Evaluation of earthquake resistance of urban buildings using image processing and machine learning techniques

Abstract

In this project, an approach has been taken to evaluate the earthquake resistance of urban buildings in Dhaka. An automated decision support system has been developed that takes the images of the buildings along with basic information as input. The system then outputs whether the building is at risk and requires structural evaluation. Data from 1106 buildings were collected during the project from 12 different areas of Dhaka. The output decision of the system is determined using a machine learning algorithm. Specifically, a CNN-based deep learning model was trained on the data collected during this project. Every deep learning model needs a baseline to make the predictions on. In this project, the baseline was developed from the FEMA P-154 report that deals with the visual screening of buildings to assess their risks during earthquakes. FEMA is the Federal Emergency Management Agency (FEMA) of the USA, a renowned agency that works with seismic hazards. After experimenting with different parameter combinations, the maximum accuracy achieved by the model was 71%. The latest deep learning models operate on millions of instances to make their predictions. Comparatively, our model was trained on only 1106 instances. With the introduction of more data points, we can achieve an accuracy of over 90% with this model.