Deep Learning Based Detachment Segmentation: the MIRET Approach

Among infrastructure diagnostics, the maintenance of Transportation Tunnel (TT) is of paramount importance. On the one hand, a predictive maintenance process would allow for safety compliance, preventive corrective actions, and increased lifetime for the asset, which results in improved efficiency of the overall infrastructure. On the other hand, existing methods to perform diagnosis and detection of these anomalies are inexpensive and time-consuming. Thus, ETS and RMT foresee the development of deep learning-based methods for the segmentation of defects on TT images. This stands as the technological pillar of two innovative ETS projects: the innovative multi-dimensional survey system (ARCHITA), and a new approach for the Management and Identification of the Risk for Existing Tunnels (MIRET). The focus of this communication is the segmentation of superficial detachments on data of concrete type Transportation Tunnels, using convolutional neural networks trained in an adversarial fashion.