Visual inspection of concrete structures is an important task; however, it is labor-intensive work, taking much time and effort, and it is often dangerous when inspectors work in high places. Automatic detection of deterioration such as cracks, free lime, exposed reinforcing bars, etc., from digital images has been extensively researched. Current deep learning approaches have improved the detection performance compared with conventional approaches. Most deep learning approaches use publicly available deep learning models, such as Faster R-CNN and Single Shot multiBox Detector (SSD). Although the results of deterioration detection methods have been compared with the conventional approach, the deep learning models themselves have not been compared. In this research, using the same datasets, seven deep learning models (YOLOv3, RetinaNet-50, RetinaNet-101, RetinaNet-152, SSD512, SSD300, and Faster R-CNN), were compared for detecting five types of deterioration (cracks, exposed reinforcing bars, free lime c-type, free lime d-type, and free lime e-type). YOLOv3 showed the highest mean average precision (mAP) of 85.7%, whereas the other models showed less than 80%.

 Aging of social infrastructure has become a social issue in Japan. In order to use existing structures effectively, how to make maintenance and management more efficient and advanced is an important problem to tackle. For visual inspection of concrete structures such as bridges and tunnels, crack is one of the factors that cause structural deterioration and failure, and it should be evaluated accurately. Recently, many methods based on image processing and machine learning, including deep learning, have been proposed. It is expected to evaluate cracks efficiently and accurately for practical use in the field. In this article, we propose a method of applying two deep learning models. The first model, which is based on VGGNet, classifies if cracks exist or not in a local area. Another model, which is based on FCN, extracts crack pixels from the detected area. Experimental result shows that the proposed method extracts cracks automatically and accurately, compared to conventional methods.

 Currently, the ground condition of the tunnel cutting face is evaluated by engineers manually. However, the evaluation of ground conditions based on the visual observation of engineers includes subjective factors. Therefore, methods that can quantitatively evaluate the ground condition of the tunnel cutting face is necessary. In this paper, a quantitative evaluation of the tunnel-cutting face’s geological condition is realized by estimating the drilling energy from the tunnel face image. We learn and update the estimator for each construction site using online learning that takes tunnel construction characteristics into account.

 The basic policy of bridge management is preventive maintenance which contains to execute bridge inspection, diagnosis, repair design, and records continuously. Deep learning which has been widely used in bridge inspection is not well examined in repair design because it is more complex and requires explainability. This study aims to propose a bridge repair decision model consisting of a damage grade assessment model and a damage progress decision model for repair design. The model trained by using periodic inspection data including images and texts achieved about 0.7 recall rate in bridge repair decision task.

 In the past years, a large number of buried pipes such as sewer pipes, gas pipes, and other pipes for infrastructure development have been an obstacle to design and construction. One solution to this is an adoption of ground penetrating radar which is a nondestructive measurement technique that locates under-ground targets by electromagnetic waves. However, there has been a problem of costs and objectivity due to visual observation of investigation results. In order to solve this problem, this paper proposes the method which automatically estimates a position of buried pipes from ground penetrating radar data using YOLOv3 and U-Net which are types of deep learning. This automatic estimation enables analysis of enormous ground penetrating data, contributing to construct a database of buried pipes.

 Permanent Scatterers Interferometry (PSI) can measure accurately ground deformation along the radar line-of-sight (LOS) direction. To detect three-dimensional (3D) ground deformation, PSI results along two different LOS directions and GNSS survey data are integrated. However, in many cases in Japan, the GNSS data available for the analysis are limited to the data of the electronic reference points operated by Geospatial Information Authority of Japan. The geolocation of the points are approximately 10 to 20 km away from each other, and the data may not reflect the deformation of the area of interest. Thus, in this paper, we examined the approach to effectively estimate the 3D deformation around Niigata airport under the framework of the data integration. Furthermore, we selected Distributed Scatterers (DS) to increase the number of the analysis points and set the upper limit of the distance for matching analysis points in each LOS. We compared the estimated vertical deformation velocity with the velocity derived from leveling survey data, and the Root Mean Square Error (RMSE) was improved 2.4 mm/year from 4.0 mm/year. It was found that our proposal is effective for monitoring ground deformation even though the GNSS data available are far away from the area of interest.

 Rapid response to natural hazards, such as floods, is very important to establish an initial response system. Therefore, it is expected to establish a disaster investigation method using synthetic aperture radar (SAR) data that can be observed even in cloudy weather or at night. However, the conventional threshold method has low accuracy in inundated area extraction. In this study, inundated area extraction using U-Net on preand post-event COSMO-SkyMed images were performed for the flood in Ayutthaya, Thailand in 2011. The accuracy was evaluated by comparing with a reference map derived from Terra/ASTER NDWI images. As a result, the F-measure was 92.0% in the inundation area extraction by U-Net using the three layers of the pre-event, post-event and the difference between pre- and post-event COSMO-SkyMed images. The F-measure of the U-Net was superior to those of the threshold method. In conclusion, the U-Net is more effective than the threshold method for inundated area extraction using X-band SAR images.