The appraisal of aroma types of tobacco usually depends on olfaction, the veracity of its result is sometimes hard to be guaranteed. In view of this, sensory evaluation models have been constructed at home and abroad by using BP neural network or other methods, but they are inefficient in recognition. According to the relationship between chemical composition and the aroma types of tobacco, the recognition model of tobacco aroma types has been constructed by using Tabu search-based Bayesian network. Experimental results showed that it can attain a better Bayesian network structure, and has higher training efficiency and better accuracy in classification compared with BP neural network or other methods.%烟叶香型通常是靠人的嗅觉评定的,评定结果的准确性往往难以保证.针对该问题,国内外建立了BP神经网络等感官评估模型,但识别效率不高.根据烟叶中化学成分与烟叶香型关系,使用基于Tabu搜索的贝叶斯网络建立烟叶香型识别模型.实验结果表明,使用该方法能得到较好的贝叶斯网络结构,与BP神经网络等方法相比训练效率更高,分类的结果也更加准确.
展开▼