Estimating demolition waste from residential interior photos: A Large Language Model solution

Demolition waste management is a critical challenge during the final stages of construction projects, particularly concerning fine-grained waste at the room level. This paper investigates the predictive potential of visual elements for demolition waste estimation, with an emphasis on interior room design. A framework integrating deep learning models with Large Language Model (LLM) is proposed to automatically classify interior design types and predict demolition waste. By utilizing both residential interior photos and floor plans, the framework can identify design types, segment walls, match rooms, and estimate demolition volumes at the room level. The results show high accuracy in design classification and waste prediction, surpassing traditional methods based on waste generation rates. The framework also provides valuable insights into specific waste materials, enhancing waste management at a detailed level. This research advances micro-level demolition waste prediction, promoting the broader application of LLMs in the construction industry.