A hybrid classification and evaluation method based on deep learning for decoration and renovation waste in view of recycling

Abstract

The escalating volume of decoration and renovation waste (D&RW) amid the rapid urbanization in China has posed significant challenges for the effective recycling of this waste stream, primarily due to the difficulty of accurately assessing its precise composition. To enhance the recycling of high-value materials from D&RW, a comprehensive understanding of its composition and quality is crucial before sorting. In this study, we propose a hybrid method that combines instance segmentation deep learning (DL) models with morphological machine learning (ML) models to automate the classification and evaluation of D&RW. A meticulously labeled dataset comprising 53,000 individual grains is curated for classification and instance segmentation. Subsequently, the ML model predicts the equivalent thickness of a grain according to the grain morphological data vector. The D&RW grains are then evaluated for weight based on the model outputs. The proposed method exhibits remarkable accuracy, as indicated by a relative low error of 2.8% in total weight evaluation between the model’s predictions and manual sorting.