Research on Methods for Determining and Understanding the Soundness of Retaining Walls Using Image Analysis

In high-density urban areas, retaining walls have been used for retaining earth when a housing lot is built in sloping terrain by cutting and filling earth. Although standards for residential retaining walls are set by the Building Lots Development Regulation Law, many retaining walls built before the law came into effect remain in unqualified conditions. It has been pointed out that when a large earthquake occurs, aging retaining walls can be damaged, affecting evacuation and disaster relief activities, restoration of residential areas, and reconstruction of daily life. The 2016 Kumamoto earthquake collapsed retaining walls and broke down into gutters in residential areas mainly in the Kumamoto metropolitan area and the Aso region. When a housing lot gets damaged from a disaster, a retaining wall needs to be restored before any of the damaged house. Therefore, it is important to promote reinforcement of existing aged retaining walls, which will lead to the resilience of cities, smooth rescue, and cost reduction for restoration. In this study, we suggest a method to assess vulnerability of a retaining wall in a simple way, aiming to develop an application for judging the danger level from images by utilizing machine learning. We used CNN as a deep learning method, which is an effective method for image recognition, to identify a type of retaining walls. The number, condition of drainage holes and the method of identifying cracks using median filters were also investigated.