Wavelength Determination And Application Of Near Infrared Sesame Fat Detector On LD

There were several methods for measuring the oil content in oilseeds, such as the conventional Soxhlet extraction, near infrared spectroscopy and so on. Near infrared analysis technology has been applied in all industries for it’s non destructive, rapid, non polluting, on-line analysis. The common near infrared spectrometer not only has complex system and high price, but also requires specialized technology. What’s more, the traditional near infrared spectroscopy could not adapt to the online detection and the field testing without the laboratory environment. Near infrared has not been really applied in our country’s food industry, agriculture and agricultural products processing sector so far. A near infrared quality detector for produce does not appear that countryspecific, low cost, easy to use and portable.LD as the light source for produce quality detection of near infrared analysis, the light source and optical system were combined. So as to improve the stability of the detection system, reduce the size and cost, promote the application of near infrared spectroscopy in agriculture, agricultural products and food processing industry. Monitored the quality of produce in production, circulation and processing,improved the added value of agricultural products. Designed a Near infrared sesame fat detector on LD, verified the rationality of the direct detection method, verified the validity of the characteristic wavelength points, and explored the suitable modeling method. Provided reliable technical solutions, data processing methods for low cost, high precision produce detector of near infrared.Eighty kinds of white sesame samples were collected from throughout provinces. FOSS NIRSystem 6500 spectrometer, S400 spectrometer of Lengguang, CONTROL DEVELOPMENT NIR256-2.2T2 fiber optic spectrometer, JDSU MicroNIR-1700 spectrometer, and self-developed mini spectrometer based on Hamamatsu C11708 MA were utilized to scan the spectra of these samples. The oil content was detected by Soxhlet extraction method according to the national standard of NY/T1285-2007.In order to establish the model with more reliable, less background interference and high SNR, the pretreatment of spectral data is always to applying in near infrared spectroscopy analysis. Each spectrometer achieve good results after spectral data pretreatment. Interval Partial least squares(iPLS), genetic algorithm(GA), successive projections algorithm(SPA),uninformative variables elimination(UVE), regression coefficient(RC), competitive adaptive reweighted algorithm(CARS)were used to select characteristic wavelength of the oil in sesame from spectral data obtaining from five spectrometers. Combined with the theoretical analysis, the wavelength of 1 210 nm, 1 410 nm, 1 730 nm, 2 300 nm were recognized as characteristic wavelengths of the oil content in sesame.Combined with the selected wavelengths of the oil, the wavelength region that the laser diode in low cost, and the moisture characteristic wavelengths 1 310 nm, 1 450 nm,1 550 nm, which was explored early. In order to measuring other components, 1 270 nm and 1 610 nm were inserted into the detector. Finally, 8 laser diode were confirmed as the light source. Then designed a Near infrared sesame fat detector on LD and scanned spectrum of sesame in same way. Math model is established with a variety of linear methods and nonlinear methods. The model established by linear methods such as multiple linear regression(MLR), stepwise regression(SWR), principal component regression(PCR) and partial least squares(PLS) were unconspicuous. While the results of BP neural network, least squares support vector machine(LSSVM), Gaussian process regression(GPR), relevance vector machine(RVM) and random forest(RF) were better than linear methods. Among these methods, the result of random forest(RF) was best, Rc and Rp were respectively 0.9638 and 0.9295, RMSEC and RMSEP were 0.0094 and 0.0125.For Near infrared sesame fat detector on LD, the structure is good, the detection method is reasonable. Characteristic wavelength point of sesame is correct and effective. The nonlinear algorithm can overcome the shortcoming of the linear regression equation, and it is modeling suitable for a few independent wavelengths. Low cost, small size, fast and simple operation, the Near infrared sesame fat detector on LD could be used to predict the oil content in sesame quickly and exactly.