Near‐infrared reflectance spectroscopy (NIRS) was used to develop calibration curves for determining the fat acidity of whole‐kernel and ground rough rice with 13% moisture content at 25°C. Partial‐leastsquares regression (PLSR) uses the optimal calibration curve for wholekernel rough rice to measure the coefficient of determination (r2) of validation and standard error of prediction (SEP) of 0.87 and 0.83 mg of KOH/100 g of dry matter, respectively. However, the optimal calibration curve for ground rough rice has a higher r2 of validation and lower SEP of 0.94 and 0.73 mg of KOH/100 g of dry matter, respectively. From 10 to 40°C, the temperature effect causes an increase of 0.24 mg of KOH/100 g of dry matter/°C in the predicted fat acidity of whole‐kernel rough rice.

Cereal Chem. 82(6):649-653From five paddy rice cultivars grown in Taiwan and harvested in the summers of 1997, 1998, and 1999, five calibrated models were established by backpropagation neural network program through different morphological and color features selection for classifying paddy rice harvested in the summer of 2000. With 60 features, the average classification rates of Model 1 and Model 5 were 92 and 99.8%, respectively. With the most effective 50 features, by loading in the first principal component, the average classification rate of Model 2 was 90.0%. With 35 features selected from the correlation coefficient matrix, the average classification rate of Model 3 was 91.0%. With the most effective 20 features of area, area/ perimeter, 48th width, shape factor, maximum length/maximum width, average intensity of blue, maximum length, average intensity of green, 47th width, 50th width, average intensity of red ,1st width, 19th width, 5th width, 6th width, 29th width, perimeter, 46th width, 42nd width, and 4th width based on the contribution of the training model, the average classification rate of Model 4 was 91.8% and would be recommended for classifying five paddy rice cultivars of set trading prices because it required fewer features and held a stable classification rate.

Using five paddy rice cultivars grown in Central, Eastern, and Southern Taiwan and harvested in the summers of 1997, 1998, and 1999, eight calibrated models were established by discriminant analysis and back‐propagation neural network with four wavelength selection methods. Randomly adding 80 samples of the 2000 year crop in the three‐crop‐year calibrated models for annual recalibration, eight models were used to classify paddy rice harvested in the summer of 2000. With 351 wavelengths of models 1 and 2, the average classification rates by discriminant analysis and backpropagation neural network were 98.1 and 92.5%, respectively. With 69 wavelengths selected by stepwise discrimination of models 3 and 4, the average classification rates by discriminant analysis and backpropagation neural network were 98.5 and 85.5%, respectively. With 69 wavelengths selected by correlation matrix of models 5 and 6, the average classification rates by discriminant analysis and neural network were 72.0 and 72.2%, respectively. With 69 wavelengths from loading values in the first and second principal components of models 7 and 8, the average classification rates by discriminant analysis and neural network were 69.1 and 60.6%, respectively. Model 3 would be recommended for classifying paddy rice to set trading prices because of its highest classification rate (98.5%).