Optimization and comparison of models for prediction of soluble solids content in apple by online Vis/NIR transmission coupled with diameter correction method

The online system can achieve high efficiency towards fruit quality determination in postharvest period. Thus, developing online and nondestructive technology for inspecting and grading fruit is meaningful and profitable in the existing robotic sorting systems. In this study, the effect of fruit diameter differences on online prediction of soluble solids content (SSC) of 'Fuji' apples based on visible and near-infrared (Vis/NIR) spectroscopy was studied. Partial least square (PLS) regression was employed to establish calibration models based on three wavelength regions (675-1025, 710-980, 750-1025 nm) and two fruit orientations (stem-calyx axis vertical with stem upward (T1) and stem-calyx axis horizontal with stem towards light source (T2)), respectively. A novel diameter correction method was proposed to reduce the effect of fruit diameter differences on original spectra. Combined with pretreatment and effective wavelength (EWs) selection methods, models were optimized and compared to determine the best calibration strategy. Diffuse transmission spectra in 710-980 nm and diameter correction method with calculated attenuation coefficient were testified much better than other corresponding regions and correction methods, respectively. Baseline offset correction (BOC) and 7-point Savitzky-Golay smoothing (SGS) of pretreatments and competitive adaptive reweighted sampling (CARS) of EWs selection methods were proved to be outstanding among other methods. 59 and 63 EWs achieved the best detection accuracies with correlation coefficient of prediction (r(p)) and root mean square error of prediction (RMSEP) of 0.92 and 0.50 degrees Brix, 0.89 and 0.56 degrees Brix for T1 and T2, respectively. The overall results indicated that online Vis/NIR transmission spectra after BOC and 7-SGS with proposed diameter correction method can make the variation of fruit diameters a small interference for SSC determination, and CARS-PLS would be effective to simplify models and promote computing efficiency to make this nondestructive detection technique promisingly applied.