摘要
提出了一种用近红外光谱技术快速无损鉴别杨梅品种的新方法,首先用主成分分析法对典型的四个杨梅品种进行聚类分析,获取杨梅的近红外指纹图谱,再结合人工神经网络技术进行品种鉴别.主成分分析表明,以主成分1和2对样本的得分值做出的得分图,对不同种类杨梅具有较好的聚类作用,可以定性分析杨梅种类.利用主成分分析可以把原始波长变量压缩成能代表原始变量的少数相互正交的主成分,用这些新变量作为神经网络的输入,建立3层BP人工神经网络模型.四个杨梅品种共100个样本用来建立神经网络品种鉴别模型,对未知的20个样本进行预测,结果表明,品种识别准确率达到95%.说明综合主成分分析和人工神经网络的方法具有很好的分类和鉴别作用,为杨梅的品种鉴别提供了一种新方法.
A new nondestructive method for discriminating varieties of waxberry by visible and near infrared spectroscopy （Vis/NIRS） was developed. First, the spectral data were analyzed by principal component analysis （PCA） for varieties clustering. Then diagnostic information was obtained from original spectra, these informations were used for pattern recognition based on ANN model. The score plot provided the reasonable clustering of the varieties of waxberry. Small quantities of principal components from PCA were used as inputs of a back propagation neural network （BPNN） with one hidden layer. 100 samples were selected randomly from four varieties, then they were used to build BPNN model. This model had been used to predict the varieties of 20 unknown samples. The recognition rate of 95% was achieved. This model is reliable and practicable. So this method could offer a new approach to the fast discriminating varieties of waxberry.
出处
《红外与毫米波学报》
SCIE
EI
CAS
CSCD
北大核心
2006年第3期192-194,212,共4页
Journal of Infrared and Millimeter Waves
基金
国家自然科学基金项目(30270773)
教育部高等学校优秀青年教师教学科研奖励计划(02411)
关键词
近红外光谱
杨梅
主成分分析
人工神经网络
聚类
near infrared spectra
waxberry
principal component analysis
artificial neural network
clustering
