A new method in wavelength selection for texture prediction of Mozzarella by Hyperspectral imaging

• Low fat Mozzarella by process modification in a new method was produced • The effect of process condition on fat and moisture content was proved which imapced on some rheological characteristics • While genetic algorithm meet remarkable performance in wavelengths selection for free oil prediction, all of algorithms showed high performances in other characteristics except cohesiveness. • Selected wavelengths were connected to some special wavebands which were corresponded to proteins and fats in Mozzarella as the spectral signature of HSI In order to select the most important wavelengths to predict the rheological properties of low fat Mozzarella, five common feature selection methods were put in to a competition through developing a PLSR model. Eight rheological characteristics (hardness, adhesiveness, springiness, cohesiveness, gumminess, chewiness, meltability and free oil formation) were measured as true value beforehand of hyperspectral imaging. After analysing by IBM SPSS Statistics, it was cleared that, hardness, adhesiveness, cohesiveness and gumminess were affected by fat, while meltability was significantly correlated to the moisture content. Developing PLSR, it was found that, except β-coefficient method which selected 32 wavelengths with a not high accuracy (R 2 p=0.661, RMSEp=0.07) for cohesiveness, other methods successfully were able to select the most important wavelengths with acceptable accuracy while represented the optimum condition for prediction of rheological properties. In summary, as if the fat played a significant role in hardness, adhesiveness, cohesiveness as such gumminess, as well as moisture in meltability, the special wavebands identified in connection with functional groups N-H and C-H corresponding to the proteins and fats in Mozzarella structure, demonstrated the spectral signature of hyperspectral imaging as well.