2019 年 60 巻 1 号 p. 33-39
Existing image processing programs for detecting structural damage such as cracks have required the fine-tuning of numerous parameters and experience-based expertise. A method for distinguishing different types of cracks applying deep learning has been developed using tunnel lining images. A classifier was created after learning from a large volume of images in two groups - either with "presence of a crack" or "absence of a crack." The classifier successfully recognized the presence or absence of cracks in images at a rate of more than 90%. Using a color-coded pixelated image to show the position of probable cracks, this paper proposes a hybrid detection method for analyzing cracks with a focus on their location and direction of progress.