基于改进ACGAN样本增强的太赫兹时域光谱隐匿危险品识别方法

A THz Time-Domain Spectral Hidden Dangerous Goods Recognition Method Based on Improved ACGAN Sample Enhancement

摘要: 针对太赫兹时域光谱数据匮乏导致基于深度学习算法的太赫兹时域光谱识别准确率较低的问题，提出了一种基于改进ACGAN样本增强的太赫兹时域光谱隐匿危险品识别方法。改进ACGAN在生成器中引入残差单元，以提高生成高保真的数据。在判别器中加入长短时记忆网络提高判别数据真伪的能力。实验首先采用反射型太赫兹光谱仪系统测量酒精、煤油、食用油、乳香油、松节油、松香油、樟脑油等7类易燃易爆液体的太赫兹时域光谱数据共1260条并喂入深度学习分类模型进行训练；随后将增强后的数据集分别注入训练好的分类模型，对识别精度指标进行分析测试，并与ACGAN和Mixup进行实验对比。使用改进ACGAN对原始样本增强扩充后ResNet、CNN、FCN和MLP分类模型的识别准确率分别提高了1.4%、1.63%、0.96%、1.07%，比ACGAN、Mixup提升的幅度更高。结果表明，改进ACGAN能够有效改善训练样本不足的问题，提高模型识别精度。

Abstract: Aiming at the problem that the lack of terahertz time-domain spectral data leads to the low accuracy of terahertz time-domain spectral recognition based on deep learning algorithms, a terahertz time-domain spectral concealment identification method based on improved ACGAN sample enhancement is proposed. Improve ACGAN to introduce residual units in the generator to improve the generation of high-fidelity data. Adding a long and short-term memory network to the discriminator improves the ability to discriminate the authenticity of the data. The experiment first uses a reflec-tion terahertz spectrometer system to measure 1260 terahertz time-domain spectral data of seven types of flammable and explosive liquids, such as alcohol, kerosene, edible oil, frankincense oil, turpentine, rosin oil, and camphor oil, and feed them into deep learning. The classification model is trained; then the enhanced data set is injected into the trained classification model, and the recognition accuracy indicators are analyzed and tested, and compared with ACGAN and Mixup. The recognition accuracy of ResNet, CNN, FCN, and MLP classification models after enhanced and expanded original samples using improved ACGAN increased by 1.4%, 1.63%, 0.96%, and 1.07%, respectively, which was higher than that of ACGAN and Mixup. The results show that improving ACGAN can effectively improve the problem of insufficient training samples and improve the accuracy of model recognition.