[A hyperspectral assessment model for leaf chlorophyll content of Pinus massoniana based on neural network]

The relationships between the leaf chlorophyll content (LCC) of Pinus massoniana at different growth stages and their chlorophyll content were analyzed. 7 of 36 red edge-based parameters were finally selected as the typical spectral response parameters which held the most significant statistical relationship with LCC, and then the hyperspectral assessment model for retrieving the LCC was built based on stepwise regression analysis method and B-P neural network, respectively. In the same way, four different vegetation indices (VIs) were selected as typical spectral parameters, in the meantime, the first four components of the principal component analysis (PCA) transformed from original spectral measurements were inputted into the B-P neural network, and then the hyperspectral assessment model for retrieving the LCC was built based on stepwise regression analysis method and B-P neural network, respectively. The results showed that R² of the red edge-based stepwise regression model and the red edge-based B-P neural network model were 0.5205 and 0.7253, RMSE were 0.1004 and 0.0848, and relative errors were 6.3% and 5.7%, respectively. R² of the VIs-based stepwise regression model and the VIs-based B-P neural network model were 0.5392 and 0.7064, RMSE were 0.0978 and 0.0871, and relative errors were at 6.2% and 6.0%, respectively. The prediction effect of PCA-based B-P neural network model was the best, R² was 0.7475, RMSE was 0.0540, and the relative error was 4.8%.