利用太赫兹时域光谱技术(THz-TDS)对11种常见毒品进行了实验探测并得到它们在0.2～2.6 THz范围内的吸收谱, 然后利用径向基(RBF)神经网络对11种毒品的太赫兹吸收光谱进行了训练和识别。训练和识别的光谱在输入网络之前都经过归一化处理, 识别率达到了96%。RBF神经网络模型是用Matlab语言编制。识别结果表明, 利用RBF神经网络可以实现对不同种类毒品的识别, 而且训练和识别的速度非常快, 几乎是实时的。这也是RBF神经网络优于误差逆传播(BP)神经网络和自组织特征映射(SOM)网络的特点。因此, RBF神经网络为太赫兹技术用于毒品的检测和识别提供了一种有效而快捷的方法。

Absorption spectra of 11 illicit drugs were measured by using the terahertz time-domain spectroscopy technique (THz-TDS) in the range 0.2～2.6 THz, and then the absorption spectra of the 11 different illicit drugs were identified successfully by well trained radial basis function (RBF) neural networks. The spectra used for training and identification were normalized before input to the network and the positive identification rate is 96%. The model of RBF neural network was processed in Matlab. The results indicate that it is feasible to apply RBF neural network to the identification of illicit drugs, and the speed of training and identification is fast. The character makes RBF network superior to back propagation neural network (BP) and self-organizing feature map (SOM). Therefore, RBF neural network provides a fast and effective method in the secure inspection and identification for illicit drugs using THz-TDS technique.