基于人工神经网络的傅里叶变换中红外光谱法对食用油油烟种类识别研究

发布时间：2018-06-14 20:47

本文选题：FTIR + 人工神经网络　；参考：《光谱学与光谱分析》2017年03期

【摘要】：随着餐饮业的发展,餐饮烟气已经成为某些城市三大空气污染源之一。由于餐饮烟气对人体健康威胁很大,近年来对餐饮烟气的研究愈来愈热。餐饮烟气中包含有大量食用油加热过程中裂解而产生的不饱和烃类,危害着人类健康。不同食用油裂解出来的成分以及含量有所不同,通过构建一定的分类识别数学模型,从而实现对食用油分类识别。采用自主研发的傅里叶变换红外光谱仪,采集了不同食用油油烟烟气红外光谱数据。同时构建了主成分分析(PCA)分别结合概率神经网络(PNN)以及误差反向传播人工神经网络(BPANN)的分类识别算法。将两种分类识别算法对不同食用油油烟烟气的傅里叶变换红外光谱数据进行分析。通过样本数据对数学模型进行训练,将训练好的数学模型对未知光谱数据进行分析,来确定产生油烟烟气的食用油种类。实验结果表明,两种算法都能对不同的油烟种类进行较好地分类识别。在全波段识别时,识别率分别达到90.25%和97.0%。通过对烟气光谱数据的吸收波段进行分析,提取大气窗口并且具有较强可挥发性有机物(VOCs)吸收特征的波段(1 300~700 cm~(-1)以及3 000~2 600 cm~(-1));将吸光度数据分成两个分离的吸收波段,两种算法在3 000 2 600 cm~(-1)波段都有较好的识别效果,PCA-PNN算法识别率为90.25%,PCA-BPANN算法识别率为92.25%。可见,两种人工神经网络算法都能有效对食用油烟种类进行识别。
[Abstract]:With the development of catering industry, fume has become one of the three air pollution sources in some cities. Because the fume of food and beverage is a great threat to human health, the research on the fume of food and beverage is becoming more and more hot in recent years. There are a lot of unsaturated hydrocarbons produced by cooking oil pyrolysis in the fume of food and beverage, which is harmful to human health. The components and contents of different edible oils are different. A certain mathematical model of classification and recognition is constructed to realize the classification and recognition of edible oils. Fourier transform infrared spectrometer (FTIR) was developed to collect infrared spectrum data of different cooking oil fumes. At the same time, the classification and recognition algorithms of principal component analysis (PCA) combined with probabilistic neural network (PNNN) and error back-propagation artificial neural network (BPANN) are constructed. The Fourier transform infrared spectral data of different cooking oil fumes were analyzed by two classification recognition algorithms. The mathematical model is trained by sample data, and the unknown spectral data are analyzed by the trained mathematical model to determine the type of cooking oil that produces fume. The experimental results show that the two algorithms can be used to classify and identify different types of oil fumes. The recognition rates are 90.25% and 97.0% respectively. By analyzing the absorption bands of smoke spectral data, we extracted the atmospheric window and have a strong absorption characteristic of volatile organic compounds (VOCs) (1 300 ~ 700 cm ~ (-1) and 3 000 ~ (2 600) cm ~ (-1) ~ (-1). The absorbance data are divided into two separate absorption bands. The recognition rate of PCA-PNN algorithm is 90.25 and the recognition rate of PCA-BPANN algorithm is 92.25. Therefore, the two artificial neural network algorithms can effectively identify the types of cooking oil smoke.
【作者单位】：中国科学院安徽光学精密机械研究所中国科学院环境光学与技术重点实验室;中国科学技术大学;
【基金】：中国科学院战略性先导科技专项(XDB05050300);中国科学院战略性先导科技专项(XDB05040500) 国家重大科学仪器设备开发专项(2013YQ22064302) 工业区VOCs排放通量遥测方法研究项目(41405029)资助
【分类号】：X831;O657.33

【相似文献】