A new method of construction waste classification based on two-level fusion

Area of Science:

• Environmental Science
• Computer Science
• Materials Science

Background:

• Automatic sorting of construction waste (CW) is crucial for efficient and safe CW recycling.
• Accurate CW classification is the primary step guiding automated sorting processes.

Purpose of the Study:

• To propose a novel two-level fusion method for construction waste classification.
• To enhance the performance and accuracy of automatic construction waste sorting.

Main Methods:

• Utilized hue histograms (HH) for global hue information and histograms of oriented gradients (HOG) for local features.
• Applied bag-of-visual-words (BoVW) to create a B-HOG vector from HOG descriptors.
• Implemented feature-level fusion combining HH and B-HOG, followed by decision-level fusion of two base classifiers.

Main Results:

• The proposed two-level fusion method demonstrated effectiveness and feasibility in CW classification.
• Experimental results on a dataset of five CW types confirmed the method's performance.
• Comparison with existing methods indicated superior or competitive results.

Conclusions:

• The developed two-level fusion technique significantly improves construction waste classification accuracy.
• This method offers a viable solution for advancing automatic construction waste sorting and recycling.
• The fusion of global and local visual features provides a robust approach for complex classification tasks.