近红外光谱协同BP神经网络的泰国茉莉香米掺伪含量快速鉴定

利用近红外光谱协同BP神经网络算法，对泰国茉莉香米及其掺伪样品的近红外光谱进行多元散射校正预处理，挑选出48 个特征波长；以特征波长的吸光度为BP神经网络输入层神经元，以样品中泰国茉莉香米的含量为输出层神经元，获得BP神经网络算法的最优结构模型，即为单层隐含层、隐含层神经元数7、隐含层传递函数logsig、输出层传递函数tansig、训练函数trainlm、网络学习函数learngdm和学习速率0.35。所建立模型的均方根误差、校正集相关系数、验证集相关系数、测试集相关系数分别为0.000 830、0.992 9、0.976 1和0.975 5，呈现出优良的预测效果，实现了泰国茉莉香米掺伪含量的快速鉴定。

Rapid Quantitative Determination of Adulterated Thai Jasmine Rice Using Combined Near-Infrared Spectroscopy and Backward Propagation Neural Network

Here, a rapid method for identifying adulterated Thai jasmine rice was developed using near-infrared spectroscopy combined with backward propagation (BP) neural network. Near-infrared spectra of pure and adulterated rice samples were pretreated by multiplicative scatter correction (MSC) and 48 characteristic wavelengths were selected by competitive adaptive reweighted sampling (CARS). Then, the optimal structure model for BP neural network algorithm was established using absorbance values at these wavelengths as the input layer neurons and Thai jasmine rice contents in samples as the output layer neurons, involving a single hidden layer, seven hidden layer neurons, logsig as the transfer function of hidden layer, tansig as the transfer function of output layer, trainlm as the training function, learngdm as the learning function of the network, and learning rate of 0.35. The model showed an excellent prediction performance with a root mean square error (RMSE) of 0.000 830, correlation coefficient of the calibration set of 0.992 9, correlation coefficient of the verification set of 0.976 1, and correlation coefficient of the test set of 0.975 5.