Recognition Based Explosives Classification Of Thz Spectra Database

The terahertz(10¹²Hz) spectroscopy developed in 1990s is a new technique for spectroscopy measurement.THz region of the electromagnetic spectrum lies between the infrared and the microwave.As new non-invasive examination methods,with lower energy and non-ionization THz is being applied to various research fields:condensed matter physics,medicine, manufacturing,and space and defense industries.Pattern classification is one crucial domain of THz measurement and a key tech for practicality of THz.This dissertation mainly focuses on speciality of the terahertz spectroscopy about explosives and making further research on the key tech—the algorithm of recognition and the management of spectra database,which aim at upgrading the accuracy of the explosives classification.Therefore,the dissertation is of relative value both in theory and practice.The following is a brief summary of the works conducted in this thesis and the results achieved:1.Based on a model developed by T.D.Dorney and D.Duvillaret et al,the dissertation firstly transformed THz time-domain spectroscopy into THz frequency-domain spectroscopy by FFT for four kinds of explosives(TNT,DNT,RDX,HMX),and then work out curve in frequency from sampling data.After analyzing physical features a new algorithmic idea which combines physical features with mathematic features was brought forward.2.Analyze physical features and mathematic features of frequency-domain spectroscopy of explosives.The dissertation Extract five features:peak,amplitude,span,slope, curvature.According to the advantage and disadvantage of the five features by lots of experiment,it raises a spectroscopy matching algorithmic flow using maximum peak and curvature of similarity measure.The experimental results indicated that identifying ratio of RDX,DNT and HMX is 100 percent,and that identifying ratio of TNT reaches 95 percent.3.This dissertation try to design a classifier using Learning Vector Quantization(LVQ) for the explosives classification based on spectroscopy.For the same experimental condition,the same kind of explosive has relative consistency in absorption spectroscopy.But in diverse condition the absorption spectroscopy of explosive show diversity,thus the fixed output type is an obstacle for compatibility and expansibility of system.In allusion to above problem,this dissertation carries out a auto-producing nerve cell algorithm based on minimum learning incremental error.Make experiments on computers with this method,the results indicated that identifying ratio of RDX,DNT is 100 percent using a classifier of maximal peak,and identifying ratio for all sample exceeds 97 percent using a classifier of curvature.4.Build up an explosives classification system(ETRS) based on THz database.Some design patterns are deployed in this system to make it expansible and reusable.ETRS which is developed by Visual C++ and ACESS can edit,query,analyze and identify spectroscopy.Combined the classifier of maximal peak and the classifier of curvature, identifying ratio of explosives can be reached beyond 99 percent and algorithmic fuse is supplied.