基于图像和光谱信息融合的红茶萎凋程度量化判别

为了实现对红茶萎凋程度量化判别,该研究提出了一种将图像和光谱信息融合后分别与线性判别分析法和偏最小二乘法结合的技术,进行工夫红茶萎凋程度定性判别及儿茶素与氨基酸比值定量预测研究。通过对图像进行主成分分析,筛选出5个特征波长和对应的光谱特征值,基于灰度共生矩阵提取5个特征波长图像的纹理特征值,并采用连续投影算法优选出14个纹理特征值,然后分别以光谱和纹理特征值融合数据建立红茶萎凋程度的线性判别模型和儿茶素与氨基酸比值的偏最小二乘预测模型。结果表明:采用所研究的方法和建立的模型对工夫红茶萎凋程度判别准确率达到94.64%,儿茶素与氨基酸比值预测相关系数为0.8765,预测均方根误差为0.434,预测结果较好。证明应用这两种方法能实现对红茶萎凋程度量化判别。

Discriminant of withering quality of Keemun black tea based on information fusion of image and spectrum

Abstract: Withering is the first procedure and the key step in processing of black tea. It is crucial for the quality of black tea product. Usually, the judgment of the withering degree relies on the processor''s judgment, rather than a quantitative analysis by fast evaluation method. In order to develop the digitized discrimination on withering degrees, different degrees of withering samples were collected in our research. In this study, 168 samples provided by Jindong tea factory in Qimen County were investigated. All of the samples belonged to different withering degrees (55 samples of mild withering, 61 samples of moderate withering and 52 samples of excessive withering). The samples were randomly divided into two subsets at the ratio of 2:1. 112 samples were chosen as the calibration set and the remaining 56 samples were prediction set. The calibration set was used to develop the model, while the prediction set was applied to test the robustness of the model. The withering degree was nondestructively evaluated by hyperspectral imaging technology at the range of 908-1735 nm. It was suggested that the ratio of catechins/amino acids was correspondingly decreased with the development of withering degrees. Furthermore, the contents of catechins and amino acids of these samples were detected by high-performance liquid chromatography (HPLC). The characteristic spectra were extracted from the region of interest (ROI), and standard normal variate (SNV) method was preprocessed to reduce background noise. All of the hyperspectral images of tea samples with different withering degrees were analyzed by principal component analysis (PCA). The first two principal component (PC) images were selected because PC1 and PC2 contributed to 99.59% variance of the total. Therefore, the first two PC images were used for selecting dominate wavelengths. And five dominant wavelengths (1 040, 1 182, 1 249, 1 449 and 1 655 nm) were selected as spectral features. Textual features were collected by Grey level co-occurrence matrix (GLCM) from five dominant wavelengths of images. Fourteen dominant textual features were selected by successive projections algorithm (SPA). Subsequently, linear discriminant analysis (LDA), support vector machine (SVM) and extreme learning machine (ELM) classification models were developed based on spectral features, textural features and data fusion, respectively. Compared with the results of the models built with spectral features or textural features, the LDA, SVM and ELM models based on data fusion showed higher correct discrimination rate in prediction set. The correct discrimination rate of LDA, SVM and ELM based on data fusion were 94.64%, 91.07% and 92.86%, respectively. The results indicated that hyperspectral imaging combined with LDA was a potent tool in the discrimination of withering degrees. At the same time, catechins/amino acids ratio was also applied in the discrimination of withering degrees. The study showed that correlate coefficient of prediction set by catechins/amino acids ratio was 0.8765, and root mean square error of prediction was 0.434. The results in this study provide a new method with fast and scientific of digitized discrimination for withering degree during black tea processing.