Development of a Process for Identification of the Operational Mode of Industrial Sites Using High Dimensional Multi-modal Data

Many algorithms exist to determine the physical contents of an image. Target detection or anomaly detection algorithms, for example, use statistical and geometric approaches in high dimensional space to locate objects within a scene. Instead of target detection, however, it has become of interest of late to delve deeper into the field of remote sensing in order to perform process detection. Process detection refers to the ability to identify the operational mode of an industrial facility. To accurately complete this task will require a new set of analysis tools.;This thesis discusses a method that can be used to perform process detection with multi-modal remotely sensed data. Using a local industrial facility, operational modes were identified, as well as the subtle differences between them. Combinations of hourly data, sparse data, and latent variables were combined through analytical tools and a prediction of the process taking place at different moments was performing using both real and simulated data sets.;An advanced analyst environment is also discussed, with a few demonstrations from a test environment developed by a small team at RIT. Temporal analysis, multi-modal data integration, and the use of process models to make latent observables are discussed. This thesis shows the utility of such an environment and demonstrates the need for the further development.