Abstract
In this study, we developed the high-resolution detector for crack and patching on pavement of expressway by using convolutional neural network (CNN), and culculated the crack ratio by this method. First, small images such as crack and patching area was randomly sampled using sketch images. We set several image sizes to make the judgment by CNN. Then, the size and the judgment accuracy of the model were compared. As a result, over learning of models was observed for larger sizes, and accuracy of model reached a peak at an intermediate size. Cracks and patching were detected by the learned model, and the crack rate was calculated based on the conventional method. From the above results, it was shown that the accuracy of crack ratio is high with t he judgment image size of 90 pixels, and the crack shape can be visualized in detail.
