基于红外光谱的烟叶自动分级研究

烟叶的自动分级一直是国内外学者智能化研究的一个重要方向。通过分析烟叶的主要组成成分和烟叶光谱信息特征，发现烟叶红外光谱可作为烟叶分级特征，并通过神经网络模型验证了红外光谱作为其分级特征的可行性，通过对比分析选取最佳红外光谱间隔、光谱范围以及最必要的光谱预处理方法。利用概率神经网络对9个等级的烟叶进行分组分级，首先对选光谱数据进行减均值的预处理以消除基线漂移，然后将其作为神经网络的输入模式，相应的等级或组分作为理想输出训练网络。选择近半数的样本作为训练样本，其余为测试样本；网络对于训练样本的正确吻合率为100%，测试样本的平均正确吻合率91%以上。结果表明烟叶的红外光谱可以作为烟叶的分级特征，概率神经网络可以用于烟叶自动分级，为烟叶的自动分级提供了新方法。

Study on automatic classification of tobacco based on infrared spectrum

The automatic classification of tobacco leaves has been an important intelligent research field.By studying on the components of tobacco leaves and analyzing the spectrum of tobacco leaves,the infrared spectra of tobacco leaves can be used as stamp of the tobacco grading.This is verified by the neural network model.The best intervals of the infrared spectra and the necessary preprocessing algorithm are studied and decided.400 infrared spectra of tobacco leaves are adopted to group or grade leaves by neural network model.Almost half leaves are used as training samples and others as test samples.The mean rates of correct fit classification are 100% and 91% for training and test samples respectively.The grouping and grading results of neural network model based on infrared spectra show that the infrared spectra can be applied to grade tobacco leaves automatically.This indicates it is a new method for the automatic grading tobacco leaves.