Abstract:

The long-term gradient experiment of nitrogen fertilizer was selected as the study area, and the collected 42 samples of chlorophyll contents from maize leaf were treated as training samples. Unmanned Aerial Vehicle (UAV) hyperspectral remote sensing system was used to obtain maize hyperspectral images, and seventeen spectral parameters were extracted from these hyperspectral images. Using correlation analysis result, four spectral parameters, which closely related to the SPAD values or the values of after mathematical transformation (R, lg^R, 1/lg^R,1/R,\sqrt[]{R} and\sqrt[]{\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$R$}\right.}) were selected and used to build unary or multiple regression equation. The results showed that λr, Rg, SDb, NDVI were the most closely related to SPAD values. The coefficients of determination between SPAD values and these four parameters were relatively high, which were 0.19, 0.83, 0.71 and 0.80, respectively. What’s more, the determination coefficient of multivariate regression equation between the four parameters and SPAD values was 0.93. The test results for the unitary or multivariate diagnostic model by using 21 independent samples from the training samples showed that the multivariate diagnostic model of SPAD values based on λr, Rg, SDb and NDVI had higher prediction accuracy. Compared with previous research results, the diagnostic model established in this study has the advantages of simple parameters and higher determination coefficient, which could be used as the data support for related field research.

Key words: hyperspectral, maize, chlorophyll, diagnostic model, regression analysis