Surface Crack Detection: Advances in Automated Visual Inspection for Infrastructure Safety

Surface crack detection is an essential method in structural health monitoring systems, which is very crucial for maintaining infrastructure safety and durability. Traditional methods mainly comprise manual inspection, but they have many disadvantages in terms of inconsistency, subjectivity, and capability to access hazardous locations; therefore, the interest in automated detection technologies is increasing considerably. Although much progress has been achieved in developing computer vision- and machine learning-based approaches, existing solutions still face challenges in terms of reliably performing in variable environmental conditions, diverse surface textures, and computational constraints toward practical field deployment. The proposed research tries to develop a robust and efficient surface crack detection system by embedding specialized attention mechanisms into CNNs to enhance feature discrimination without hindering computational efficiency. The methodology introduced applies a dual attention architecture by incorporating channel and spatial attention modules within an optimized CNN framework. The network was trained and validated using the Concrete Surface Crack Detection dataset containing 40,000 diverse surface images. Experimental results illustrate the exceptional performance of the proposed model, with 99.69% accuracy and an F1 score of 99.69%, thus outperforming the state-of-the-art techniques. This will allow for the more reliable monitoring of infrastructure, which may reduce maintenance costs and catastrophic failures while extending the service life of civil infrastructure through early detection of structural deterioration.