Coarse-to-Fine Seismic Assessment Based on Computer Vision

Recently, remote sensing satellites, unmanned aerial vehicles (UAVs), and smartphones have been extensively utilized in non-contact post-earthquake inspection at different scales with cutting-edge computer vision and machine learning techniques. This study establishes a computer-vision-based coarse-to-fine seismic assessment framework to localize dense buildings in urban areas, classify collapsed and non-collapsed states, recognize multi-type surface damage on structural components, and evaluate seismic performances. First, a Transformer-CNN deep learning architecture is designed for semantic segmentation of dense buildings and binary classification of collapsed states using large-scale remote sensing satellite images. It consists of a Swin Transformer encoder, multi-stage feature fusion module, and UPerNet decoder to extract global correlations and local features of dense buildings synchronously. Then, a multi-task learning strategy is proposed to simultaneously recognize multi-type structural components (column, beam, wall), seismic damage (concrete crack, spalling, and rebar exposure), and multi-level damage states (minor, moderate, major) using medium-scale UAV images. It contains a CNN-based encoder-decoder backbone with skip-connection modules and multi-head segmentation subnetworks for different tasks. The geometric consistency loss of split line, area, and curvature is further designed to refine the semantic segmentation of local details, increase boundary smoothness, and suppress inner voids. Finally, a machine learning neural network is established to quantify the seismic damage index of structural components using damage-related parameters (lengths, areas, and numbers of concrete crack, spalling, and rebar exposure) and design-related parameters (axial compression ratio, shear span ratio, and volumetric stirrup ratio) as inputs. A seismic damage indicator with an explicit bound of [0, 1] can be obtained to reflect the nonlinear accumulation of seismic damage. The effectiveness and applicability under real-world post-earthquake scenarios have been validated by the 2017 Mexico City Earthquake M7.1, 2008 Beichuan Earthquake M8.0, 2010 Yushu Earthquake M6.9, 2015 Nepal Earthquake M7.8, 2016 Ecuador Earthquake M7.8, and 2016 Meinong Earthquake M6.7.