基于地面LiDAR的水稻生物量高精度反演

发布时间：2018-06-24 06:12

本文选题：水稻 + 三维激光扫描技术　；参考：《新疆大学》2017年硕士论文

【摘要】：随着中国的发展,现代化进程的加快,工业化的兴起,农业的发展不可避免地受到影响。中国可谓是一个人口大国,农作物总产量位居世界排名前列,但是人均占有量却很低。因此,如何实时的掌握农作物的生长及监测其生物量,以便为政府提供粮食的基础信息,成为现阶段的研究方向之一。现在国内外对森林生物量的研究颇多,但是用地面LiDAR技术对农作物生物量的研究是一个新的研究方向。从传统的技术到遥感监测技术,再到现在的激光雷达技术,人们对科学的监测农作物的生物量做了大量的研究。激光雷达技术自开发以来被广泛的应用于军事、医学、工程和森林等领域。本文运用FARO Focus3D三维激光扫描仪获得水稻点云数据,该仪器具有携带方面,高性能、易于操作等优点。应用三维激光扫描技术可获得水稻植株的三维点云数据,再对点云数据进行拼接、精简和去噪等预处理。三个处理过程中去噪最为重要,本文依据噪声的特点应用了半径滤波、统计滤波和移动最小二乘法这三种方法去除噪声,并用F-score方法验证去噪精度,通过验证统计滤波在此研究中去噪效果最好,精度最高。本文选取株高和体积两个参数用于估测生物量。株高可直接在Cloud compare软件中获取。体积(m3)分别用三维凸包法和Voxel法两种方法获取,在MATLAB中运行相应的代码即可得到体积。利用SPSS和MATLAB软件建立基于株高的生物量模型、基于三维凸包体积生物量模型、基于Voxel体积生物量模型、基于株高和三维凸包体积生物量模型和基于株高和Voxel体积生物量模型,共计建立15个回归模型。通过验证可知三维凸包和株高、Voxel体积和株高建立的复合回归模型较好,决定系数都在0.8以上。最优模型是三维凸包体积和株高建立的二元一次回归模型,决定系数达到0.87,回归结果较理想。研究结果表明用水稻株高和体积估测生物量是一个可行的方法,能够高效精确的估测以水稻为代表的农作物生物量(g/m2)。
[Abstract]:With the development of China, the acceleration of modernization and the rise of industrialization, the development of agriculture is inevitably affected. China is a populous country with the highest crop output in the world, but low per capita share. Therefore, how to grasp the growth of crops and monitor their biomass in real time, in order to provide the basic information of grain for the government, has become one of the research directions at this stage. There are many researches on forest biomass at home and abroad, but it is a new research direction to use surface LiDAR technology to study crop biomass. From the traditional technology to the remote sensing monitoring technology, and then to the present lidar technology, people have done a lot of research on the scientific monitoring of crop biomass. Since its development, lidar technology has been widely used in military, medical, engineering and forest fields. In this paper, the point cloud data of rice are obtained by Faro Focus 3D laser scanner. The instrument has the advantages of carrying high performance and easy to operate. The 3D point cloud data of rice plants can be obtained by using 3D laser scanning technique, and then the point cloud data can be spliced, reduced and de-noised. According to the characteristics of noise, this paper applies three methods to remove noise, such as radius filter, statistical filter and moving least square method, and uses F-score method to verify the denoising accuracy. It is proved that the statistical filtering has the best denoising effect and the highest accuracy. In this paper, plant height and volume were selected to estimate biomass. Plant height can be obtained directly from Cloud compare software. Volume (m3) is obtained by three dimensional convex hull method and Voxel method respectively. The volume can be obtained by running the corresponding code in MATLAB. The biomass model based on plant height was established by using SPSS and MATLAB software. The model was based on 3D convex hull volume biomass model, Voxel volume biomass model, plant height and three-dimensional convex hull volume biomass model and plant height and Voxel volume biomass model. A total of 15 regression models were established. The results showed that the complex regression model of three-dimensional convex hull, Voxel volume and plant height was better, and the determination coefficient was more than 0.8. The optimal model is a binary first-order regression model based on three-dimensional convex hull volume and plant height. The determination coefficient is 0.87, and the regression result is ideal. The results show that it is a feasible method to estimate biomass with rice plant height and volume, which can estimate the biomass of crops represented by rice (g/m2) efficiently and accurately.
【学位授予单位】：新疆大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN958.98

【相似文献】