Results From Application of Time Series Concepts to Vehicle Gamma Count Profiles

Abstract

Algorithms based on time-series analysis techniques were explored for maximizing the effectiveness of pass-through radiation portal monitors for detection of special nuclear material (SNM). Time-series properties of vehicle count profiles such as stationarity and autocorrelation within energy windows were characterized. Vehicle count profiles were nonstationary but were found to be made stationary by first-differencing. Autocorrelation functions showed consistent differences between NORM alarm and non-alarm vehicles. Injection studies were performed to assess the performance of time-domain detection algorithms based on stationarity tests and on the CUSUM change-point detection test. Results indicated possible roles for detection algorithms based on statistical process control and on time series concepts.