Variable Selection from Image Texture Feature for Automatic Classification of Concrete Surface Voids.

Area of Science:

• Computational intelligence
• Machine learning applications in civil engineering
• Image processing for material defect detection

Background:

• Surface voids (bugholes) are common defects in concrete casting.
• Manual inspection of concrete bugholes is inefficient and inconsistent.
• Automated classification is crucial for accurate concrete surface inspection.

Purpose of the Study:

• To develop an automatic classification system for concrete bugholes.
• To improve accuracy and feature interpretability in bughole detection.
• To provide a reliable method for identifying surface voids in concrete.

Main Methods:

• Texture features were extracted from concrete surface images using Gabor filters and gray-level run lengths.
• A variable selection strategy based on cumulative voting was employed for feature interpretability.
• An automatic ensemble classifier was designed for bughole classification.

Main Results:

• The proposed method achieved comparable prediction accuracy to existing techniques.
• The selected variables enhanced the interpretability of the classification model.
• Experimental results on 1000 image samples demonstrated the method's effectiveness.

Conclusions:

• The developed machine learning approach offers an effective solution for automatic concrete bughole classification.
• The combination of feature extraction, interpretable variable selection, and ensemble classification improves defect detection.
• This method provides a more consistent and efficient alternative to manual inspection of concrete surfaces.