Automated Explosive Threat Discrimination Using Enhanced Multiple Input Detection Systems with Decision Fusion
Landmines and improvised explosive devices pose a significant problem in many countries around the world. However, the technologies behind explosive-orientated decision systems are still limited and prone to high false alarm rates and low probabilities of detection. Some particular problems include distinguishing between threat and clutter signals, and the heavy reliance on human operators to perform these discrimination tasks. The aim of this research is to design an automated decision system for detecting explosive threats. This will involve implementing to a certain degree a number of decision system components such as alarm detectors, feature extractors, classifiers, and decision fusers. The input data to the system will range from metal detectors, ground penetrating radar, to possibly infrared. The involvement of multiple sources may also lead to high level fusion using the decisions generated from the different detector inputs. The intended outcome is to produce an automated decision system capable of achieving improved probabilities of detection and reduced false alarm rates.
A wavelet transform based technique has been experimented with as a possible feature extraction method for metal detector signals [1]. This has also been coupled with Fuzzy ARTMAP (FAM) neural networks for target classification, together with majority voting for decision fusion from an ensemble of FAM networks. The experiments using these techniques have produced promising results. Further research is being conducted in the areas such as improving the generated feature vectors of the wavelet technique, utilizing more sophisticated methods for decision fusion (eg. Bayesian decision theory, Dempster-Shafer theorem, fuzzy integral), and incorporating additional sources into the decision system (eg. ground penetrating radar, infrared).
References
[1] Tran, M.D.J., Abeynayake, C.: Evaluation of the Continuous Wavelet Transform for Feature Extraction of Metal Detector Signals in Automated Target Detection. New Advances in Intelligent Decision Technologies, Springer-Verlag, 245-253, 2009.