Surface mortar detection and performance evaluation of recycled aggregates based on hyperspectral technology

The quality of recycled aggregates is affected by the residual mortar. It is significant to detect the surface mortar distribution of recycled aggregates after mortar removal by mechanical crushing. From this perspective, a method to accurately detect the surface mortar distribution of recycled aggregates is proposed. The processed hyperspectral features were obtained by applying data filtering and screening, L2 norm processing, feature transforming and dimensionality reduction. Then these features were put into the extreme learning machine (ELM) for offline training, and a sliding window processing mechanism was added to the trained model, which was used to detect the recycled aggregates and output the category images. Finally, two characterization parameters of the proportion of mortar area and the mortar volume were extracted from the images. The regression models of water absorption (WA) and apparent density (AD) of recycled aggregates were obtained based on the proportion of mortar area and the mortar volume, with the determination coefficients of 0.99. The results demonstrated that the proposed approach could be profitably applied to evaluate the quality of the recycled aggregates, which lays a foundation for visual identification and intelligent sorting of recycled aggregates.