Model Research For FT-NIR Analysis Of Sugarcane Juice Brix With Different Pathlength And Resolutions

Cane sugar manufacture includes a number of process links,and the process quality analysis of sugar cane sugar production has always been an important task in the sugar industry.Clear juice is the product of sugar cane sugar production,and hammer is an important indicator of quality analysis.The traditional measurement method requires sampling for laboratory measurement,which is cumbersome and time-consuming,and has little guidance on real-time production.Fourier transform near-infrared spectroscopy?FT-NIR?has the advantages of fastness,convenience,good reproducibility and high wavelength accuracy.It has been applied to the rapid detection of sugarcane process products.Resolution and optical path are two important parameters.High resolution can improve the fine spectral characteristics of the sample,but often leads to a decrease in signal-to-noise ratio;long optical path can improve the spectral absorption information of low-level components,but it is easy to cause the overall saturation absorption,resulting in noise interference.The resolution of near-infrared spectroscopy and the parameter adaptability of optical path are important research topics.In this paper,the FT-NIR analysis of sugarcane juice is used as the research object,and the adaptability of different optical path and resolution models is studied.Under different optical paths?2mm,5mm,10mm?and different resolutions(4cm^-1,8cm^-1,16cm^-1,32cm^-1),12 sets of spectra were obtained,and quantitative analysis models were established and compared.Firstly,based on the multiple division of calibration and prediction sample sets,the partial least squares?PLS?regression model of FT-NIR analysis of sugarcane juice is established,and the accuracy of the model is evaluated by the average prediction effect.Secondly,the derivative order is constructed.?d?,smoothing points?s?,differential spacing?g?full-circulation Norris-PLS modeling platform,preferably optimal Norris parameters for each set of samples,for spectral preprocessing;finally,constructing the starting wavelength?I?The number of wavelengths?N?,the number of PLS latent variables?LV?full cycle MW-PLS modeling platform,the optimal wavelength model for each set of samples is preferred.The models are compared according to the same optical path and the same resolution.The main results of this paper are as follows:?1?Modeling and comparison of the same optical path and different resolutions1)Full-spectrum PLS model:The optimal resolution of the model is 4cm-1 under 2mm,5mm optical path.The optimal resolution of the model is 32 cm^-1 at 10 mm optical path.The average deviations?SEP⁺?between the predicted values and the measured values of the above three optimal models are 0.522,0.523,and 0.531?%?,respectively,and the relative errors?RSEP⁺?are 3.3,3.3,and 3.4?%?,respectively.2)Norris-PLS model:Under the 2mm optical path,the optimal resolution of the model is 8cm^-1.The optimal resolution of the model is 4 cm^-11 at 5 mm and 10 mm optical path.The SEP⁺of the above three optimal models are 0.481,0.487,and 0.473?%?,respectively.RSEP⁺ is 3.06,3.10,and 3.01?%?,respectively.3)MW-PLS model:Under the 2mm optical path,the optimal resolution of the model is 32cm^-1.The optimal resolution of the model is 4 cm^-1 at 5 mm and 10 mm optical path.SEP⁺ is 0.472,0.470,and 0.464?%?,respectively.RSEP⁺were 3.00,2.99,and 2.95?%?,respectively.?2?Modeling and comparison of the same resolution and different optical paths1)Full-spectrum PLS model:The optimal optical path of the model is 2mm at 4 cm^-1,8 cm^-1,16cm^-1,32 cm^-1 resolution.The SEP⁺of the above four optimal models are 0.522,0.526,0.527,and 0.489?%?,respectively.RSEP+is 3.32,3.35,3.35,3.11?%?.2)Norris-PLS model:The optimal optical path of the model is 10 mm at 4 cm-1 resolution.At 8cm^-1,16 cm^-1,32 cm^-1 resolution,the optimal optical path of the model is 2mm.The SEP⁺of the above four optimal models are 0.473,0.481,0.496,and 0.482?%?,respectively.RSEP⁺is 3.01,3.06,3.16,and 3.07?%?,respectively.3)MW-PLS model:The optimal optical path of the model is 10 mm at 4 cm-1 resolution.At 8cm^-1,16 cm^-1,32 cm^-1 resolution,the optimal optical path of the model is 2mm.The SEP+ of the above four optimal models are 0.464,0.474,0.476,and 0.472?%?,respectively.RSEP⁺ is 2.95,3.02,3.03,and 3.00?%?,respectively.Based on the spectral preprocessing and parameter optimization of the wavelength model,the effects of different optical paths and resolutions on the FT-NIR analysis model of sugarcane juice broth are systematically studied.Through comparative study,the MW-PLS model with the measurement parameter 10mm optical path 4cm^-1 resolution achieves the optimal prediction effect,and the optimal Norris parameters and optimal bands are given.These results provide valuable reference for the establishment and improvement of sugarcane sugar near-infrared analysis model,which is of great significance for the quality analysis of sugarcane sugar production process in China.