Enhanced Construction Site Debris Management Using Deep Learning Classifiers for Future Remote Robotics Integration

Construction sites produce large amounts of waste daily, including broken building parts, packaging, and general trash. Debris accumulation creates significant safety risks for workers and reduces productivity, with manual debris collection and sorting consuming over 15% of workforce hours. This led to the design of a decision algorithm to classify different debris types. In this study, a machine learning classifier was built based on collected images from construction sites. Images of four types of debris were collected: ceramic, concrete, general trash, and wood. The most recent deep learning networks were used for training the classification process, such as ShuffleNet, Darknet19, and Darknet53. The accuracy of the classifier outperformed existing classifiers, achieving accuracies of 99.73%, 98.4%, and 99.18%, respectively. The proposed debris management system will offer a revolutionary solution for managing construction debris. With emerging technology in the robotics field, a system is needed to classify debris for the operation of future wireless self-driving control robots with arm grippers powered by solar energy.