Image-based reinforced concrete component mechanical damage recognition and structural safety rapid assessment using deep learning with frequency information.

Safety assessment of post-event damaged structures is vital and significant because it directly affects life security, structural repair, and economic loss, especially in earthquakes. This study applies deep learning (DL) approaches to structural health monitoring and post-earthquake reconnaissance. An image-based procedure for reinforced concrete (RC) component damage recognition and structural safety rapid assessment is established. An RC component mechanical damage image dataset is built first, and comprehensive component damage recognition tasks (CDRTs) are proposed. EfficientNet-V2 is selected as the baseline model to perform CDRTs. The discrete wavelet transform module that integrates frequency information is then proposed to improve the performance and interpretability of the baseline model. Finally, component and structural damage are linked and structural safety rapid assessment methods based on the results of CDRTs are proposed. This study overcomes the limitations of manual inspections and advances the applications of DL techniques in civil engineering by incorporating frequency information. • Proposing a procedure for RC component damage recognition and structural assessment. • Establishing a dataset for RC component mechanical damage images. • Frequency information introduction can assist CNN models in image classification. • Linking the damage type, component damage level, and structural safety state. [ABSTRACT FROM AUTHOR]