Publications Details

Class 2 MIVS image processing system (MIPS) evaluation

Horton, R.D.; Whichello, J.V.

The Modular Integrated Video System (MIVS) Image Processing System (MIPS) was developed to assist the International Atomic Energy Agency (IAEA) Inspectorate in the safeguards review of MIVS surveillance data. MIPS is designed to review MIVS surveillance data automatically; firstly detecting significantly changed images and secondly identifying if the changed images show IAEA defined objects of safeguards interest. To achieve this, MIPS uses both digital image processing and neural network techniques. A change detector uses image processing techniques to identify significantly changed images. The MIPS neural network classifier detects images which may show an important object(s). The neural network is trained , i.e., given example images showing the objects that it must recognize, for each different facility. The MIPS change detection algorithms reduce the original MIVS data by eliminating images without significant activity. The MIPS neural network algorithms further reduce this data by eliminating images which may not show a safeguards significant object. The images analyzed by both the change detector and the neural network are available for inspector review. The MIPS algorithms are implemented in commercially available hardware. A high-level menu-driven system interface allows inspectors to train the neural network and to operate both the change detection and neural network classification. An evaluation program was conducted jointly by Sandia National Laboratories (SNL) and the IAEA to determine the systems capabilities on a variety of MIVS data. The MIPS processing techniques and the user interface were evaluated by IAEA inspectors. Performance tests were also completed on a variety of MIVS data. This paper provides a description of the Class II MIPS and the evaluation program and reports on the results of this joint evaluation. 7 refs., 4 figs., 3 tabs.