Towards a circular economy: Recapturing battery, metal, and plastic from soil-size and gravel-size municipal solid waste incineration bottom ash using convolutional neural networks

Abstract

Utilizing municipal solid waste incineration bottom ash (MIBA) to replace natural aggregate as building material substitutes is running under a circular economy worldwide. Applying pretreatments on MIBA to achieve manageable environmental impacts due to the complex composition of it becomes as a necessary route. We explored the convolutional neural network (CNN) techniques to identify quality scenarios of soil-size (<2 mm) and gravel-size (4–19 mm) MIBA, using 10,122 customized images divided into groups of 5060 normal/visible images and 5062 infrared images with different image resolutions. The CNN models achieved an interpretation accuracy ranging from 83.1% to 97.33% for scenarios of quality consideration under normal and infrared illumination. In addition, the visibility analysis revealed that well-trained CNNs models could detect the unsuitable remains/residues features of metals and batteries, in MIBA. The findings highlighted this model's considerable potential in the field of waste recycling and the recapturing of battery, metal, and plastic from soil-size and gravel-size MIBA.