Classifying construction and demolition waste by combining spatial and spectral features

Abstract

The generation of construction and demolition waste (C&DW) has increased on a yearly basis, and the damage caused to the environment is significant. To save resources, C&DW needs to be managed and recycled. Currently, waste-classification methods are mainly based on wind selection, water selection, screening and manual sorting. These methods are inefficient and the classification results are not clean. To achieve efficient and intelligent recycling of C&DW, it is essential to classify the wastes effectively. In this paper, four methods are reported: characteristic reflectivity and extreme learning machine (ELM); first-order derivative of characteristic reflectivity and ELM; grey level co-occurrence matrix and ELM and convolutional neural network. These methods were used to classify typical types of hard-to-distinguish waste: wood, rubber, brick and concrete. It was found that each method had inadequacies, and the correct rate was between 82·22 and 89·33%. Therefore, a weighted fusion of membership matrix, which combine all four methods, was proposed. As a result, the correct rate in repeated experiments significantly improved to 95%. The classification results can be used in further study about automatic sorting of C&DW using robotics instead of people.