Calculation Method Of Three-dimensional Nitrogen Distribution In Rice Plant

Nitrogen(N)is one of the essential mineral elements in the normal growth of crops.Crop growth and yield can be affected by the nitrogen nutrition status.Influenced by translocation and allocation,crop N usually shows a vertical distribution in canopy.The lower leaves are more easily to show a deficiency symptom when plants lack of N.The three-dimensional(3D)modeling technologies based on imaging can be used to rebuild crop morphology and the estimating methods of crop N based on imaging spectroscopy can be used to monitor N status in crop plant.Both of them are all capturing data by imaging technology,and also they are becoming more and more mature.So,in this study,these two technologies were combined to calculate the 3D distribution of N in rice plant based on multi-view RGB images and near-infrared(NIR)images.The main works and results were as follows:1)The multi-view imaging system were designed and implemented,which included hardware platform establishment and system control software development.The results showed that:a)the software was compatible with the hardware and the whole system worked well,also,the work status in each hardware module can be controlled precisely with the user interface;b)the imaging system can be used to capture batch of RGB images and NIR images whose wavelength were 720nm,760nm,850nm,950nm and 1000nm respectively in multi-view automatically.2)The shape-from-silhouette(SFS)technology was firstly used to build 3D model of a toy block,and then applied to rebuild 3D morphology of rice plant.Finally,the effect of SFS and REVscanTM laser scanner in establishing 3D model of rice were compared.The results showed that:a)the lager the size of a morphological index in a same object,the higher the precision of its 3D model;b)the SFS was suitable for 3D modeling of rice plant with well precision,for example,the accuracy of rice leaf length was 0.997(R2);c)for the morphology and texture of rice plant,3D model based on the SFS was more similar to the real rice plant than that based on REVscanTM laser scanner.3)Vegetation indexes(VIs)were selected and calculated from RGB images and NIR images of rice captured by multi-view imaging system firstly.Then,the best regression model was established between VIs and SPAD readings based on 5 function prototypes which were linear,power,exponential,logarithmic and quadratic respectively.Finally,the best prediction model of N was transformed from estimating model of SPAD readings.The results showed that:a)the internal relationship between VI and SPAD readings or N concentration was closer to quadratic function;b)maximum accuracy(R2=0.809)of SPAD readings regression model was determined by GVI760;c)the best VI of N was GVI760 and the sensitive spectral band of N were G and NIR760R in this study.4)By using the multi-view imaging system,images that contained sensitive bands of N in multi-view were acquired firstly.3D models which were texture mapped with sensitive band were established secondly.3D N distribution in rice plant was calculated by applying the best estimating model to the texture data on 3D model.It was found that the result can identify N distribution trend not only in a single leaf blade but also in the whole canopy or in different rice plants.