Potential of hyperspectral imagery for nitrogen content retrieval in sugar beet leaves
Potentiel de l'imagerie hyperspectrale pour l'estimation de la teneur en azote dans les feuilles de betteraves sucrières
Résumé
Leaf nitrogen content (LNC) is one of the most important limiting key nutrients in sugar beet crops, so plant nitrogen status has to be carefully monitored throughout the plant life. In this study, close-range hyperspectral imaging was used to infer LNC from reflectance spectra in a non-destructive way and under in-field conditions. First, after acquisition, images were preprocessed in order to remove some sources of variability that were not correlated to LNC, such as specular reflection and spectral noise. For every hyperspectral image, the mean leaf spectrum was then evaluated and associated to the actual average LNC value measured on the same plants. Partial Least Square regression was used to calibrate a regression model. With six latent variables, LNC was accurately predicted with a low error and a high coefficient of determination (RMSECV = 1.72 g/kg; R² = 0.86). When applied to individual spectra of hyperspectral images, this model led to a consistent LNC map of sugar beet leaves, i.e., LNC was low in old nitrogen-deficient leaves and it was high in young wide leaves. Such a mapping is therefore a valuable non-destructive evaluation tool to better understand how LNC is distributed within plants and to identify LNC-deficient zones.
Origine | Fichiers produits par l'(les) auteur(s) |
---|