Deep Learning For Nuclear Safeguards

Erik Wolfart - European Commission - DG JRC
Carlos Sanchez Belenguer - European Commission - Joint Research Centre
V. Sequeira - Joint Research Center -- Ispra
File Attachment
a349.pdf1.18 MB
Nuclear inspectors benefit from location-based services in several ways: i) to independently verify the current position and navigate within nuclear facilities; ii) to access to information that is contextual to the current position, for example to measurements, notes and observations acquired during previous inspections; iii) to carry out specific tasks according to the current location. Furthermore, the inspector can tag measurements and observations taken during the inspection with the current position to facilitate later analysis of the data and increase the efficiency of follow-up inspections. Current mobile devices typically use GPS to acquire position information, i.e. they cannot be used for inspection activities inside nuclear facilities. This paper presents an indoor localization system developed for nuclear safeguards applications, which recognizes the current location using visual information. In an off-line mapping phase, we use a 3D laser sensor mounted on a backpack with a calibrated spherical camera i) to generate the data for training a deep neural network and ii) to build a database of georeferenced images for an environment. Thanks to the 3D laser measurements and the spherical panoramas, we can efficiently survey large indoor areas in a very short time. The underlying 3D data allows us to identify images observing the same place and effectively train a deep neural network that maps an image to a signature, which is representative for the given location. During the online localization phase, the inspector acquires an image and queries the trained network to efficiently retrieve the location of the most similar signature in the database of georeferenced images. The paper presents the architecture of the underlying neural network and shows how the concept can be applied to other applications in nuclear safeguards,e.g.for verifying spent nuclear fuel using gamma emission tomography. In this case, the network can be trained to map the sinogram generated by the tomograph to a unique signature, which is representative for the given fuel assembly layout and which can then be used to verify the operator declaration.