Deep Learning Approach for Automated Rapid Visual Screening

Rapid visual screening (RVS) is crucial for identifying potentially unsafe structures and mitigating urban earthquake vulnerability, but conventional methods are labor-intensive, time-consuming, costly, and susceptible to subjective errors due to human involvement in the process. This study introduces a novel automated deep learning (DL) approach, leveraging convolutional neural networks (CNNs), including residual network (ResNet), visual geometry group (VGG), inception, and densely connected convolutional networks (DenseNet). The methodology employs hard voting ensemble methods and hierarchical classification to extract and classify necessary visual features from street view images for RVS. This approach addresses the limitations of previous research in direct screening applications. The developed program significantly accelerates the process, producing final scores in just ۵۳ seconds, making it ۱۷ times faster than conventional methods, with accuracies up to ۹۶%. This approach drastically reduces labor costs, enhances feasibility for large-scale projects, and offers a robust tool for both pre-earthquake evaluation and post-earthquake rapid screening of damaged structures, thereby improving urban earthquake resilience.

کلیدواژه ها:

Pre-earthquake evaluation ، Rapid visual screening (RVS) ، Deep learning ، Convolutional neural network(CNN) ، Seismic damage assessment