基于时间序列神经网络的山核桃化学成分分析

利用果实化学成分含量与光谱的非线性模型,实现果实化学成分含量的快速无损鉴定,成为林业研究的热点之一。果实在生长发育过程中,化学成分的含量随时间的递增而不断增加,常规方法不能较好拟合和预测果实化学成分含量的变化。本研究提出一种基于时间梯度的神经网络方法(TSNN),以6个时间梯度山核桃果实蛋白质和脂肪含量的光谱和实测数据为研究对象,分别与偏最小二乘法(PLS)和人工神经网络(PLS-ANN)方法比较,检验TSNN方法的建模和预测效果。结果表明:TSNN方法对蛋白质含量的预测,均方根误差分别比PLS和PLS-ANN方法降低了18.82%和7.39%;TSNN方法对脂肪含量的预测,均方根误差分别比PLS和PLS-ANN方法降低了39.95%和35.02%。TSNN方法的校正相关系数平方(R_c~2)和预测相关系数平方(R_p~2)比PLS和PLS-ANN均有提升。因此,TSNN方法是一种比较准确实用的定量分析方法。

Neural Network of Time Series in Chemical Content of Hickory

We proposed a neural network of time series method (TSNN),and compared with the partial least squares (PLS) and the artificial neural networks of partial least squares (PLS-ANN) to test the results of modeling and prediction.The measured data of spectra and proteins and fat contents in six growth stages were studied.The PLS,PLS-ANN and TSNN method were used to establish the model,and the results were compared.For the contents of protein,the root mean square error (RMSEP) of the TSNN was reduced by 18.82％ and 7.39％ PLS and PLS-ANN,respectively.For the contents of fat,the RMSEP of the TSNN were reduced by 39.95％ and 35.02％ compared with PLS and PLS-ANN,respectively.The correlation coefficient squared and the prediction correlation coefficient squared of TSNN were improved compared with PLS and PLS-ANN.