The technique makes use of so-called ‘speckle imaging’ where a material illuminated by a laser will produce a unique pattern of reflected spots, or speckles into a camera. By training a deep neural model with a large set of samples, it was found possible to detect up to 30 types of material with 98% accuracy.
The pre-baked model runs on a Raspberry PI zero with an off-the-shelf camera all powered from a power bank. This allows the whole assembly to simply drop onto an existing laser cutter head, with no wiring needed.
Even if you’re a seasoned laser cutter user, with a well-controlled stock pile, the peace-of-mind this could give would definitely be worth the effort. A more detailed description and more videos may be found by reading the full paper. Here’s hoping they release the system as open source, one day in the not-to-distant future. If not, then, you know what to do 🙂