基于t分布函数的玉米群体三维模型构建方法

为利用少量实测数据快速构建能够反映因品种、环境条件、栽培管理措施等因素产生形态结构差异的玉米群体三维模型,提出基于t分布函数的玉米群体三维模型构建方法。通过实测数据构建主要株型参数的t分布函数,在其约束下生成群体内各植株主要株型参数,通过构造株型参数相似性度量函数调用玉米器官三维模板资源库中的器官几何模板,结合人工交互或图像提取的各植株生长位置与植株方位平面角2组群体结构信息生成玉米群体几何模型。利用三维数字化仪获取的玉米群体田间原位三维数字化数据所构建玉米群体计算得到的LAI与该方法构建玉米群体计算得到的LAI进行对比验证,结果表明:该方法所生成玉米群体叶面积指数与原位三维数字化数据所构建玉米群体计算得到的LAI相比,误差在±2%以内,可以满足面向可视化计算的玉米结构功能分析研究需求。方法可为玉米株型优化设计、耐密性鉴定、品种适应性评价等虚拟试验研究提供技术手段。

Construction method of three-dimensional model of maize colony based on t-distribution function

Abstract: Crop colony is the organization system which performs photosynthesis and dry matter production function. Its morphological structure has important influence on light interception ability, canopy photosynthetic efficiency and crop yield. The morphological characteristics of crop colony have always been the most basic way for people to recognize, analyze and evaluate crops. Therefore, it is of great practical significance to rapidly and accurately model and analyze the morphology of crop colony in a digital and visual way. Morphological data acquisition of maize colony is labor-intensive and time-consuming, and thus a t-distribution based three-dimensional (3D) maize colony modeling method was proposed using a few measured data. The method constructs t-distribution function of primary plant morphological parameters using measured data and generates random plant morphological parameters under the constraint. The main plant morphological parameters include plant and phytomer scale. Here plant scale parameters include plant height, total leaf number, and first leaf index, and phytomer scale parameters include leaf growth height, leaf insertion angle, leaf length, leaf width, and leaf azimuthal angle. Particularly, leaf azimuthal angles are generated using the deviations between the plant azimuthal plane and leaf azimuths. High quality geometric models in 3D template resource database of maize organs are selected by constructing a similarity assess function of plant morphological parameters. Leaf length, leaf insertion angle, leaf index, and plant cultivar are the control parameters in the function. Then geometric models of individual plants in target colony are generated. Interactive design or field image extraction method is used to allocate the growth positions and plant azimuthal planes of each plant in the colony. Maize colony is generated by moving and rotating operations of each plant according to the designed or extracted growth positions and plant azimuthal planes. Leaf area index (LAI) is used to validate the generated maize colony model. Three in-situ field measurement experiments in Qitai County of Xinjiang using 3D digitizer were carried out to reconstruct geometric models of maize colony, and the cultivar was Xianyu 335 and the planting densities were 105, 135, and 165 thousand plants/hm2, as true values for LAI calculating. Corresponding plant morphological parameters of the corresponding colonies were measured. The maize colony modeling method based on t-distribution function was used to construct 3D models and LAI was also calculated for the colonies. Results show that the LAI errors are less than ±2%. In addition, generalized LAI of different heights of plant colony was proposed to provide more detailed verification in different height levels. The averaged RMSE (root mean square error) of Xianyu 335 with the density of 135 thousand plants/hm2 is 0.023, and the averaged NRMSE (normalized root mean square error) is 0.425, which demonstrate that it has a good consistency of spatial leaf distribution between the in-situ measured field colony and reconstructed colony using t-distribution. These results show that the proposed maize colony modeling method could meet the needs of plant functional-structural analysis. Compared with the existing methods, the proposed method is more effective and highly realistic, and the constructed maize colonies are capable of reflecting the agronomic characteristics of the target colony, such as the differences caused by intrinsic cultivar, environment, planting, or management factors. Maize colony model could be rapidly generated by simple modification of morphological input parameters. Combined with the light distribution simulating algorithm, a large number of maize colony models will be designed for virtual experiments. It has great importance for the research and application of maize plant morphology optimization, estimation of planting density, adaptability evaluation of different cultivars, and cultivation strategy decision. Due to the complexity of maize colony structure morphology, there are still many subsequent colony modeling issues that will be addressed in future research, such as adjacent phytomer parameters constraint model construction, plant collision detection and collision response, and colony mesh simplification and optimization for visual computing.