基于神经网络和傅里叶光谱分析的易制毒混合气体检测方法

易制毒气体识别对于抑制毒品流通具有重要作用，但目前关于易制毒气体浓度检测的研究还不成熟。针对易制毒混合气体检测的问题，通过采集傅里叶红外光谱信息建立了反向传播(Back Propagation, BP)神经网络模型。以乙醚和丙酮的混合气体实验为例，对BP-傅里叶红外变换光谱(Fourier Transform Infrared Spectroscopy, FTIR)模型进行了验证和分析。结果表明，利用BP-FTIR吸收系统采集的多组分混合气体的光谱数据的总体回归R值为0.99273，相关性强。在混合气体测试中，乙醚气体的最大预测误差为28 ppm，丙酮气体的最大预测误差为11 ppm，总体预测误差较小，说明该模型能够较好地预测乙醚丙酮混合气体的浓度。因此，神经网络模型对多组分易制毒气体进行浓度反演的预测结果精度较高，本研究也为易制毒及其他混合气体检测提供了新的思路。

Poisonable Mixed Gas Detection Method Based on Neural Network and Fourier Spectrum Analysis

The identification of gas which is easy to produce drugs plays an important role in inhibiting the circulation of drugs, but the current research on the concentration detection of gas which is easy to produce drugs is not mature. In this paper, the back propagation (BP) neural network model is established by collecting Fourier infrared spectrum information for detecting the gas mixture that is easy to produce drugs. The model of BP-FTIR is verified and analyzed by taking the mixed gas experiment of ether and acetone as an example. The results show that the global regression R value of the multi-component gas spectrum data collected by BP-FTIR absorption system is 0.99273, and the correlation is strong. In the mixed gas test, the maximum prediction error of ether gas is 28 ppm, and the maximum prediction error of acetone gas is 11 ppm. The overall prediction error is small, indicating that the model can predict the concentration of ether acetone mixture well. Therefore, the prediction result of concentration inversion of multi-component gas which is easy to produce drugs by neural network model is highly accurate, and this study also provides a new research idea for the detection of gas which is easy to produce drugs and other mixed gases.