Article Preview
TopIntroduction
The increasing concern over terrorist attacks with nuclear weapons in the last decade necessitated the development of new countermeasures for enhancing national security, safeguards and public safety. Detection and identification of illicit materials potentially convertible into nuclear weaponry, known as special nuclear material (SNM), is one of the last barriers against nuclear terrorism. The search for hidden SNM focuses on two main points: in scanning cargo containers in a country’s entry ports and in urban searches with mobile detectors.
One example of nuclear signals is the gamma spectra. Measurement of gamma radiation is one of the most common techniques for detection of threats. Gamma signals are processed in order to extract a nuclide’s gamma signature from the overall signal. In general, a gamma detector provides a signal which is the aggregation of nuclides’ signatures and ambient background radiation.
Analysis of measured signals is a significant aspect of the overall detection process. Its goal is to recognize the signatures of interest and subsequently identify the respective material. Efficient analysis requires a high positive detection and a low false alarm rate. Computationally, the efficiency of the analysis must be able to handle the large volume of incoming data, which requires fast processing and decision-making. Clearly, every developed algorithm aims to maximize efficiency and minimize processing time.
Nuclear spectrum processing is an active research area and a significant amount of methodologies have been proposed. Artificial intelligence tools have found wide use in analyzing detector signals and recognizing hazardous materials. Neural networks have been used for analysis of gamma spectra as presented for germanium detectors by Yoshida et al. (2003) and for scintillator detectors by Kucuk & Kucuk (2003). Furthermore, the use of neural networks for real time analysis of gamma-ray signals and subsequent identification of radioactive isotopes is introduced in (Keller & Kouzes, 1993), while a multi-level perceptron was used for improving signal-to-noise-ratio in measurements as presented in (Yearworth & Miller, 1993). Genetic algorithms were also applied for detector signal analysis as shown in (Carlevaro et al. 2008) with the drawback that those methods are seemingly effective only for simple spectra. A methodology integrating evolutionary computation and fuzzy logic was introduced in (Huang et al. 2006), while expertise if-then rules were introduced in (Lehner at al. 2003) and in (Alamaniotis et al. 2009) for intelligent processing for gamma-ray signals. Other methodologies based on fuzzy logic approaches were proposed in (Murray et al. 2000) for CZT signals, and in (Alamaniotis et al. 2011) for Nuclear Resonance Fluorescence (NRF). The intelligent technique of LG-graphs was applied on radiation signals by Bourbakis et al. (2008) and by Pantelopoulos et al. (2009).
In a previous paper (Alamaniotis et al. 2011), the authors have applied their developed smart algorithm for processing of signals acquired with a nuclear detector. The underlying principles of intelligent information processing of their algorithm are based on the well-developed tools of fuzzy logic (FL) (Tsoukalas & Uhrig, 1997). The FL tools are used for maxima identification as well for making inferences over the sources identified in a signal. The advantage of FL Radiation Isotope Identifier (RIID) algorithm is that it does not require training and supports automated and expedient analysis of spectral content without regard to the complexity of the measured signal. At the same time, uncertainty and ambiguity over detected results is handled with the computation of decision certainty factors.