*3.5. SSC and Soluble Sugars Contents Predictions*

Table 4 generalized the statistics of PLS model results of calibration and prediction using μ*<sup>a</sup>* and μ- *<sup>s</sup>* at 905–1600 nm for SSC and the contents of soluble sugars of Fuji apple. The *Rc* <sup>2</sup> of the modeling set was higher than the *Rp* <sup>2</sup> of the prediction set, thus these models were reliable. Among the three prediction models of soluble sugars based on μ*<sup>a</sup>* and μ- *<sup>s</sup>*, the best model was for sucrose content (*Rp* <sup>2</sup> > 0.7365, *RMSEP* < 1.458 g/kg). In addition, the models relied on μ*a*, and the μ- *s* for fructose and glucose content were slightly worse, the *Rp* <sup>2</sup> values of which were in the range of 0.635–0.781 and 0.593–0.725, and the *RMSEP* values were 1.570–2.216 g/kg and 1.344–1.906 g/kg. A previous study reported similar results in that the prediction results on optical properties for sucrose were better than fructose and glucose [36].


**Table 4.** Model results of soluble solid contents and soluble sugar contents for apples.

The root mean square error of calibration (*RMSEC*); the coefficient of determination of prediction (*Rp* 2); the root mean square error of prediction (RMSEP).

In addition, the *Rp* <sup>2</sup> values of the five prediction models based on μ*<sup>a</sup>* were in the range of 0.752–0.851, and the *RMSEP* values of the SSC and soluble sugar models were 0.329% and 1.047–2.062 g/kg, respectively. Li et al. [47] assessed the SSC of apple based on near infrared spectroscopy, with the results of *Rp* <sup>2</sup> and RMSEP in the ranges of 0.576–0.854 and 0.547–0.673%, and the prediction accuracy of SSC was equivalent to the present study. In addition, according to the model results (Table 4), the prediction results based on μ*<sup>a</sup>* were better than the model results based on μ- *<sup>s</sup>*, which further proved that SSC and soluble sugars were more correlated with μ*<sup>a</sup>* than μ- *s*. The findings obtained from this study made a contribution to the fact the light absorption links to the chemical compositions of materials, and that scattering of photons is a physical procedure that greatly depends on the wave characteristics of photons and the structures of mediums.

In general, the prediction models for sucrose based on μ*<sup>a</sup>* in this study performed well (*Rp* <sup>2</sup> = 0.851, *RMSEP* = 1.047). The models for sucrose in 400–1050 nm built with the AIS system, according to the report of Wei et al. [36], were second to the models of our study. According to their study, the models' *Rp* <sup>2</sup> and *RMSEP* values were 0.804 and 1.099, respectively. The result might be caused by the presence of the more obvious bonds C-H, O-H, C-C with a longer wavelength, which indicated that the absorption properties of the mid-wave near infrared region are more suitable for the quantitative detection of soluble solids and soluble sugars.
