Search

Conventional practices of bridge visual inspection present several limitations, including a tedious process of analyzing images manually to identify potential damages. Vision-based techniques, particularly Deep Convolutional Neural Networks, have been widely investigated to automatically identify, localize, and quantify defects in bridge images. However, massive datasets with different annotation levels are required to train these deep models. This paper presents a dataset of more than 6900 images featuring three common defects of concrete bridges (i.e., cracks, efflorescence, and spalling). To overcome the challenge of limited training samples, three Transfer Learning approaches in fine-tuning the state-of-the-art Visual Geometry Group network were studied and compared to classify the three defects. The best-proposed approach achieved a high testing accuracy (97.13%), combined with high F1-scores of 97.38%, 95.01%, and 97.35% for cracks, efflorescence, and spalling, respectively. Furthermore, the effectiveness of interpretable networks was explored in the context of weakly supervised semantic segmentation using image-level annotations. Two gradient-based backpropagation interpretation techniques were used to generate pixel-level heatmaps and localize defects in test images. Qualitative results showcase the potential use of interpretation maps to provide relevant information on defect localization in a weak supervision framework. [ABSTRACT FROM AUTHOR]

Cracks are prominent defects that significantly impact concrete bridges' structural safety and serviceability assessments. The conventional manual bridge inspection is often limited by multiple challenges related to the heavy workload of data processing and the subjective judgment of inspectors. Several vision-based methods, from traditional image processing techniques to segmentation deep learning models, have been employed for pixel-level crack detection to detect cracks in various scenes and surfaces automatically. The present paper proposes an end-to-end framework for crack segmentation combining UNet, Gabor filters, and Convolutional Block Attention Modules leveraging the complementary strengths of both spatial and frequency domains to enhance crack feature extraction and mitigate the impact of image disturbances. The designed network is trained and evaluated on a multi-source annotated crack dataset established by the authors to expose the model to multiple instances of crack appearance and background complexity and enhance its generalization ability. The proposed model achieves an Intersection over Union (IoU) of 60.62 % and an F1-score of 74.49 %, outperforming benchmark segmentation networks. Experimental results demonstrate the effectiveness of cross-domain feature fusion and multi-source data utilization in segmenting cracks with diverse patterns and background interference scenarios. • An end-to-end crack semantic segmentation framework is proposed based on a methodology combining UNet, image filtering using Gabor filters, and Convolutional Block Attention Modules, which showed a significant gain in performance compared to benchmark segmentation models. • A multi-source annotated dataset is established based on three different data sources, including an author-constructed dataset. • The proposed method is applied to high-resolution bridge images captured by a UAV for crack segmentation in complex and noisy inspection scenes. [ABSTRACT FROM AUTHOR]