Improved Recurrent Convolutional Neural Network for Crack Detection in Concrete Structures

Detecting cracks in concrete structures involves advanced imaging and deep learning techniques to identify and assess the extent of structural damage. The problem statement in crack detection methods like an identification of cracks, complex background noise, intricate patterns, and environmental conditions are overcome by segmentation and detection process of this approach. The SDNet2018 dataset and the proposed crack detection method, like U-Net for segmentation, distinguish between cracks and non-cracks. Inception ResNetV2 for pixel-wise feature extraction enhances the identification of crack patterns. The final step involves crack-type detection using an Improved Recurrent Convolutional Neural Network (RCNN). The proposed method achieves performance metrics with an accuracy of $99.67 \%$, precision of $98.78 \%$, recall of $99.45 \%$, and an F1-score of ${9 9. 2 1 \%.}$ Existing methods like Convolutional Neural Network (CNN+U-Net), K-means with Genetic Algorithm (GKA), Multiscale Crack Detection Network (MSCNet), and Transfer Learning (TL) overcome limitations and provide accurate solution for crack detection in surface areas.