Enhanced Intelligent Identification of Concrete Cracks Using Multi-Layered Image Preprocessing-Aided Convolutional...

Area of Science:

• Civil Engineering
• Computer Science
• Materials Science

Background:

• Concrete crack identification is crucial for structural health monitoring.
• Conventional convolutional neural networks (CNNs) show promise but are limited by image noise.
• Image noise from various sources (light spots, blurs, stains) impacts CNN accuracy.

Purpose of the Study:

• To enhance the accuracy, noise immunity, and versatility of CNN-based concrete crack identification.
• To develop a novel framework integrating CNNs with a multi-layered image preprocessing (MLP) strategy.
• To investigate the impact of noise on crack identification performance.

Main Methods:

• Developed a hybrid framework combining CNNs with a multi-layered image preprocessing (MLP) strategy.
• MLP strategy incorporates homomorphic filtering and Otsu thresholding.
• Built, trained, and tested CNN models for crack detection and classification on a large dataset.

Main Results:

• Comparative studies demonstrated the effectiveness of the MLP strategy in improving crack identification.
• The proposed framework showed enhanced accuracy and noise immunity compared to conventional CNNs.
• Performance was evaluated under various noise conditions and levels.

Conclusions:

• The hybrid framework significantly improves concrete crack identification accuracy and robustness against noise.
• The MLP strategy is effective in mitigating noise interference in CNN-based crack detection.
• This approach offers a more versatile and reliable solution for structural health diagnosis.