Appearance Modeling Of Wheat And Maize Leaves

Appearance of plant leaves is the process of interaction between light and leaf surface, which could be represented by image, hyspectral reflectance data and plenoptic function. Because the leaf appearance reflects the distribution of inside substance, it plays an important role in plant physiology, plant growth prediction and visualization of virtual plant. Recently, appearance modelling of plant leaves is mostly based on digital images, and hyperspectral data. Also, different models could be used in different research including plant nutritional status prediction, diseased recognition and visualization. The research objects of this paper are the leaves of wheat, corn, etc. This thesis builds different appearance model of plant leaves for different researches and applications by using different techniques including digital image processing, hyperspectral data analysis, image measurement and realistic rendering, also the corresponding softwares are developed for different model.The content of this thesis is as following:(1) Disease recognition model for plant leaves: We acquired digital images of four kinds of wheat diseases in complicated background. After image preprocessing, background segmentation, spot segmentation, feature exaction and selection, multiple classifier system design, we developed a disease recognition model. Also, we developed an application by VC++ and encapsulate the classes into DLL. In the recognition model, we designed a SVM-based multiple classifier system with stacked generalization framework, which could bring in high-level knowledge of plant physiology and realize a high recognition rate.(2) Realistic appearance model for plant leaves: Firstly, we built a realistic appearance model for plant leaves which is based on Ward BRDF. Then using optics mechanical facilities such as linear source, backlighting source, camera and stepper motor, a BRDF acquire system for plant leaves is built. This system, which is a image-measurement system, could get reflectance and transmission properties of a leaf sample through analysing its image sequence. Finally, we analysed an appearance model of aging leaf by independent component analysis (ICA) and separated two independent components including two relative quantity of green pigment and aging pigment. Also an aging appearance synthetizing model was built. The results of visualizing simulation demonstrate that the proposed appearance model could render realistic image of plant leave, also the ICA-based appearance synthetizing model could simulate the forward/backward aging process through changing the quantity of two pigments in aging leaves.(3) Nutrition prediction model of appearance for plant leaves: We acquired 125 corn leaf samples in three different growth period including silking period, pustulation period, mature period through pushbroom imaging spectrometer (PIS). The spectral reflectance data are extracted using the way of combining image with spectrum, also the corresponding chlorophyll data of biochemical component, and then partial-least-square regression analysis (PLS) is applied to build regression model in three different periods and the whole growth period. The simulation experiments demonstrate good fitting performance for test data and these four prediction models.