Pattern Recognition Algorithms for Comparing NWIS Signatures for Weapons Components

Previous experience with the use of Nuclear Weapons Identification System (NWIS) time and frequency signatures has shown that their large number (19 for a source and two detectors) makes manual and/or visual examination impractical. Pattern recognition methods can be used to compare signatures more quickly and reliably. The pattern recognition algorithm utilized compares a particular NWIS signature with that for another item, as follows. The deviation of the signature of the item of interest (Ui), at a frequency, time lag, or multiplicity i, from the reference (Ri) in units of standard deviations ( i), is calculated as [(Ui – Ri)/ i]2. If multiple signatures are compared, this average is taken over all signatures. This can be applied to all 19 signatures. The standard deviation of the signature i is obtained by repeated measurements on reference weapons components in containers. An unknown component in a container matches the signature class when the average deviation lies within the statistical bounds prescribed by the statistical degree of confidence desired. The application of this technique on measurements on Uranium castings and weapon components in containers is illustrated.