Click on “Download PDF” for the PDF version or on the title for the HTML version.

If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.

Combination of variable selection methods and Vis/NIR spectroscopy for assessing soluble solids content of “Gannan” navel oranges

Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan

Citation:  2012 Dallas, Texas, July 29 - August 1, 2012  121341101.(doi:10.13031/2013.42053)
Authors:   Tong Sun, Wenqiang Wu, Wenli Xu, Muhua Liu, Wanhuai Zhou
Keywords:   Visible/near infrared spectroscopy, Variable selection, Soluble solids content, Navel orange

Visible near infrared (Vis/NIR) spectroscopy has become a popular technique for non-invasive quality assessment of intact fruits. The objective of this research was to assess soluble solids content (SSC) of Gannan navel oranges by Vis/NIR spectroscopy and variable selection methods. The semi-transmission spectra of Gannan navel oranges were collected by a QualitySpec spectrometer in the wavelength range of 350~1000 nm. Different spectral pretreatment methods were used to remove spectra noise. After that, several variable selection methods such as successive projections algorithm (SPA), uninformative variable elimination (UVE), and successive projections algorithm combined with uninformative variable elimination (UVE-SPA) were used to find sensitive wavelengths, partial least squares (PLS) and multiple linear regression (MLR) were used to develop calibration models for SSC using sensitive wavelengths. The results indicate that standard normal variate is the best pretreatment method for semi-transmission spectra of Gannan navel oranges, the correlation coefficient (r) and root mean square error (RMSE) of PLS in prediction set are 0.922 and 0.381%, respectively. PLS combined with UVE obtains the comparative results to full-spectrum PLS, the r and RMSE in prediction set are 0.917 and 0.392%, while the number of variables used in the calibration model is reduced from 351 to 134.

(Download PDF)    (Export to EndNotes)