Abstract:

The quantitative estimation of chlorophyll content in canopy leaves of Hami melon by spectral technology can provide a theoretical basis for water and fertilizer regulation and field management. On the basis of the original spectrum, this experiment used multivariate scattering correction, standard normal variable intersection, standardization, Savitzky-Golay convolution smoothing, normalization and moving average smoothing to preprocess the original spectrum data, and then combined the feature interval selection with the feature wavelength selection to achieve the purpose of data dimension reduction and model simplification. The regression models of partial least squares and extreme learning machine were established. The results showed that the preprocessing effect of multiple scattering correction was the best. On this basis, 13 characteristic wavelengths were selected by combining the inverse interval partial least squares (BiPLS) and the competitive adaptive reweighting sampling (CARS) algorithm, which were used as the input variables of the model. The model established by partial least squares (PLS) had the best effect. The correlation coefficient R_p and root mean square error (RMSEP) of the prediction set were 0.942 4 and 1.006 2, respectively. Therefore, the spectral quantitative analysis model established by BiPLS and CARS combined with PLS could realize the quantitative estimation of chlorophyll content in Hami melon canopy leaves.

Key words: canopy of Hami melon, SPAD value, feature interval selection, characteristic wavelength selection, partial least squares, estimation model