OPTIMAL GEOMETRIES FOR THE DEVELOPMENT OF RICE QUALITY SPECTROSCOPIC CHEMOMETRIC MODELS

Three sample geometries, two different instrument types, and two spect ral collection modes (reflectance and transmission) were used to asses s rice quality and develop chemometric models for composition and sens ory characteristics. Rice samples (120) including three cultivars, two growing locations, five drying treatments, two moisture levels, and t wo levels of milling were scanned in two locations. Data collected for modeling included amylose, protein, moisture, whiteness, transparency and milling degree. Taste and texture were determined with the use of separate trained sensory panels. The NIR models show that composition is best modeled in the 1,100-2,500 nm range, while the physical prope rties of whiteness, transparency and milling degree are best modeled i n the 750-1,1350 nm range. Additional models were developed using limi ted data subsets of the spectral data points. In some cases, adequate models were generated with as few as 20 wavelength data points. Result s show that no one spectroscopic protocol is best for all analytes in rice and that for any complex food matrix more than one preprocessing or spectral range protocol is needed.