Abstract:

To solve the problem in diagnosing nitrogen nutrient in rice leaves based on hyperspectral reflectance, we extract weak information from the spectral signal to estimate nitrogen content by applying wavelet analysis to decompose the reflectance spectra. Hyperspectral data were collected in China’s northeast city Changchun. The data were used to develop predictive models of nitrogen contents and test the accuracy of the models. We decompose the reflectance and derivative spectra of rice canopy into eight levels using the Db5 function of Daubechies wavelets, and establish 192 models among the obtained wavelet coefficients at different levels and different decomposition positions. By comparison, wavelet coefficients with high precision are selected to establish the best model. The results indicate that wavelet coefficients can be used to obtain accurate prediction of nitrogen content with a high correlation coefficient up to 0.99. The wavelet-based approach outperforms predictive models based on a range of existing spectral indices, showing good prospects in applications for estimating biochemical components of crops that need to extract weak information from hyperspectral data.

Key words: hyperspectral remote sensing, wavelet analysis, weak information extraction, nitrogen content, rice leaf