基于近红外光谱的神经网络预测大米直链淀粉含量

摘要

借助主成分分析，确立了用于近红外光谱分析的BP神经网络的输入输出模式对;并用BP神经网络方法建立了不同类型、不同粒度的大米样品直链淀粉含量预测模型;考察了模型的预测能力，其预测值与用标准方法取得的化学测定值间具有良好线性关系(相关系数达0.9)。用BP神经网络可降低因样品粒度的不同而对预测结果造成的差异。%The apparent amylose content (AAC) is one of the important parameters to affect the cooking and taste characteristics. The chemical measurement of rice AAC is of high expense and time consuming. So the algorithm of BP neural networks (BPNN) combined with principal component regression(PCR) using for dealing with the near infrared spectra of rice samples was presented. The principal components of near infrared spectrum data were calculated by PCR algorithm. These principal components were used as input nods of BPNN， and the prediction model of AAC in rice was built by optimizing the studying vector of BP neural networks. The correlation coefficient between the evaluation of rice AAC computed by BPNN algorithm and chemical value was 0.9. The standard error of prediction was 0.56 for milled rice， the mean relation error and the mean absolute bias were 2.0% and 1.5 respectively.