Diversion Detection using Optical Surveillance based on Deep Learning

Automatic detection technique of diversion in optical surveillance without supervision or very limited supervision would be very helpful for the effective and efficient safeguards. Recently the deep learning techniques were applied to the automatic detection of anomaly in the optical surveillance. A generative model was learned only using the normal pattern of optical surveillance, and the anomaly in the optical surveillance can be captured using the generative model. In the present presentation, we introduce our effort to apply these approaches to the timely detection of the diversion in the pyroprocessing facility. We made virtual optical surveillance data of normal and abnormal pattern inside electro- refining process cell of the pyroprocessing facility. A generative model based on autoencoder is produced. The model is learned only using the normal surveillance data. We attempt to detect the abnormal moment in the surveillance data mixed with normal and abnormal scenes, and our model can produced quite notable result. Our effort could give some hint about the surveillance based on deep learning could help the efficient and effective safeguards also security.JustificationsOptical surveillance is very important and traditional safeguards technology. However, it is difficult to detect the diversion only using the optical surveillance in timely manner. Recently the deep learning technique is applied to the anomaly detection including the video surveillance, and they produce notable result. Since the presentation or papers about application of the deep learning technique to optical surveillance for safeguards purpose is not many, our work is unique.Short biographical sketchSe-Hwan Park received his Ph.D. degree (Nuclear Physics) in 2002 at Seoul National University in Korea. He has worked at the Korea Atomic Energy Research Institute (KAERI) since then. He has research interests in the development of frontier technology for safeguards and security, and design of safeguards system of the pyroprocessing facility.