Effect of measurement position on prediction of apple soluble solids content (SSC) by an on-line near-infrared (NIR) system

Near infrared spectroscopy has been widely applied in the area of rapid determination for fruits’ internal qualities. Therefore, an on-line near-infrared detection system was established to predict the SSC of apples in this study. Due to random measurement positions of apples, negative influences will be exerted on the performance of prediction models. With the aim of learning more about these influences and also compensating for them, spectra were taken at different measurement positions in the present work, including six fixed positions and a random position. Besides, the relations between these positions were investigated as well. It is also found that when the concave surfaces at apple’s calyx and stem interfered with the light path in the detection system, model’s robustness and accuracy would be deteriorated. At last, average and global spectra were used to build prediction models with comparison purpose. The optimal prediction model was established by the average spectra of seven measurement positions (RMSEC 0.356%; rc 0.947; RMSEP 0.370%; rp 0.906), which was superior to results in many previous studies. Last but not least, several suggestions on compensation for these influences were made to improve the model performance in some practical applications.