摘要
建立一种基于红外光谱的快速无损地检验洗发用品的分析方法。利用傅里叶红外光谱对60个常见的洗发用品样品进行检验,分别采用Savitzky-Golay(S-G)平滑、快速傅里叶变换(FFT)、降噪等方法对光谱数据进行预处理,并结合主成分分析法对光谱数据进行降维处理。同时建立多层感知器神经网络和贝叶斯判别分析两种分类模型,对光谱数据进行分析验证。多层感知器神经网络对原始数据、经过S-G平滑、FFT、降噪后的分类准确率分别为86.67%、88.33%、80%、90%,贝叶斯判别的分类准确率为83.33%、85%、83.33%、95%。结果显示,降噪处理效果较佳,贝叶斯判别具有更高的准确率。该方法重现性好、样品用量少、无损样品,可为洗发用品类物证鉴定提供科学依据。
To establish a rapid and non-destructive analytical method for testing shampoo products,sixty common shampoo samples were tested using Fourier transform infrared spectroscopy.The spectral data were preprocessed using Savitzky Golay smoothing,fast Fourier transform FFT,and noise reduction methods,respectively.The spectral data were then dimensionally reduced using principal component analysis.At the same time,two classification models,multi-layer perceptron neural network and Bayesian discriminant analysis,were established to analyze and verify spectral data.The classification accuracy rates of the multi-layer perceptron neural network for raw data,S-G smoothing,FFT,and noise reduction are 86.67%,88.33%,80%,and 90%,respectively.The classification accuracy rates of Bayesian discrimination are 83.33%,85%,83.33%,and 95%.The results show that the effect of noise reduction processing is better,and Bayesian discrimination has a higher accuracy rate.The method has good sample reproducibility,small sample consumption,and non-destructive samples,which can provide scientific basis for the identification of physical evidence of shampoo products.
作者
姜红
周贯旭
周飞翔
郝小辉
JIANG Hong;ZHOU Guanxu;ZHOU Feixiang;HAO Xiaohui(Criminal Investigation Department,Gansu Police Vocational College,Lanzhou 730046)
出处
《分析科学学报》
CAS
CSCD
北大核心
2024年第1期75-80,共6页
Journal of Analytical Science
基金
甘肃省哲学社会科学规划项目(2022YB129)。
关键词
傅里叶变换红外光谱
洗发用品
主成分分析
多层感知器神经网络
贝叶斯判别分析
Fourier transform infrared spectroscopy(FTIR)
Shampoo products
Principal Component Analysis
Multilayer Perceptron Neural Network
Bayesian discriminant
