茶叶外形品质的高光谱图像量化分析  

The Quantification of Tea Appearance Quality Using Hyper-spectral Imaging

作　　者：吴瑞梅[1] 吴彦红[1] 艾施荣[2] 刘木华[1] 赵杰文[3] 严霖元[1] 

机构地区：[1]江西农业大学工学院,江西南昌330045 [2]江西农业大学软件学院,江西南昌330045 [3]江苏大学食品与生物工程学院,江苏镇江212013

出　　处：《江西农业大学学报》2013年第2期413-418,共6页Acta Agriculturae Universitatis Jiangxiensis

基　　金：江西省科技计划项目(20112BBF60019);江西省教育厅科学基金项目(GJJ11081)

摘　　要：为弥补感官审评方法评定茶叶品质存在的不足,提出采用高光谱图像来量化分析茶叶的外形感官品质。以碧螺春名优绿茶为对象,采集其高光谱图像,利用主成分分析法从原始高光谱图像中优选出3个波长(768.74,827.54,886.83 nm)下的特征图像。分别提取每个特征图像的颜色特征和纹理特征,3个特征图像共提取90个特征变量。利用BP神经网络方法建立特征变量与外形感官评分之间的相关模型,模型对预测集样本的相关系数为0.859,预测均方根误差为3.611。在预测集中,对模型的预测值与实际评分值进行t检验时,预测值与实际值无显著差异,说明所建模型是准确可靠的。Hyper-spectral imaging technique was used for the quantitative evaluation of tea quality so as to make up the deficiency of sensory evaluation method for evaluating green tea quality.＂Biluochun＂,a famous green tea was used as the test object.Hyper-spectral image data of tea were obtained by using a hyper-spectral imaging system based on spectrometer.Three feature images at 768.74 nm,827.54 nm and 886.83 nm were optimized by principal component analysis method from raw hyper-spectral images.Color feature and texture feature were extracted from each feature image,and 90 feature variables in all were obtained from the three feature images.The model between the feature variables and sensory scores of tea appearance was developed by BP-ANN method.The correlation coefficient（R） and root mean square error of prediction（RMSEP） of BP-ANN model were 0.859 and 3.611 in the prediction set,respectively.25 samples in prediction set were used to do T-test,and the result indicated that there was on significant difference between the actual sensory scores and the prediction values of the model.The result showed that the developed model was accurate and reliable.

关 键 词：茶叶外形 高光谱图像 BP神经网络 主成分分析 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]