Neural network analysis of Terahertz spectra of explosives and bio-agents

A proposed, non-invasive, means to detect and characterize concealed biological and explosive agents in near real-time with a wide field-of-view uses spatial imaging of their characteristic transmission or reflectivity wavelength spectrum in the Terahertz (THz) electro-magnetic range (0.1-3 THz). Neural network analyses of the THz spectra and images will provide the specificity of agent detection and reduce the frequency of false alarms. Artificial neural networks are mathematical devices for modeling complex, non-linear functionalities. The key to a successful neural network is adequate training with known input-output data. Important challenges in the research include identification of the preferred network structure (e.g. multi-layer perceptron), number of hidden nodes, training algorithm (e.g. back propagation), and determination of what type of THz spectral image pre-processing is needed prior to application of the network. Detector array images containing both spectral and spatial information are analyzed with the aid of the Neurosolutions™ commercial neural network software package.