Research on the Estimation Model of Calcium Content in Guanxi Honey Pomelo Leaves Based on Spectral Index

By constructing the estimation model for calcium (Ca) content in honey pomelo leaves based on spectral analysis, it could provide a theoretical basis for monitoring and rapid non-destructive diagnosis of Ca content in honey pomelo leaves.

The original spectral and first-order derivative spectral characteristic bands and spectral characteristic indices (difference spectral index (DSI), ratio spectral index (RSI), and normalized difference spectral index (NDSI)) were analyzed and extracted. Single variable estimation model, partial least squares estimation model (PLS), backpropagation neural network estimation model (BPNN), random forest estimation model (RF), and support vector machine estimation model (SVM) for honey pomelo leaf calcium content were established, and the optimal spectral estimation model for honey pomelo leaf calcium content was evaluated and verified.

There was a significant multi band correlation between the original spectrum and first-order derivative spectrum of pomelo leaves and calcium content. Based on the correlation coefficients of the original spectrum and first-order derivative spectrum, the maximum wavelengths were 553, 714 nm and 528, 699, 602 nm, respectively. The spectral indices with significant correlation between the original spectrum, first-order derivative of pomelo leaves and calcium content were DSI_790,1040, RSI_910,990, NDSI_900,990 and NDSI′_350,580, DSI′_560,570, RSI′_350,580. The polynomial estimation model constructed with spectral indices such as RSI_910,990, NDSI_900,990, NDSI′_350,580, DSI_790,1040, DSI′_560,570, RSI′_350,580, DSI′_528,602 as independent variables had relatively high determination coefficient R² (R²>0.60). A hyperspectral estimation model for calcium content in honey pomelo leaves was established using the above four machine learning methods. The R² of PLS, BPNN, RF and SVM estimation models were 0.79, 0.82, 0.85 and 0.84, respectively, and the root mean square errors (RMSE) were 4.33, 4.11, 3.81 and 3.93, respectively; the R² of the validation models were 0.77, 0.80, 0.87 and 0.83, respectively, and the RMSE were 4.50, 4.28, 3.67 and 3.90, respectively. The order of estimating the accuracy of the model was RF>SVM>BPNN>PLS.

The accuracy comparison analysis of four models for calcium content in honey pomelo leaves showed that the RF estimation model had better predictive performance than the other three estimation models. This result could provide a new method for rapid diagnosis of calcium content in honey pomelo leaves for reference.