Research On Detection Of Nitrogen Nutrition Levels Of Oilseed Rape Based On Multispectral Machine Vision

In this research,we use nutrition liquid in order to have the oilseed rape in different nitrogen level,we use multispectral vision system for taking image of oilseed rape canopy which are under different nitrogen level,processing the image and then extract the features which are highly correlated with SPAD values using the machine vision technology.For figuring out the diagnosis model for oilseed rape nitrogen,we use the linear regression.The main research results are below:Designed the hardware of multispectral machine vision system of diagnosis of oilseed rape nitrogen content,Moreover,we designed a obturation illumination light box for experimenting and taking image of different rape sample canopy.By analyzing the 1-D histogram and 2-D histogram of every channel's multispectral image of rape sample,finally we choose the IR channel's image to perform the image segmentation.We segment the IR image of rape canopy in the way of maximize the 2-D entropy,and we got a good result.Meanwhile,we choose the OTSU method to contrast,and it was shown that the entropy algorithm keep more information about plant canopy when it segment the soil background,the segmentation error of it is 9%,it is better than OTSU.In the aspect of feature extraction,after the image segmentation,average the intensities of each multispectral image of oilseed rape canopy.We got the three features of AGI,AIR and IR/R,these three features are highly correlated with rape nitrogen content.After a further analysis,we discovered thatwe can diagnose the rape nitrogen in general using feature IR/R,if 1＜IR/R＜1.4,then therape is normal status,if IR/R＜1,then the rape is in lack of nitrogen nutrition.But we can not diagnose nitrogen content quantitatively by using IR/R.Using the multi linear regression between several image features and SPAD values of rape canopy in different life span.Each prediction model is built for diagnois of rape nitrogen content in each life span.The R~2 of the prediction model is 0.76,0.94,0.84 and 0.85 repectively in the seedling time,bud time,flower time and fruit time.To validate the model,we get the relative average error are 13%,9%,16%and 16%respectively.This shows the feasibility and accuracy of our multispectral prediction model.