| Abstract | Rice ( Oryza stiva L.) trade is important in economic terms as it is a staple food in many countries. Not only the varietal diversity, environmental differences and agricultural handling di sparity, but also the individual customer’s preference commonly leads to variation in the conception of rough rice quality. Moreover, the rough rice quality results in milling quality. Hence, inspection of rough rice quality is extremely important . General ly, the quality is assessed from the core primary constituents, namely moisture and varietal purity, not to mention the assessment of deterioration, foreign matter contamination, maturity and milling quality. In some cases, secondary and tertiary constitue nts such as proteins and amylose in brown rice and milled rice are measured because of nutritional and physico - chemical property effects. This research investigated the feasibility of alternative near infrared spectroscopy (NIRS) for evaluation of rough ri ce quality based on moisture, protein and amylose content simultaneously . V arietal authentication and adulteration was determined with whole - grains and single - kernel measurements. Moist rough rice samples used in this study were derived from five Thai fa mous varieties , namely, Khao Dawk Mali 105 (KDML105), Pathum Thani 1 (PTT1), Suphan Buri 60 (SPR60), Chainat1 (CNT1) and Pitsanulok2 (PSL2). The samples were randomly collected from 259 different locations all over Thailand during December 2009 and March 2 010. The whole - grains and single - kernel spectra of rough rice samples were tested using near infrared reflectance spectroscopy (NIRS_R), Fourier transform near infrared reflectance spectroscopy (FT - NIRS_R), and near infrared transmittance spectroscopy (NIR S_T) and used for developing prediction models. Model testing was accomplished by independent spectra to verify the performance. Based on constitutional prediction, the NIRS_R and FT - NIRS_R models for the measurement of whole - grains by the partial least s quares regression (PLSR) method were successfully developed with high coefficient of determination of calibration (R 2 c ) and low standard error of prediction (SEP). The FT - NIRS_R model showed better performance than NIRS_R model to predict rough rice moist ure content ( R 2 c = 0.990, SEP = 0.30 % ), brown rice crude proteins ( R 2 c = 0.921, SEP = 0.23 % ), milled rice crude proteins ( R 2 c = 0.863, SEP = 0.23 % ) and apparent amylose content ( R 2 c = 0.966, SEP = 1.37 % ). Likewise, indirect instrumental measurement of roug h rice moisture content based on single - kernel technique by NIRS_R and NIRS_T was reasonable. The R 2 c of 0.986 and SEP of 0.27 % were obtained using the PLSR model of NIRS_R, while the model derived from NIRS_T spectra showed the R 2 c of 0.979 and SEP of 0.5 3%. The analysis of varietal authentication was possible using with NIRS also both whole - grains and single - kernel s . Pure whole - grains of rough rice samples containing 13.07 to 27.78% moisture content could be differentiated with 71.32% of accuracy based o n NIRS_R spectra omitting water - bands in intermittent wavelength ranges of 1100 - 1402, 1690 - 1815, 2055 - 2191, 2248 - 2500 nm and using the soft independent modeling of class analogies classification (SIMCA) method. More comparative model with 96.60% of accurac y was developed from FT - NIRS_R spectra in wavelength range of 1600 - 2500 nm. The partial least square discriminant analysis (PLSDA) method resulted in superior models and had 100 and 99.22% of accuracy for NIRS_R and FT - NIRS_R application (1100 - 2500 nm), re spectively. Based on individual kernel measurement, the PLSDA method was applied to entire rough rice spectra of NIRS_R (1100 - 2500 nm) and NIRS_T (665 - 955 iv nm). The obtained models had a potential to identify their varieties with high accuracy of 91 and 100 %, respectively. The investigated moisture in rough rice kernels for calibration modeling ranged from 13.61 to 28.16% for NIRS_R and from 5.13 to 28.60% for NIRS_T analysis. Based on the analysis of off - variety adulteration in bulk KDML105 rough rice sam ples with single - kernel technique, the results strongly confirmed the possibility of using NIRS_T to predict the contamination (% w/w) of PTT1, SPR60, CNT1 and PSL2 with SEP of 2.47 - 3.31 % . The models were developed by using PLSR method. The performance of predictive model for determining PTT1 - adulteration ( R 2 c = 0.997) was almost the same with that for adulterated CNT1 ( R 2 c = 0.994), SPR60 ( R 2 c = 0.994) and PSL2 ( R 2 c = 0.996) determination . The accomplishment of model development could be collaborated with the differences of spectra depending on cellulose, starch, protein and oil content among those varieties. In conclusion, the NIRS provides a state - of - art technology that will allow the successful inspection of rough rice quality by determining moisture co ntent, screening secondary and tertiary constitutions, identifying varieties and detecting off - variety adulteration based on either whole - grains or single - kernel approach . The test can also be done more rapidly than the traditional methods. Approximately 3 min per a sample is required for NIRS_R and FT - NIRS_R testing, while the NIRS_T testing required only 10 s. It is also worthwhile mentioning that these approaches can be accomplished whenever the undried or dried materials are supplied, because of the ex pedient calibration models. Moreover, the single kernel approach is capable of automate on - line inspection of the quality at drying, milling and breeding facilities and can make more an efficient quality control system s in the future. |