We propose an integrated solution to reduce false alarms and increased positive identification in exclusive detection systems (EDS). To effectively combat complexity and uncertainty in a major interpretation, the project concentrates on to novel approaches: (1) hyperspace data mining algorithms to determine the non-linear relationships between feature overlap/variations (FOV models) and classification performance criteria. And (2) a soft expert classifier (SEC) and by multisource-multisensor decision fusion using multiple imaging modalities (CT, x-ray and NQR). The FOV models characterize how false alarms and positive detection are related to FOV, and they offer vital insights into how EDS performance is affected by material properties, thus helping designed more effective classifiers. In decision fusion, SEC offsets the demerits of one paradigm by the merits of another, by applying neuro fuzzy expert systems and evolutionary computing to multi-source information. Combining the human knowledge and non-linear FOV models obtained from explosive data, SEC is expected to outperform other classifiers based on any single technology. Advanced algorithms using statistics and fuzzy sets will also be developed for image segmentation, feature selection and reduction, and object recognition. Anticipated results and potential commercial application of results: the result is a prototype PC software with GUI's and algorithm modules for segmentation, material property modeling, and explosive classification. The multisource-multisensor information fusion algorithm is expected to work with image features generated by most imaging modalities (sensors) in the EDS industry. Phase I research on algorithm design will pave the way for developing a real-time, machine-independent product for most EDS is in Phase II. The product has immediate application to carry-on luggage inspection, weapons inspection, narcotics detection, quality controls, aircraft subsurface inspection, equipment diagnoses and vision systems.
Keywords: explosive detection, data mining, decision fusion, neuro fuzzy classifier move in