基于地基激光雷达的全生育期玉米叶面积指数提取方法研究

Leaf area index of maize is a key parameter of maize growth model using remote sensing technique, which is an important indicator for maize cultivation and maize breeding. Aiming at decrease the error of leaf area index extraction, this study exploring the method of estimating leaf area index accuratly, automaticly, using terrestrial LiDAR points. The terrestrial LiDAR points echoing on maize leaves and stalks are identified accurately from the massive, discrete raw LiDAR points which don't have spatial relationship using the cylinder model segmentation algorithm and random sample consensus (RANSAC) segmentation algorithm. The LiDAR points which should echoing on the leaves within internal canopy of middle and later growth stages are predicted using the improved random forest (RF) algorithm, which solving the problem of incomplete data in middle and later growth period. Based on the expression method of canopy structure and the graphical interpretation of turtle in L-system, we reconstruct the real three-dimensional canopy model of maize. These realistic three-dimensional canopy models in the whole growth period are used to compute the real leaf area index. At the same time, the leaf clumping indexes in early stage and middle and later stages are computed based on these real three-dimensional canopy models, then the effective leaf area index in the whole growth period are computed. The leaf area index of canopy exposure to light in the whole growth period are computed based on light energy distribution which is depicting through the radiation flux of all leaf surface stacks and the energy exchange between them. Three kinds of leaf area index in the whole growth are estimated which has the discrete and continuous features at the same time. This project will lay the foundation for growth monitoring and yield prediction of maize, and there is great significance in maize cultivation and maize breeding.

玉米叶面积指数是玉米生长遥感模型的关键参数，也是玉米栽培和育种需考虑的重要指标，针对目前该指数提取误差大的问题，本项目以地基激光雷达为数据源，开展该指数的自动、精确提取方法研究。探索圆柱体模型分割和随机采样一致性分割方法，准确、快速的识别玉米叶片和茎秆回波点云；改进RF预测方法，预测玉米生长中后期冠层内部叶片回波点云，建立玉米冠层结构框架的表示方法和龟形基础的图形解释方法，重建全生育期玉米冠层真实三维模型；基于重建的三维模型提取真实叶面积指数，并分别计算生长前期与生长中后期的叶片集聚指数，从而提取全生育期玉米有效叶面积指数；基于RGM辐射度模型求解每个玉米叶片小面元的辐射通量及相互之间的能量交换，计算冠层内光能分布特征，提取全生育期玉米受光叶面积指数。项目旨在提高同时具有离散与连续双重特征的全生育期玉米叶面积指数提取精度，为玉米长势监测、产量预测奠定基础，同时也对玉米栽培和育种具有重要意义。

玉米叶面积指数是玉米生长遥感模型的关键参数之一，也是玉米栽培和育种需考虑的一种重要指标，针对目前利用光学遥感技术提取误差大的问题，本项目联合使用以地基激光雷达数据和光学遥感影像，开展区域范围内全生育期玉米冠层叶面积指数的自动、精确、连续提取方法研究。本项目围绕全生育期玉米叶面积指数提取，从以下几方面进行了研究：（1）针对无效散射点、粗差点和玉米茎秆回波点云影响玉米叶面积指数提取精度的问题，利用曲面拟合的方法分离地面回波点云后，利用DON法线差分方法准确识别玉米冠层叶片回波点云，从而解决了玉米茎秆等非光合器官回波点云影像叶面积指数提取精度的问题。（2）利用改进的Delaunay三角网生成算法，采用从局部三角化出发来构造全局三角化的方法，建立玉米冠层三维模型，并利用L-系统对封垄后单株玉米的表达方法进行预测、补充，解决生长中后期冠层内部数据不完整的问题。（3）基于多个关键生育期的玉米叶片回波点云，提取关键生育期的玉米冠层叶倾角概率密度分布函数，改进PROSAIL模型中的Compell叶倾角概率密度分布函数，建立查找表并通过代价函数反演全生育期玉米叶面积指数，并以野外实测的不同生育期的玉米叶面积数据为真值，对全生育期的玉米叶面积指数提取结果进行精度评价与验证。本项目的全生育期玉米叶面积指数提取方法的研究，为玉米生长过程监测、产量预测、玉米生长过程中的潜在胁迫监测奠定基础；同时，玉米叶面积指数也是一种重要的表型参数，该研究成果也对玉米栽培和育种具有重要意义。