Appearance modeling and visualization of maize leaf with agronomic parameters

Abstract: In this paper, we presents a method for modeling and visualization of crop leaves based on agronomic parameters in order to improve the production efficiency of three dimensional digital media for agricultural subjects. The field experiments were conducted in the experiment farm of Beijing Academy of Agricultural and Forestry Sciences, Beijing City. During maize growth periods of jointing stage, horn mouth stage, florescence stage, grouting stage, maturation period and senility period, SPAD values of maize leaves were obtained by using SPAD-502 and four appearance parameters were collected by an apparent image acquisition system using linear light source. The four appearance parameters were diffuse reflectance, transparency, specular reflectance and roughness. In order to finely model the maize leaf appearance, leaf surface was divided into three structures including mesophyll, main vein and secondary vein. With the data, we analyzed the relationship between SPAD, growth period parameters and various kinds of appearance parameters of the three structures, respectively. We applied a variety of mathematical models to fit the data to describe the potential relationship between SPAD, growth period and diffuse, transparency appearance of mesophyll. Based on R2 and RMSE, we selected quadratic function for diffuse reflectance and negative exponential function for transparency. The coefficients of determination for the RGB channels of final diffuse model of mesophyll were 0.9137, 0.8511, and 0.6146, respectively, and their corresponding values of RMSE were 0.01549, 0.0128, and 0.01533. The coefficients of determination for the transparency model were 0.9975, 0.9913, and 0.9907, respectively, and their corresponding values of RMSE were 0.0064, 0.0096, and 0.0207. Data showed diffuse reflection and transparency of secondary vein could be classified into four sections by SPAD values and there was little difference between each class of appearance values. We used mean values of each section as diffuse and transparency parameters. The diffuse reflection and transparency of main vein has little change in whole growth periods. As such, we used mean values as unique and invariant apparent feature in diffuse and transmission reflection. We found that specular reflectance and roughness parameters of the three structures were not related to SPAD and growth period, therefore, mean values of these two appearance parameters obtained from all samples were used as a glossy appearance. Based on the above analysis, a series of quantitative appearance models for three structures were established respectively with SPAD and growth period as input parameters, and appearance material parameters as output. The spatial-varying appearance of leaf surface formed specific texture patterns. We abstracted the texture pattern and constructed a parameterized structure method to generate a structure texture image for synthesis of the geometric features of mesophyll, main vein and secondary vein structures. Each pixel of structure texture image represented a structure category of the point at this position, and we used the appearance models of each structure to assign appearance parameters for each pixel. Finally, four appearance textures were generated for simulating the appearance patterns of the leaf surface including diffuse texture, transparency texture, specular texture and roughness texture. A real-time rendering framework was developed to simulate the subtle interaction between plant leaves and light. In order to achieve a WYSIWYG display result, we simplified the light computing by decomposing light environment into a directional light source (sun) and some environment light sources (sky). A complete realistic effect was finally developed by combing the appearance model and illumination computation using a deferred lighting rendering framework which can simulate real-time shadows, ambient occlusion, and dynamic radiance from sun and sky. The experimental results demonstrated that the proposed approach was capable of generating different appearance of crop leaves by controlling the agronomic parameters and achieving a visually satisfactory display result. Our method showed diffuse reflectance and transmittance were the most affected by the parameters of SPAD and growth stage, while the specular reflection and roughness parameters were always stable. There were differences in the fitting results of diffuse and transparency model of mesophyll in RGB channels, while simulation result indicated there was little impact on the final visualization effect. Our method has a great potential to become an effective visualization tool for agricultural application, such as crop model, agricultural education system, and digital media for agricultural subjects.