Multi-Spectral Images Estimation Models for Detection of Rape Nitrogen Contents

This paper adopted the multi-spectral image analysis method to quantitatively analyze the oilseed rape total nitrogen contents. The reflection intensity distribution information of images in the visible and near-infrared spectral bands is used for nondestructive testing the nitrogen contents of oilseed rape. The multi-spectral image features of the nitrogen contents of oilseed rape in different growth stages are preliminarily proved up. The multi-spectral camera was used to get the multi-spectral images of oilseed rape canopy and the got images were preprocessed using the median-filtering method. Then two-dimensional maximum entropy segment method was used to carry out background segmentation of rape multi-spectral images. By extracting mean & ratio features of the multi-spectral images of oilseed rape canopy, it was found that the features of ARV1, AVS560, ADV1, AVS660 & g are highly correlated with nitrogen content during the whole rape growth period. The prediction model of nitrogen content of oilseed rape in different growth stages was built by stepwise regression method considering the serious multi-collinearity between multi-spectral variable. The result shows that the multi-spectral image analysis method can be use to quantitatively analyze the oilseed rape total nitrogen contents. This provides good supporting information for the scientific management of rape nutrition.