Near-infrared Spectroscopy Interpretation And Modeling Based On Feature Extraction Technology

Near-infrared spectral bands overlap seriously,and there is a large amount of redundant information and nonlinearity,which makes it difficult to mine and analyze the spectral feature information by common methods such as first-order and second-order integer order derivatives.In this study,a fractional order derivative(FOD)is combined with the size and variation range of chemical component content,and the FOD spectral analysis method is proposed.The depth interpretation of spectral and the establishment of universal model are studied,focusing on the feasibility of application in samples with different gradient distribution of chemical component content.The main research contents are as follows:(1)Establish a sample library with different gradient distribution of chemical component content,put aside the similarities and differences of sample chemical properties,focus on the size and variation range of the abstract content,and break the traditional method of sample classification based on chemical properties.Samples of cotton-polyester blended fabrics,diesel and pork from many industries such as textile,petroleum and food are collected.The component contents of these samples have the characteristics of large gradient or small gradient distribution,which is suitable for the detailed characterization of spectral data and chemical feature extraction by FOD.And a new idea proposed for the establishment and transmission of sharing model from the quantitative attributes of samples.(2)The 0～2 order FOD spectra of cotton-polyester with large gradient distribution are obtained in 0.1 order step.The chemical characteristic attribution of absorption peaks of cotton and polyester components are analyzed,1194nm and1790nm are the absorption peaks of cotton,which are related to the vibration absorption of C-H and O-H in cotton cellulose fiber,1681nm and 1136nm are the absorption peaks of polyester,which belong to the first overtone and second overtone of aromatic C-H fundamental stretching bands respectively.The cotton content analysis model based on FOD is established,and its prediction effect is the best at1.2 derivative,R_P is 0.9888 and RMSEP is 0.0225.The results indicate that compared with the integer order derivative,the fractional order derivative can mine the detail information of spectral feature more deeply,and can effectively improve the performance and prediction accuracy of the model.(3)The FOD spectra of diesel with small gradient distribution in 0～2 order are obtained.The different absorption modes of various C-H groups related to the chemical composition of diesel are analyzed,760nm,1146nm and 1434nm are the fourth overtone of methylene C-H fundamental stretching band,the second overtone of aromatic C-H fundamental stretching band and the combined band of methyl C-H,respectively.The aromatics content analysis model based on FOD is established,and the prediction effect is the best at 0.6 derivative,R_P is 0.9876 and RMSEP is 0.0153.The results indicate that the FOD spectral analysis method also has good analytical advantages in the application of small gradient samples,and proves the universality of this method in the samples with different gradient distribution of chemical component content.(4)The near-infrared spectral intelligent analysis system based on FOD is developed by MATLAB software.The system includes four functional modules:spectral preprocessing,spectral interpretation,modeling analysis and content prediction.Cotton-polyester blended fabric and diesel samples with different gradient distribution are selected from the sample library to test each function of the analysis system.The system can effectively achieve the spectral interpretation and modeling analysis of samples with different gradient distribution,and has intuitive and friendly interface,simple operation and comprehensive functions,which can meet the needs of online detection.