Cracks in infrastructure pose serious risks to public safety and require timely detection for effective maintenance. This study presents Deep Crack, a deep learning-based approach for image-based crack prediction. The proposed method utilizes Convolutional Neural Networks (CNNs) with the Rfcnb architecture as the backbone, combined with transfer learning to improve detection performance. Extensive data preprocessing techniques, including image augmentation, are applied to address data limitations and enhance model generalization. The model is trained and validated on a diverse dataset, enabling it to accurately distinguish between cracked and non-cracked images. A customized classification layer, incorporating global average pooling and fully connected layers, is integrated to further improve performance. The effectiveness of the model is evaluated using metrics such as accuracy, precision, and recall, along with confusion matrix analysis and classification reports. Experimental results demonstrate that the proposed approach achieves high classification performance, making it suitable for real-world infrastructure monitoring applications. This work highlights the effectiveness of combining deep learning and transfer learning techniques for automated crack detection and emphasizes their potential applications in civil engineering and infrastructure management.
Naik et al. (Thu,) studied this question.