Neural Network Recognition Of Terahertz Spectra Of Explosives

Terahertz (THz, 1THz=1012Hz) radiation with a frequency range of 0.1THz~10THz, sandwiched between microwave and infrared, belongs to far infrared region. The THz spectroscopy technique, as a new THz detection technique, has been widely used in the sensing and imaging field since it has the advantage of insensitive to thermal background and high signal-noise ratio comparing with the traditional Fourier transform infrared spectroscopy (FTIR).The detection of explosives and related compounds (ERCs) is more important for the national security and defense because of increasingly rampant international terrorism. For a variety of explosive molecules rotation and vibration spectra are in the range of 100GHz to 10THz, the THz-TDS has been used in the detection and identification of explosives non-destructively, non-ionized and high-sensitivity. The measurement of explosives spectra is an important part of THz spectroscopy research and an important basis for detection and identification of explosives, drugs and other illegal substance with artificial neural networks.Chapter one introduces the basic theories of THz spectroscopy technique, including the terahertz pulses generation, the terahertz pulses detection and its characteristics. It specifies some explosive security technology for nowadays and lists research status for terahertz spectroscopy of explosives in domestic and foreign countries. Our major work and signification for this thesis were presented.The principle of optical path which was built by our laboratory, experimental measurement procedure and sample preparation were presented in detail in chapter two. Moreover, the chapter two especially emphasized on the principle part which were used in the article, including how to calculate optical constant of sample, the development of artificial neural network, the basic principle of MLP neural network and SOM neural network. Chapter three illustrates the identification of THz spectra of explosives with ANNs, 5 kinds of pure explosives and 6 kinds of composite explosives and their THz spectra are presented. After repetitive modeling and adequate training we got very high positive rate with ANNs, the results indicate that it is feasible to apply these two ANNS on the identification of different types of explosives, and it also provides an effective method in the inspection and identification for explosives using THz-TDS.In Chapter four we introduced a new proposed measure-reference-free THz transmission spectroscopy. Unlike traditional transmission spectroscopy where a sample beam and a reference beam are both needed, in this work the absorption peaks of targets are extracted from the phase of the sample alone and the phase spectra have the same feature peaks with the absorption spectra. We then applied three pattern recognition methods(SOM,MLP and Mahalanobis distance classifier) in explosive spectra classification and identification with high accuracy rate and low alarm rate. The results indicate that the reference-free technique is a viable and valid spectroscopic modality by replacing the conventional absorbance curves with phase curves as features for identification and classification. This is a new method for detecting and identifying explosives from standoff distance.In the last chapter, we summarized our work of experiment and simulated results. Terahertz science and technology for explosive detection and its potential applications are discussed.