King's College London

Research portal

Detection of extremely weak NQR signals using stochastic resonance and neural network theories

Research output: Contribution to journalArticle

Original languageEnglish
Pages (from-to)96-103
JournalSIGNAL PROCESSING
Volume142
Early online date12 Jul 2017
DOIs
Publication statusE-pub ahead of print - 12 Jul 2017

Documents

King's Authors

Abstract

Nuclear Quadrupole Resonance (NQR) signal detection is widely used for searching related substances of interest, such as explosives, petroleum, drugs, etc. NQR responses from these substances are usually very weak compared to background noise. Moreover, in some applications such as landmine detection, NQR responses decay with time quickly, and the required scanning times are usually prohibitively long. This paper presents a novel approach which can detect NQR signals of very low SNRs in such scenarios, by combining a stochastic resonance framework and neural network theory. Firstly, the approach relies on the design of a stochastic resonance (SR) system which can transform the original data into a nonlinear waveform with special SR features. Secondly, a (feedforward) robust neural network is trained to discern this nonlinear waveform accurately, in order to identify the NQR signal. Our results demonstrate that the neural network approach outer-performs traditional signal processing detection and estimation methods. Moreover, this stochastic resonance neural network approach (SRNN) can be designed to detect a variety of NQR signals which have similar NQR parameters but not identical resonant bands. The SRNN approach can also be effective in cases where both noise and radio frequency interference are strong relative to the NQR response.

Download statistics

No data available

View graph of relations

© 2018 King's College London | Strand | London WC2R 2LS | England | United Kingdom | Tel +44 (0)20 7836 5454