藏匿爆炸物的太赫兹光谱智能化识别方法

The goal of this proposal is to establish a coupled recognition system of explosives. The two critical issues of this mater are intelligent identification of explosives and the algorithm for reducing the interference caused by the barriers. Firstly, a terahertz time-domain spectroscopy (THz-TDS) system in transmission mode will be built on the basis of trace and hiding features of explosives. Then the THz spectra of the typical pure and mixed explosives will be investigated based on a large mount experiments on this system, and also the influences of different barriers on the THz spectrum of explosives will be investigated. Secondly, we will analyze the associated relationship between the different barriers and the spectra of explosives, and investigate the algorithm for reducing the interference caused by the barriers, and study the approach for automatic identification of explosives based on artificial neural network algorithm. Lastly, we will investigate the intelligent identification of explosives under different barriers in the simulative scenario of security checks, and establish the coupled recognition system. The innovations of this research are joining the quantum chemistry calculation to the detection of THz spectra of mixed explosives, and combining the principal component analysis with the artificial neural network, which provides a new effective method for the identification of THz spectroscopy. It is anticipated that this research can provide a referable method for identification of substances using THz technique and promote the development of THz technology.

项目以建立藏匿爆炸物的太赫兹光谱耦合识别系统为研究目标，拟解决的关键问题是爆炸物的太赫兹光谱智能化识别方法和规避不同包装物影响的屏蔽算法。首先基于爆炸物的痕量和藏匿特征，建立透射式太赫兹时域光谱研究平台。然后通过大量实验及量子化学计算，分别研究典型单质及混合爆炸物的太赫兹指纹谱性和不同包装物对光谱的影响。其次分析不同包装物与爆炸物太赫兹光谱之间的关联关系，研究规避包装物影响的屏蔽算法；同时引入神经网络工具箱，建立基于人工神经网络组合算法的爆炸物识别方法。最后模拟机场安检场景，研究爆炸物在不同包装物下的智能化识别方法，建立耦合识别系统。项目创新之处在于，不但将量子化学计算方法全面引用到混合爆炸物太赫兹光谱探测试验中，详细解析太赫兹光谱；而且将结合人工神经网络算法与主成分分析方法，为太赫兹光谱识别方法带来创新方向。并期望为太赫兹技术用于物质识别提供可借鉴的光谱分析方法，推动太赫兹技术发展应用。

近年来社会政治经济形势复杂，恐怖事件时有发生，因此研究针对爆炸物等危险品快速无损的检测方式具有重要意义。爆炸物等高危化学品在太赫兹频段具有独特的光谱特征，且较红外光谱而言，特征峰更加具有“唯一性”，因此太赫兹技术非常适用于爆炸物等高危化学品的快速准确探测。本项目设计了一种基于差分吸收原理的远距离太赫兹探测仪器。采用二阶微扰理论和密度泛函理论两种方法，结合非谐性校正计算，对芥子气及其模拟剂进行了详细的结构研究和太赫兹光谱模拟，并分析了特征峰的起源。结果显示二者太赫兹谱特征峰不同，说明对于这种难以通过红外光谱进行区分的结构相似化合物，能够通过太赫兹光谱进行鉴别。采用独立成分分析 (ICA) 方法对混合物太赫兹光谱进行了定性及定量分析。对于定性分析，ICA方法很好地从混合物光谱中提取各组分光谱，计算光谱与实测光谱相关性最高可达0.96，表现出强相关性，可进行混合物中各组分的精确识别。对于定量分析，ICA方法计算所得预测值与实际值的均方根误差可低至0.05，预测值与实际值吻合度较高，说明ICA方法可以高精度地预测混合物中各组分的相对浓度。为了提高指纹谱识别的性能，将经验模态分解 (EMD) 方法应用到太赫兹光谱分析中， IMF1清晰地显示了高频区的指纹谱，特别是对于低浓度和弱吸收峰的物质。结果还显示，无论数据质量如何，EMD结合傅里叶变换都能在光谱指纹谱识别中具有优越的性能。