The crack ratio is one of the indices used to quantitatively evaluate the soundness of asphalt pavement. However, since the inspection of pavement requires much labor and cost, automatic inspection of pavement damage by image analysis is required in order to reduce the burden of such work. In this study, a system was constructed that automatically detects and evaluates cracks from images of pavement using a convolutional neural network, a kind of deep learning. The most novel aspect of this study is that the accuracy was recursively improved through retraining the convolutional neural network (CNN) by collecting images which had previously been incorrectly analyzed. Then, study and implementation were conducted of a system for plotting the results in a GIS. In addition, an experiment was carried out applying this system to images actually taken from an MMS (mobile mapping system), and this confirmed that the system had high crack evaluation performance.

To ensure the safe operation of bridges, it is necessary to carry out repair and strengthening based on appropriate inspections and damage records. However, in conventional bridge maintenance, the inspection results are recorded as 2 D data, which makes it difficult to carry out appropriate repair and strengthening due to the small amount of information. Therefore, research has been conducted to utilize Building Information Modeling (BIM) for bridge maintenance to record damage in more detail, and various concepts have been proposed. However, since no concrete method has been established to automate the entire process from damage detection to linking the damage to the BIM-based model, these tasks currently have to be carried out manually. For this reason, the use of BIM in actual maintenance work has not progressed. This study proposes a method to automate the process from damage detection to linkage to a BIM-based model. In the proposed method, each damage is detected using deep learning, local location information is obtained using Structure from Motion, and the location information is integrated into the 3 D data of the entire bridge. In addition, a case study using an actual bridge is also presented to demonstrate the usefulness of the proposed method.

It is necessary to assess damage properly for the safe use of a structure and for the development of an appropriate maintenance strategy. Although many efforts have been made to measure the vibration of a structure to determine the degree of damage, the accuracy of evaluation is not high enough, so it is difficult to say that a damage evaluation based on vibrations in a structure has not been put to practical use. In this study, we propose a method to evaluate damage by measuring the acceleration of a structure at multiple points and interpreting the results with a Random Forest, which is a kind of supervised machine learning. The proposed method uses the maximum response acceleration, standard deviation, logarithmic decay rate, and natural frequency to improve the accuracy of damage assessment. We propose a three-step Random Forest method to evaluate various damage types based on the results of these many measurements. Then, the accuracy of the proposed method is verified based on the results of a cross-validation and a vibration test of an actual damaged specimen.

We proposed an automatic detection method of slope failure regions using a semantic segmentation method called Mask R-CNN based on a deep learning algorithm to improve the efficiency of damage assessment in the event of slope failure disaster. There is limited research on detecting landslides by deep learning, and the lack of training data is an important issue to be resolved, as aerial photographs are not taken with sufficient frequency during a disaster. This study attempts to use CutMix-based augmentation to improve detection accuracy. We also compare the detection results obtained by augmentation of multiple patterns. In the comparison of the not augmented data case, the recall increased by 0.186 in the case using the augmented data with the shape of the slope failure region maintained. When the image data was augmented while maintaining the shape of the slope failure region, the recall score indicated the low oversights in the prediction result is 0.701. This is an increase of 0.186 compared to the case where no augmentation was performed. In addition, the F1 score was 0.740, this also increased by 0.139, and high values were obtained for other indicators. Therefore, the method proposed in this study is greatly useful for grasping slope failure regions because of the detection with high accuracy, as described above.