Objective   A model based on the leaf chlorophyll was established to provide a rapid, convenient, and non-invasive means for evaluating and predicting the quality of varieties of JUNCAOs.

  Methods  Using a multispectral camera, spectra of leaf chlorophyll were obtained from different JUNCAOs. NDSI and RSI vegetation indices were established according to the RGB and HSV color space systems. A chlorophyll meter was used to obtain SPAD readings on JUNCAO leaves . Four regression algorithms were applied to build the model to correlate the spectral and SPAD measurements.

  Results   A significant difference on SPAD was observed between healthy and dry or dead leaves, but not between different JUNCAO. A significant correlation was found between the leaf SPAD and spectral reflection, especially, the red band of RGB color showing a coefficient of −0.79. The high sensitivity of the red band in distinguishing healthy from withering leaves allowed an accurate and stable measurements by the color systems that ranked as: 5-channel RGB＞HSV＞3-channel RGB. Among the 4 inversion methods, the rank was: random forest machine learning＞support vector machine learning＞stepwise linear regression＞univariate linear statistical. And the best fitting function in predicting SPAD was the RGB-NDSI-RF type, which had a R² of 0.95 vs. 3.04 for RMSE and 0.19 for MRE and validation R² of 0.75 vs. 8.26 for RMSE and 0.76 for MRE.

  Conclusion   As JUNCAO is becoming increasingly popular for wide applications, a reliable, rapid, non-invasive method to determine the plant quality was much desired. The current leaf chlorophyll spectra-based RGB-NDSI-RF model built by the random forest machine learning method demonstrated an applicability for evaluating varieties of JUNCAO.