Big visual data analytics for damage classification in civil engineering

Visual data provide a wealth of information to better understand the world around us. A tremendous amount of visual data is collected in civil engineering applications through efforts such as scientific experiments, field surveys, resource management, and reconnaissance missions. Among these efforts, visual data generate crucial and abundant information evaluating the condition of a civil structure. As a typical example, during a disaster such as a natural catastrophe or industrial explosion, vast amounts of perishable image data are collected that may be used to generate new knowledge from the consequences of that event. However, not only does this process require timeconsuming data collection by human engineers, it is also tedious and expensive to manually search through these data sets to find the most informative images. Autonomous collection, processing and analysis offer great potential to support structural evaluation. In this study, we propose a novel autonomous evaluation method to examine large volumes of images. Recent deep convolutional neural network (CNN) algorithms are applied to extract visual information from the collected images. Task-oriented engineering knowledge and experience are incorporated into the procedures to increase accuracy. The target application addressed in this study is post-disaster building damage evaluation. Illustration of the technique and capabilities for collapse classification is demonstrated using large-scale images gathered from past earthquake events.