Fruit Sugar Content Near Infrared Online Detection Model Establishment And Optimization

In this study, Gannan navel orange as experimental subjects, mainly sugar content of itssingle internal quality as an indicator. By sample spectrum, spectral data analysis and thedetermination to the physical and chemical indicators, on-line detection device detectioneffect was studied, different pretreatment methods of comparative analysis, the comparison ofdifferent variable selection method, using different modeling methods for modeling andsimple, the stability of the calibration model is set up. Conclusions are as follows:1. In this experiment, orange sugar content for the detection of targets, the navel of thesame species were used diffuse reflectance and diffuse transmission in two different ways ofdetecting near-infrared detection device online, near-infrared spectral data acquisition orangesugar content. Compared the diffuse reflection and diffuse transmission spectroscopy navelorange your hydrometer correction model is set up, respectively was not involved in themodeling effect of experimental samples model validation, the research results show thatusing diffuse transmission detection ways of near infrared online detection device built modelis superior to the diffuse reflection mode, detection effect is better.2. To eliminate dynamic online collection of spectral information contains the influenceof the interference of other useless information, diffuse and diffuse transmission of two kindsof detection methods collected spectral preprocessing. Contrast analysis of the standardnormal variables transform (SNV), a derivative of the S-G and multiple scatter correction(MSC) three pretreatment methods on the effect of spectral information processing, the resultsshowed that both diffuse, or diffuse transmission spectra were collected to choose MSCpretreatment method for spectral data processing effect is better, the prediction correlationcoefficient, diffuse reflection and diffuse transmission were0.67and0.87, predict root meansquare error, diffuse reflection and diffuse transmission0.85oBrix and0.40oBrixrespectively.3. Comparative analysis of the reverse interval partial least squares (BiPLS), movingwindow partial least squares (MWPLS), genetic algorithms (GA), the reverse interval partialleast squares-genetic algorithms (BiPLS-GA) and move the window partial Least Squares-genetic algorithms (MW-GA) five variable selection methods, can be a good spectrum ofredundant information removed. Which combination of variables BiPLS-GA optimalscreening method is the best one of all screening methods to predict results. The number of variable filter194, only16.4%of the original spectrum, greatly reducing the number ofvariables in the modeling, but also very effective in improving the ability to predict the PLScalibration model. The correlation coefficient Rc and Rp0.93and0.90, respectively, and theroot mean square error RMSEC RMSEP were0.30oBrix and0.36oBrix.4. To choose the best modeling method, comparative analysis of the principal componentregression (PCR), partial least squares (PLS) and least squares support vector machine(LS-SVM) modeling approach to predict the effect, from the experimental the results can bedrawn from a combination BiPLS-GA PLS modeling method for screening variables toestablish calibration model to predict the best performance.