Performance Evaluation of Convolutional Neural Networks for Automated Crack Detection in Concrete Structures

Automated visual inspection of structural integrity plays a critical role in civil engineering maintenance. This study evaluates five state-of-the-art Convolutional Neural Network architectures for detecting micro-cracks in reinforced concrete surfaces under diverse environmental conditions. Experimental results demonstrate that EfficientNet-B3 achieves 96.8% accuracy with significantly lower computational overhead.