. This worked for materials with clear, consistent signatures, but struggled with anything more variable. The shift towards machine learning changed this: instead of fixed rules, systems are trained on examples, learning to recognise patterns that would be difficult to define explicitly through rules alone.

What technologies do we use in PICVISA?

PICVISA's equipment combines multiple sensor types – near-infrared, visual, and increasingly hyperspectral – with machine learning models trained to interpret the data these sensors produce. The models are trained on large sets of labelled material samples, learning to associate sensor signatures with material types, so that when the equipment encounters new material in operation, it can classify it based on the patterns it has learned.

A promising future?

As more data is collected from equipment operating in the field, models can continue to improve – learning to handle new materials, new packaging formats, or variations in waste streams that weren't well represented in the original training data. This ongoing learning is what allows sorting equipment to keep pace with a waste stream that's constantly changing, rather than becoming less effective as the materials it encounters evolve over time.

Explore PICVISA optical sorting solutions