Computer vision to recognize construction waste compositions: A novel boundary-aware transformer (BAT) model

Area of Science:

• Computer Vision
• Artificial Intelligence
• Environmental Science

Background:

• Accurate construction waste recognition is crucial for effective waste management, including levy determination and robotic sorting.
• Existing computer vision (CV) methods struggle with low accuracy, particularly in distinguishing boundaries between mixed waste materials.

Purpose of the Study:

• To introduce a novel boundary-aware Transformer (BAT) model for precise, fine-grained composition recognition of construction waste mixtures.
• To enhance the semantic segmentation of construction waste by accurately identifying material boundaries.

Main Methods:

• A pre-processing workflow to isolate difficult edges from the background.
• A Transformer architecture with a custom cascade decoder for waste material segmentation.
• A learning-enabled edge refinement scheme to improve boundary segmentation precision.

Main Results:

• The BAT model demonstrated a 5.48% improvement in mean intersection over union (MIoU) and a 3.65% increase in Mean Accuracy (MAcc) on a benchmark dataset.
• Achieved superior performance in segmenting nine types of construction waste materials in cluttered and mixed states.

Conclusions:

• The BAT model significantly advances construction waste material semantic segmentation accuracy.
• This research facilitates the practical application of CV in construction waste management, promoting a circular economy.