Preliminary Study of Ceiling Damage Detection System Using Image Database by Deep Learning Approach (Convolutional Neural Networks)

Abstract

In Japan, large span buildings are employed as shelters for residents when earthquake or other disasters occur. Fallings of ceilings or other hanging components in these buildings usually happen in a sudden and are dangerous to human body. Damage detection and safety judgements of ceilings are mainly relied on the observation with human (especially by experts, First Class Authorized Architects in Japan, for example) naked eye. Damage detection and safety evaluation of ceilings or hanging components can be transformed into computer vision problems. The prevailing deep learning method, especially the convolutional neural networks (ConvNets), is a potent algorithm to construct a damage detection system. In this report, an architecture of deep learning model applying ConvNets is realized to evaluate the damage degree of ceilings. In the discussion part, more detailed discussions of the deep learning model are also performed to make a better understanding of it.