Objectives Cephalotaxus mannii is a state key protected plant. The construction of non-destructive growth monitoring method would provide an easy and instant tool for the warning of restricted growth conditions.
Methods A pot experiment with L9 (34) design was conducted in Yunlong Town, Haikou City from 2021 to 2022 , using 1-year-old C. mannii as test materials. The three factors were irrigation rate (W), light transmission (S), and nitrogen rate (N). The factors were designed three large gradients to make Cephalotaxus mannii seedlings grow differently, the three levels were 2500, 5000, and 7500 mL for W, and 30%, 10%, and 5% for S, and 0, 5, and 10 g/plant for N. Since the seedlings grew for one month, the seedling ground diameter, plant height, canopy length, lateral branch length, leaf chlorophyll content were monitored in frequency of every two months, and total 7 monitors. Based on ANOVA and multiple comparisons, the seedlings with significantly lower indices were defined as being in an unsuitable growth environment (recorded as 0), otherwise in a suitable environment (recorded as 1). Three representative seedling were chosen from each treatment for taking canopy images at the initial of monitoring, then the color feature parameters of the images were extracted as independent variables by segmenting the images. A classification model was constructed to diagnose whether the current environment suitable or not suitable for seedling growth. The chlorophyll relative content (SPAD) were estimated by partial least squares regression (PLSR) and stepwise regression (SR) models. Simultaneously, the dummy variables converted by binary classification was added into the model.
Results The seedlings growth of C. mannii were impacted significantly by transmission rate and irrigation level, not by nitrogen application rate. The support vector machines (SVM) effectively identified seedlings under the unsuitable environments (the false positive rate was 14%). PLSR and SR models well estimated the SPAD values using image color feature parameters as independent variables, and PLSR model adapted better than SR model in solving the collinearity and dimensionality reduction problems. The conversion of two categorical variables into dummy variables did not improve the prediction accuracy of SPAD values under suitable environment, but did that under unsuitable environments. Using this model to invert the SPAD of seedlings under unsuitable environments, the lower leaf tips appeared chlorosis, indicating the reliability of the prediction.
Conclusions The growth of Cephalotaxus mannii seedlings is mainly impacted by light and irrigation rate, with significant effects on the lower leaf tips by light and irrigation. The color feature parameters captured from the canopy images are effective in diagnosing the growth status of Cephalotaxus mannii, especially under unsuitable environments such as low light and poor irrigation. Additionally, the PLSR with dummy variables model can enhance the predictive ability of SPAD for class 0 samples. Therefore, the method can be used in non-destructive diagnose under the growth environment of Cephalotaxus mannii.