Rice Nutrition Diagnosis And Modeling Based On Digital Image

Recently, digital diagnosis technology have been well developed and widely used in our dairy life. Recognization of nutrition status is a hot topic in agricultual system. For rice system, efficient, quick and practical are the key words. In this study, we proposed two platforms to get the canopy and leaf image of paddy field. One is remote-controlled helicopter (Herakles II) set up with digital camera, the other is the scanner. Significant efforts have been made in the development of the modeling. A good relationship between digital image characteristics and NPK status of rice is developed. Additionally, auxiliary factors play an important role in modeling the recognization of nutrition pattern. The results are as follows:1. Comparison of images obtained from different sensorsThe scanner and the digital camera were selected to collect images of rice leaves. After compared with characteristics of color, contrast and informativity between these two images, there was no significant difference. Howerer, the informatin got from the scanner were more exactly, as the scanned image can make the shape and texture information more stable with less noise information. Thus, the scanner was used to extract reliable fractional information of rice leaves, while canopy information of rice was collect by the use of digital camera.2. Establishing diagnose rules of rice NPK nutrition statusRice leaves showed significant variations at different nutrition status. After analizing the characteristics of the digital image, leaf color information, leaf texture (mean, entropy) and spot area information played an important role in recognizing the N, P and K nutrition. In addition, rough set theory was used to establish the recognization rules for different nutrition status. The results showed that leaves with single nutrient deficiency can be recognized with high rate, with a precentage of 77.8% (N),97.1% (P) and 100% (K) respectively. Note that the old leaf can get a better result compared to the new one.3. Recognization of nutrition status by the use of remote-controlled canopy imagesDue to the advantages of real-time, exact and high efficiency, the remote-controlled helicopter with digital camera was selected as the monitoring platform. There was a good relationship between image characteristics and observations. Specifically, hyperspectral characteristics in the visible region and the parameter of dark green color index (DGCI) have got a significant quadric relationship with nitrogen content. Additionally, RGB and HSI color space and five texture characteristics also had outstanding relationship with N nutrition status. All the results indicated that it was practical to recognize the N nutrition status at canopy scale by the use of this proposed platform. Samples without nitrogen input had got highest accuracy, while the accuracy of samples with normal N input was lowest.4. model establishment for predicting leaf nitrogen contentThe selected factors in establishing the recognization modeling are different at leaf and canopy level. For leaf scale, the factors are color, texture and shape, with contribution rates 40.04%,29.62% and 26.44%, respectively. For canopy scale, there are color and texture factor, with contribution rates 85.55%,9.27%, respectively. A model was established as follows N=2.8967e-0.3312FZ and N=-0.01FZ2-0.39FZ+2.3155. Quantify expert experiences and its auxiliary role in nutrition diagnoseLeaf characteristics are different. In order to insure accuracy and extensive applicability, a lot of expert diagnosis experience has been used. In this study, rice samples were classified using three empirical characteristics (RLS, DL, CI5/CI4) and then quantified by the use of supervised discretization method. Interval rules were established based on the classification results.