Advanced dataset acquisition for improved construction and demolition waste classification using machine learning

Efficient sorting and recycling of construction and demolition waste (CDW) are vital to sustainable development and a circular economy in the construction industry. Building on our previous study that achieved up to 92.3% accuracy using RGB camera data, we propose an improved data set acquisition and feature extraction approach to improving classification performance. We introduce a customized measurement line with industrial RGB cameras, force transducers for volume and mass estimation, and acoustic transducers for ultrasound frequencies. By integrating these additional data sources and exploring various feature extraction techniques, such as shape indices, texture entropy, and mean intensity gradients, our approach aims to enrich the set of features for machine learning algorithms and increase classification accuracy. This research addresses the challenge of improper sorting in CDW recycling, which limits the value of recycled aggregates in high-quality applications.