REDUCTION OF NUISANCE ALARMS IN MULTIPLE PHENOMENOLOGY PERIMETER PROTECTION SYSTEMS USING A HIGH ORDER NEURAL NETWORK PROCESSOR

Multiple sensors have been used for many years for the perimeter protection of high value facilities in order to ensure high probability of detection of intrusions. Unfortunately, most exterior protection systems have a problem with environmentally induced nuisance alarms. GRC has developed and tested a high order neural network processor specifically to deal with the improvement of nuisance alarm performance in multiple phenomenology perimeter protection systems. The processor's ability to learn the distinguishing characteristics of sensor signals allow it to recognize the difference between signals caused by nuisance events and those caused by intrusions. GRC's processor has been tested on a dual phenomenology ported coax and microwave sensor field for a set of 120 intrusion events which included large animals and vehicles as nuisance sources. Results which were achieved showed a system probability of detection of 1.0 with a probability of nuisance alarm of 0.1 for real-time classification. The ability of neural networks to learn about new situations allows GRC's processor to tailor the performance of individual sensor systems to site-specific nuisance sources. Using GRC's processor, the security risk managers can now rely on their sensor systems to a greater degree to contribute to the overall security of their facilities.