Earthquake-generated construction and demolition waste recovery using hyperspectral imaging aided by shallow neural networks technique

Construction and demolition waste (C&DW) accounts for nearly one-third of total waste generation in the European Union, representing a significant environmental challenge. Although recovery rates are high (∼89%), much of the recycled material is downcycled, hindering true circular economy goals. This study proposes an integrated analytical method combining portable X-ray fluorescence (XRF), near-infrared hyperspectral imaging (NIR-HSI), and Shallow Neural Networks (SNN) for fast, accurate classification of earthquake-related C&DW from central Italy. Thirty sample sets from the 2016–2017 earthquake zones in Abruzzo, Marche, and Emilia Romagna were analyzed using portable energy-dispersive XRF to define three recycling-oriented material classes: concrete-based (CON), ceramic-rich (CER), and natural aggregates (NAT). Statistical tests and principal component analysis (PCA) confirmed significant differences among classes. NIR-HSI spectra (1000–1700 nm) were processed to train an SNN with a single hidden layer. The classifier showed excellent precision, recall, specificity, and F1-scores (≥ 0.98) across classes, with misclassifications limited to borderline cases like glazed ceramics. The goal of this work is to evaluate the best achievable performance within a controlled feasibility framework, demonstrating that the coupling of NIR-HSI with SNN provides a rapid, robust, and transferable strategy for automated C&DW classification, thereby supporting circular economy goals through improved material recovery and recycling efficiency.