An Image-Based Approach to Automated Recognition of Asbestos-Containing Components in Wall Demolition Waste

The feasibility to discriminate potentially asbestos-containing components from asbestos-free concrete based on camera images using the example of wall demolition waste is investigated. For this, three types of asbestos substitute materials and two types of concrete are crushed and photographed. The classification of the fragment images is carried out with a) morphological and texture features and b) with features automatically extracted by the pretrained MobileNetV3 network. Feret diameters, circularity, and others served as morphological descriptors. The texture was described by measures of grey-level intensity, as obtained from the grey-level co-occurrence matrix. Support vector machines are found to achieve classification accuracies above 99 % based on the automatically extracted features. The presented identification approach is promising to automate the treatment process of asbestos-containing materials from construction and demolition waste, which is effortful and requires expert knowledge to this day.