Surface Crack Detection in Concrete Using Transfer Learning with Lightweight CNNs

Surface cracks in concrete structures can significantly compromise their durability and structural integrity if left undetected. Traditional inspection methods are often time-consuming, subjective, and labor-intensive. This study presents an automated and efficient approach for detecting surface cracks in concrete using transfer learning with lightweight convolutional neural networks (CNNs). The primary objective is to achieve high detection accuracy while minimizing computational costs, enabling real-time implementation on resource-constrained devices such as smartphones and UAVs. In this research, a pre-trained MobileNetV۲ architecture is fine-tuned on a dataset of labeled concrete surface images to distinguish between cracked and non-cracked regions. Data augmentation techniques are employed to enhance generalization and mitigate overfitting. The performance of the proposed model is evaluated using standard metrics such as accuracy, precision, recall, and F۱-score. Results indicate that the lightweight model achieves competitive accuracy compared to heavier architectures, with significantly lower inference time and memory usage. The findings suggest that MobileNetV۲, when used with transfer learning, provides a practical balance between performance and efficiency for real-world applications in structural health monitoring. This study contributes to the development of cost-effective, scalable, and accurate crack detection systems that can support early maintenance decisions and reduce long-term repair costs. The proposed method demonstrates the potential of integrating AI-driven inspection techniques into smart civil infrastructure management, offering a promising solution for automated concrete damage assessment.

کلیدواژه ها:

Surface Crack Detection ، Transfer Learning ، Lightweight CNN ، Mobile Net V۲