Studying On Nitrogen Deficiency Diagnosis Model Of Rubber Seedling Based On Image Recognition

Compared with the traditional diagnosis of plant nutrition, computer vision technique is applied to lots of researches due to new, real time and convenient. In this study, I set up the regression model between nitrogen concentrations in rubber tree leaves and color feature value of leaf images by the comparison of the data of leaf images and chemical analysis. Ultimately, we have realized a rapid, quantitative method to diagnose nitrogen concentrations based on computer vision technique.In our study, the rubber tree seedlings were cultivated with deficiency nutrition’s solution culture method under the greenhouse conditions. Then the leaves we collected were chemically analyzed and taken pictures in order to illustrate their correlations between the leaf nutrient content and the image eigenvalue. The results are shown as follows:1. Different resolution camera can reflect the features of the color of the rubber blades. Compare different resolution camera to collect the same set of leaf image characteristics, found that while a camera to obtain the image characteristics of the different values are different, but their descriptions of the same set of blade changing law is similar. Therefore different resolution of the camera on the rubber blades can be used for image retrieval.2. The samples containing different nitrogen levels from deficiency nutrition’s solution culture were collected, and then analyzed through the comparison between the data of the images and nutrient contents. Our study suggested that there were significantly correlated the nitrogen (N) contents with color variables R and H, respectively.3. Eight regression models of nitrogen (N) nutrition diagnosis were instituted under RGB and HIS color system. Both the upper and lower surfaces of the blade were analyzed by regression analysis method (e.g. one variable linear, multivariate linear and nonlinear) under RGB and HIS color system, respectively. Our results showed that nitrogen (N) contents in the upper of leaves were significantly related with cubic regression equation of R/(R+G+B) (R2=0.872), R/(G+B) (R2=0.865) and H (R2=0.864),as well as nonlinear regression equation of H (R2=0.865)respectively. Besides, I found that nitrogen (N) contents in the lower of leaves were significantly related with cubic regression equation of H (R2=0.853) and R (R2=0.814), and nonlinear regression equation of R (R2=0.808) and H (R2=0.845).4. We obtained a best prediction model on rubber tree leaf nitrogen (N) nutrient diagnosis through comparing the eight models, which can exactly calculate the rubber tree leaf nitrogen (N) contents. The equation was shown below:CN= 50.11-282.9R/(R+G+B)+546.OR/ (R+G+B)2-349.8R/(R+G+B)3 (R2=0.886; S= 0.391).