In the present article, near infrared spectra of 89 wood samples of different geographical provenances and species were measured, and back propagation artificial neural networks(BPANN) and generalized regression neural network(GRNN) were used for modeling of wood species NIRS identifying. Parameters for two neural networks were chosen via analysis of variance, respectively; and networks were trained with optimum parameters. Considering the difference between spectra, spectra with different levels of white noise and different levels of bias were simulated and predicted by using the models built. It was found that both the two models had satisfactory prediction results, identification correct rates obtained by BPANN model applied to spectra with bias level no higher than 2% and noise level no higher than 4% were above 97%; correct rates obtained by GRNN model applied to spectra with bias level no higher than 2% and noise level no higher than 4% were above 99%.%测量了不同产地及品种的89个木材样品的近红外光谱,并分别使用反向传播人工神经网络(back propagation artificial neural networks,BPANN)与广义回归神经网络(generalized regression neural network,GRNN)建立了NIRS树种识别模型.通过方差分析分别选择两种神经网络所用参数,并采用最优参数进行网络训练.考虑到样品光谱的差异,对含不同水平白噪声与不同水平偏置的光谱进行模拟,并使用建立的模型对模拟光谱进行预测.发现两种神经网络模型均有较好的预测结果,其中BPANN模型,对含偏置水平不高于2％、噪声水平不高于4％的模拟光谱识别正确率在97％以上；GRNN模型,对含偏置水平不高于2％、噪声水平不高于4％的模拟光谱识别正确率在99％以上.
展开▼
