Abstract:

In order to promote the breeding of Perilla frutescens varieties with high anthocyanin and guide the production management of Perilla frutescens under the stress of adversity， Perilla frutescens was as the research object， the field leaves were collected and taken photos with a digital camera. The image color of photo was analyzed by red green blue color space （RGB） and CIELab color space. And the relationship between color parameters and leaf anthocyanin content was analyzed for screening out color parameters with high correlation coefficient. Univariate regression inversion model was established， and finally the best predictive model of anthocyanin content in the leaves of Perilla frutescens was obtained. The results showed that in RGB color space， the normalized redness intensity （NRI） and normalized greenness intensity （NGI） had significant correlations with anthocyanin content， and the correlation coefficient of NGI was greater than that of NRI. When the contribution ratio of leaves front and leaves back was 2∶1， the correlation between NGI and anthocyanin content was the highest with the correlation coefficient 0.853 2. Compared with different models，it was found that the exponential model established with NGI as the independent variable had the best fitting effect with the correlation coefficient 0.838 1 and the coefficient of determination （R²） 0.755 0. In the CIELab color space， a^* had the best correlation with anthocyanin content， and the correlation coefficient reached the maximum （0.735 6） when the contribution ratio of leaf front and leaf back was 2∶1. The power model based on a^* had the best fitting effect，and the correlation coefficient and R² were 0.743 8 and 0.679 8， respectively. The NGI model and a^* model were respectively used to estimate the content of anthocyanin in leaves. After verification， it was found that the prediction effect of the a^* model was better with higher accuracy and stability. Therefore， the model of a^* was used as the best model to predict the content of anthocyanins in Perilla frutescens leaves.

Key words: Perilla frutescens, RGB, CIELab, anthocyanin, digital camera

摘要：

为推进富含花青素的紫苏品种选育，指导逆境胁迫下的紫苏生产管理，以紫苏为研究对象，采集田间叶片并使用数码相机拍照，结合红绿蓝色彩空间（red green blue color space，RGB）和CIELab色彩空间（CIELab color space）2种图像色彩分析手段处理图片，与叶片花青素含量进行相关性和显著性分析，筛选出相关系数较高的色彩参数，建立单变量回归反演模型，最终综合建模得到预测效果最优的紫苏叶片花青素含量预测模型。结果表明，在RGB色彩空间中，红光标准化值（normalized redness intensity，NRI）、绿光标准化值（normalized greenness intensity，NGI）与花青素含量呈极显著相关，其中NGI的相关系数大于NRI。当叶片正反面色彩贡献比为2∶1时，NGI与花青素含量的相关性最大，相关系数为0.853 2。对比不同模型发现，以NGI为自变量建立的指数模型拟合效果最好，相关系数为0.838 1，决定系数（R²）达0.755 0。在CIELab色彩空间中，红度（a^*）与花青素含量的相关性最好，且相关系数同样在叶片正反面色彩贡献比为2∶1时达最大，为0.735 6。基于a^*建立的幂模型拟合效果最好，相关系数和R²分别为0.743 8和0.679 8。分别使用NGI模型和a^*模型对叶片花青素含量进行估测，验证后发现a^*模型的预测效果更好，准确性和稳定性更高，因此以a^*模型为预测紫苏叶片花青素含量的最优模型。

关键词: 紫苏, RGB, CIELab, 花青素, 数码相机