Effect of selected pre-processing methods by PLSR to predict low-fat mozzarella texture measured by hyperspectral imaging

Having improved prediction performance of PLSR, the most common methods were implemented to reduce diffusively and specular reflected radiation caused by NIR hyperspectral data of low fat mozzarella cheese. As the Hyperspectral data suffer from complicated extra scattering radiation issued from the biological origin or instrumental measurements, SG_Smoothing, MSC, SG_FD, SNV and SNV_Detrend were applied to dimminish the noises. Then developing PLSR algoeithm six aboved pre-processing methods were put in competiotion for prediction of rheological properties (hardness, adhessiveness, cohesiveness, springiness, gumminess, chewiness, free-oil and meltability) of low fat mozzarella. Results showed that, based on PLSR, the SG_Smoothing was prefered for Hardness (R2p = 0.846, RMSEp = 2911.29), springiness (R2p = 0.85, RMSEp = 0.0939) and meltability (R2p = 0.728, RMSEp = 22.08). SG_FD was selected as prefered method for prediction of adhesiveness (R2p = 0.809, RMSEp = 56.39) and Free-oil (R2p = 0.992, RMSEp = 0.442) with the highest performance while SNV for gumminess (R2p = 0.835, RMSEp = 1446.52) and MSC for chewiness (R2p = 0.817, RMSEp = 1186.2) were chosenIn spite of less accuracy, SNV_Detrend was the best option for prediction of chohesiveness (R2p = 0.654, RMSEp = 0.071). Results implied that, except the Cohesiveness, selected pre-processing method had the best performance in prediction of low fat mozzarella texture according to PLSR model outputs.