Representation Learning Using RAIN RFID Tag Backscatter Features for Material Classification in Circular Economy Applications

We present how RAIN RFID power-on-tag-reverse (POTR) features can be used to reliably differentiate between materials having different relative dielectric permittivity (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\epsilon_{r}$</tex>) and effective loss <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(\tan (\delta))$</tex>. Our approach shows how this is achieved using 15 diverse RAIN RFID tags, having an embedded Tmatch antenna design, deployed on 7 different material types. We present a data visualization and K-means clustering algorithm that can reliably differentiate between material types with 94 % accuracy. We show how our approach is particularly useful at differentiating between materials having very similar <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\epsilon_{r}$</tex> but different <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\tan (\delta)$</tex>. We also demonstrate the technique appears to be robust to reflections, wet inlay adhesives and material thickness via a limited study conducted in a non-idealized warehouse environment. Future research directions are also discussed.