Advancing recycling efficiency through hyperspectral identification of fabric blends

Textile sorting for recycling and revalorisation is a multifaceted challenge that requires accurate material classification. We demonstrate the effectiveness of short-wave infrared (SWIR) hyperspectral imaging as a method employed to address this challenge. The utilization of various data processing strategies enables us to ascertain the accuracy of textile blends. Employing the Multivariate Curve Resolution-Alternating Least Square algorithm, we establish an uncertainty range of ±2.7 - 5.0% using pure elements as a training set. To achieve this, we employ multiple pre-processing methods to enhance the spectrum and assess alternative regression algorithms, such as Multivariate Regression-Partial Least Square and Principal Component Algorithm. Additionally, we conducted tests using two hyperspectral systems with distinct spectral ranges: one extending up to 2500 nm and the other up to 1700 nm. Furthermore, a study on the influence of fabric color on regression and textile spectra was conducted.