Sustainable resource management in construction: Computer vision for recognition of electro-construction waste (ECW)

Abstract

A significant amount of Electro-construction waste (ECW) often ends up in landfills, leading to adverse impacts on the environment and human health. Although waste sorting utilises automated technologies like computer vision (CV), implementation in the construction industry remains limited. This study addresses the gap by evaluating the effectiveness of CV in ECW recognition to enhance resource recovery. A novel dataset was curated by sourcing images from web and applying background subtraction techniques to simulate realistic construction site conditions. This method significantly enhanced model accuracy by up to 16 %, demonstrating its potential for scalable and automated dataset generation. The study classified ECW into four critical categories: cables, switches, lights, and AC ducts. Performance evaluation across two model architectures Convolutional Neural Networks (CNNs) and Transformers showed competitive results, achieving classification accuracies of 91.52 %, 93 %, 89.81 %, and 93.62 % for ResNet50, ConvNeXt, Vision Transformer, and Swin Transformer, respectively, with low inference times suitable for real-time applications. The findings highlight the transformative potential of CV-driven solutions in sustainable waste management practices. By enabling accurate and real-time ECW recognition, this research contributes to enhanced recycling efficiency, reduced environmental impact, and resource conservation within the construction industry.