Qualitative detection of nitro-aromatic explosives using supervised learning access

Abstract

— Nitrobenzene and Nitrotoluene are potential explosives and pose a threat to mankind. As direct sensors for detection of these nitro-aromatic compounds are not available, an array of four gas sensors, sensing the aroma of explosives, along with a temperature and humidity sensor are exposed to varying concentrations of the explosives. An arduino based data acquisition system acquires the sensor arrays response and transmits it to a computer. Feature parameters of Area, Slope and Relative Response are extracted from the sensor response and are used to train and test for presence of explosives using supervised learning algorithms. After a comparative performance study of various such algorithms, the feedforward neural network with resilient backpropagation is employed for the detection of these explosives. The system is tested for 51 cases, where the explosive is mixed with air and not a pattern gas. The system correctly identified the presence of nitrotoluene and nitrobenzene with an accuracy of 94%. A user interface is developed for easy use of the system, which allows the user to set the training mode or testing mode of the system. This interface, pops up a message when it detects the presence of nitrobenzene or nitrotoluene before the explosion.