Interleaved Deep Artifacts-Aware Attention Mechanism for Concrete Structural Defect Classification.

Area of Science:

• Civil Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Automatic classification of concrete structural defects is challenging due to surface variations and environmental factors.
• Existing methods struggle with complex textures, lighting, and overlapping defect types.
• Accurate defect detection is crucial for structural health monitoring and maintenance.

Purpose of the Study:

• To propose a novel deep learning architecture for robust concrete defect classification.
• To address challenges posed by diverse surface textures, lighting conditions, and overlapping defects.
• To develop a fully automatic, end-to-end trainable system for defect identification.

Main Methods:

• An interleaved deep artifacts-aware attention mechanism (iDAAM) was developed.
• The architecture combines fine-grained dense modules (FGDM) for multi-scale feature aggregation and concurrent dual attention modules (CDAM) for localized feature encoding.
• FGDM captures robust information across scales, while CDAM refines feature representation for variations in defect appearance.

Main Results:

• The iDAAM architecture demonstrated superior performance in classifying multi-target, multi-class, and single-class concrete defects.
• Extensive experiments on three large public datasets confirmed the effectiveness of the proposed method.
• Ablation studies validated the contribution of FGDM and CDAM components.

Conclusions:

• The proposed iDAAM offers a significant advancement in automatic concrete defect classification.
• The method's ability to handle complex image variations makes it highly suitable for real-world applications.
• This approach paves the way for more reliable structural health assessment through automated image analysis.