Modern simulation environments generally do not contain high fidelity Civilian Infrastructure (CI) features, or they lack the necessary content (e.g., data model and attribution) to simulate the second and third order effects caused by a critical component outage. This problem can often be traced back to the Geographic Information Systems (GIS) data used to create the simulation environments. Although GIS data repositories, like OpenStreetMap or National Geospatial-Intelligence Agency's (NGA) Geospatial Repository and Data (GRiD), contain some level of CI feature content, this content is often incomplete and does not contain enough metadata to properly model the relationships and dependencies between CI features. Therefore, a new approach is needed for the simulation and training community that leverages and enhances the existing CI content. 
 

In this paper we describe an automated method for collecting CI features from various image sources, including satellite, street view, and drones by leveraging machine learning (ML) modules to detect and georegister multiple CI features simultaneously. We define a framework for orchestrating the execution of ML modules within Docker containers, which establishes a set of services for CI feature collection. This framework will be flexible to allow integration of new feature detection modules, adjustable to execute ML services in a variety of permutations, as well as scalable to handle very large areas of coverage by utilizing a distributed processing environment for feature detection. Automating the process of CI feature collection will result in significant savings in time and money when compared to manual collection efforts.