Vision Systems Design wanted to learn more about the RETEX system and its features and capabilities, so we reached out to Francesco Di Clemente, Head of RETEX Specim, at Spectral Imaging Ltd.
Editor's note: This Q&A may have been edited for style and/or clarity.
Vision Systems Design (VSD): From a system design perspective, what were some challenges in achieving real time hyperspectral processing for textile identification, and how were they solved?
Francesco Di Clemente (FDC): Hyperspectral data flow can be huge if not treated properly. The key is to focus on relevant information instead of trying to brute-force it! Also, clever use of the hardware makes the difference and since RETEX is a “Specim-based solution” we can control both hardware and software layer.
VSD: How does this system handle material variability—such as fiber orientation, fabric thickness, finishes, and wear—when distinguishing between visually similar fibers and complex blends?
FDC: First and foremost, RETEX has its very own workflow to cancel out these kinds of disturbances. Besides that, the very same architecture of the AI is designed to compensate for variability.
VSD: Elastane detection is often cited as a difficult problem in textile sorting. What spectral or AI based strategies enable this system to identify elastane, even in previously unseen blends?
FDC: I cannot tell you without revealing some industrial secret. The only thing I can say is that the closer you look, the less you see.
VSD: For integrators deploying this system into existing sorting or recycling lines, what are the main considerations around illumination, conveyor speed, field of view, and calibration to maintain classification accuracy?
FDC: RETEX is highly resistant to “noise,” and it is very easy to setup and calibrate. Also, it partially self-calibrates each time it starts. Basically, every time the system starts, it self-calibrates itself around illumination, field view, and conveyor speed.
VSD: How does the modular architecture support scaling from lab validation to full production, and what parts of the system—hardware, models, or workflows—typically require adaptation during that transition?
FDC: We tend to use the same setup in the lab that will be given to the customer so that minimum adaptation is required. What sometimes happens is that a customer might require a new material or wants to cover a new use case for which we customize one or more process. The concept is “same software, different configuration.”
Related: Condi Food Develops Hyperspectral Imaging Sorting Solution
