CAJ | 학술논문

应用近红外漫反射光谱技术结合误差反向传递人工神经网络（BP—ANN）算法，建立豆粕品质（包括水分、粗蛋白、残油）的定量分析模型。将豆粕漫反射吸收光谱数据进行SNV、DT、SG求导、SG平滑和均值中心化处理，然后采用偏最小二乘方法（PLS）降维获取主成分，并优化选择合适的隐含层节点数、隐含层和输出层转化函数，建立校正模型，并用验证样品对校正模型进行验证。结果显示，BP—ANN法建立的水分、粗蛋白和残油的预测相关系数（R）分别为0．981、0．988、0．982，预测标准偏差（SEP）分别为0．120、0．216、0．036，均优于PLS建模方法结果，且满足传统分析方法的重复性要求，表明BP—ANN方法可用于生产过程豆粕品质的快速监控。
응용근홍외만반사광보기술결합오차반향전체인공신경망락（BP—ANN）산법，건립두박품질（포괄수분、조단백、잔유）적정량분석모형。장두박만반사흡수광보수거진행SNV、DT、SG구도、SG평활화균치중심화처리，연후채용편최소이승방법（PLS）강유획취주성분，병우화선택합괄적은함층절점수、은함층화수출층전화함수，건립교정모형，병용험증양품대교정모형진행험증。결과현시，BP—ANN법건립적수분、조단백화잔유적예측상관계수（R）분별위0．981、0．988、0．982，예측표준편차（SEP）분별위0．120、0．216、0．036，균우우PLS건모방법결과，차만족전통분석방법적중복성요구，표명BP—ANN방법가용우생산과정두박품질적쾌속감공。
The models of quantitative analysis of mositure, protein and residual oil in soybean meal were established by back propagation- artifical neural network method （BP- ANN） combined with near infrared diffuse reflectance spectroscopy. Firstly, the original absorbance spectra of soybean meal samples were pretreated by SNV, DT, Savitzky - Golay derivative , Savitzky - Golay smoothing and mean - centering. Secondly, the principal components were obtained by PLS dimension - reducing, and the number of hidden node, transfer functions of hidden layer and output layer were optimized ; Finally, all the parameers were inputed into BP- ANN to easablish the calibration model. Then the models were validated by prediction set. The results showed that the correlation coefficients （R） of prediction for moisture, crude protein and residual oil were 0. 981,0. 988 and 0. 982 respectively; and the standard errors of prediction （SEP） were 0. 120,0. 216,0. 036, respectively. It shows that BP - ANN was more accurate compared with the partial least square method （PLS）. Furthermore, the results meets the repeatability of traditional analysis method, it can be applied to rapid monitoring of soybean meal quality.