基于高光谱成像和外观特征的祁门工夫红茶质量数字化评判

Digital Evaluation of the Quality of Keemun Congou Black Tea Based on Hyperspectral Imaging and Appearance Features

研究提出基于高光谱成像(Hyperspectral imaging,HSI)结合化学计量学算法对七个等级的祁门工夫红茶样品外观质量评判。利用HSI系统采集样品原始图像,提取其9个色泽特征。利用灰度共生矩阵和灰度统计矩阵等图像纹理特征提取算法获得66个纹理特征数据。将图像色泽和纹理数据进行特征融合,以融合数据作为构建模型的特征向量。利用支持向量机、随机森林和最小二乘支持向量机(Least squares support vector machine,LSSVM)算法结合特征融合数据,构建祁门红茶样品的等级识别模型。实验显示,基于融合特征的LSSVM模型的判别精度最高,为94.85%。结果表明,HSI结合色泽、纹理特征和化学计量学方法作为一种快速和有效的策略,可以成功地应用于祁门工夫红茶等级质量的准确判别。

Hyperspectral imaging(HSI)combined with chemometric algorithm was applied to evaluate the appearance quality of seven grades of Keemun Congou black tea samples.The original images of the samples were acquired using the HSI system for extracting their 9 color features.66 texture feature data were obtained using image texture feature extraction algorithms such as gray-level co-occurrence matrix and grey-level gradient co-occurrence matrix.The image color and texture data were fused,and then were utilized as the feature vector to build the model.The support vector machine,random forest and least squares support vector machine(LSSVM)algorithms were employed to combine the feature fusion data to build the grade recognition models of Keemun black tea samples.The experimental results showed that the LSSVM model based on fused features had the highest discriminative accuracy of 94.85%.It is demonstrated that HSI combined with color,texture characteristics and chemometric methods can be successfully applied as a rapid and effective strategy to accurately identify the grade quality of Keemun Congou black tea.