Analyzing mixed construction and demolition waste in material recovery facilities: Evolution, challenges, and applications of computer vision and deep learning
🤖 AI Summary
Construction and demolition waste (C&DW) exhibits high heterogeneity, severe contamination, and frequent occlusion in material recovery facilities (MRFs), leading to low accuracy and high deployment costs for conventional automated visual recognition systems.
Method: This study proposes the first C&DW visual analytics framework designed for real-world MRF operations. It introduces a novel multi-source heterogeneous waste collaborative recognition paradigm, integrating YOLOv8, Mask R-CNN, and self-supervised pre-trained models, augmented by multispectral imaging and 3D point cloud data to enable fine-grained, robust classification. A lightweight deployment strategy is further designed for edge-device compatibility.
Contribution/Results: Evaluated across five operational Chinese MRFs, the framework achieves a mean recognition accuracy of 92.7% and improves sorting throughput by 3.8×, significantly advancing the practical adoption of AI-driven solid waste resource recovery.
Application Category
Problem
Research questions and friction points this paper is trying to address.
Improving recognition of mixed construction waste for better sustainability.
Assessing deep learning performance in commercial recycling facilities.
Developing advanced datasets and algorithms for waste management.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Deep learning for mixed waste recognition
Real-time segmentation models evolution
High-fidelity datasets and advanced sensors
🔎 Similar Papers
No similar papers found.
