【作者】 李鹏飞；

【导师】 张晓东；

【作者基本信息】 江苏大学 ， 农业工程（专业学位）， 2021， 硕士

【摘要】 番茄是我国种植最广的设施果菜之一,是菜篮子工程的重要组成部分。在设施环境中合理的种植和管理番茄是一直以来的重点,利用信息化技术手段获取番茄作物的长势信息不仅能够为水肥管理和调控提供依据,还能够提高番茄的产量和质量,以及有利于设施农业的绿色可持续发展。因此,为充分获取番茄的长势信息,本文综合利用3D光场成像、太赫兹成像及信息融合技术,开展对番茄作物的长势检测方法及应用研究。主要研究内容如下:(1)获取不同尺度下试验数据。使用3D光场成像系统采集不同钾浓度下番茄作物株高、茎粗、冠幅面积、叶面积、作物体积等外部表型信息;使用太赫兹时域光谱成像系统获取0-4.0THz范围内番茄叶片的太赫兹波谱层析图像信息;同时使用传统方法获取外部表型数据和内部营养含量数据。(2)对3D光场成像系统和太赫兹成像系统分别获取不同尺度的番茄长势信息进行处理。对于利用点云数据计算的茎粗、叶面积和冠幅面积特征参数,先对点云数据使用双边滤波算法进行点云去噪,然后再使用Space算法进行点云稀疏,对于茎粗特征参数寻找水平方向点云的最大值与最小值,然后计算差值;对于叶面积和冠幅面积特征参数使用德劳洛三角剖分算法将点云由独立点变成可用于计算面积的面图,然后再使用Python结合VTK库进行编程计算;针对植株体积特征参数,先对从不同方向获取的植株点云数据进行数据拼接,然后使用凸包算法进行体积计算;针对株高特征参数,先使用标板进行高度标定获取不同高度下对应的深度图颜色,然后对不同高度的番茄样本进行深度图获取,利用样本最高点和标板深度图进行比对算出样本高度;接着利用相关性分析筛选出与番茄植株营养长势相关性较高的特征参数,最后使用多元回归模型对番茄长势特征与作物营养含量建立多元回归模型。基于获取的太赫兹光谱图像数据,先使用滑动平均算法、迭代自适应加权惩罚最小二乘算法、多元散射校正算法进行数据预处理,然后使用主成分分析法进行数据降维,找出代表数据大部分信息的前三主成分进行图像分析并建立多元回归模型。(3)考虑到作物营养无损检测的单一检测方法具有局限性,将两个检测装置测量得到的数据一起代入到多元回归算法和BP神经网络算法中,然后分别得到不同模型算法下的相关系数,其中通过多元回归算法得到的不同时期的不同浓度下的相关系数在0.903-0.931之间,通过BP神经网络得到的相关系数在0.917-0.940之间,将同一数据分别利用两种模型算法计算时,使用BP神经网络得到的相关系数均优于使用多元回归算法。(4)由于上述试验设备价格比较昂贵,搭建了适用于温室的可搭载多种营养和长势检测设备的可移动检测平台,并介绍了该平台主要功能和部件的结构。更多还原

【Abstract】 Tomato is one of the most widely planted protected fruits and vegetables in China,and it is an important part of the vegetable basket project.Reasonable planting and management of Tomato in facility environment has always been the focus.Using information technology to obtain the growth information of tomato can not only provide the basis for water and fertilizer management and regulation,but also improve the quality of tomato,and is conducive to the green and sustainable development of facility agriculture.Therefore,in order to fully obtain the growth information of tomato,this paper uses 3D light field imaging,terahertz imaging and information fusion technology to carry out the research on the detection method and application of tomato growth.The main research contents are as follows:(1)The experimental data of different scales were obtained.In order to make the sample data closer to reality,we used perlite as the matrix,and used 3D light field imaging system to collect the external phenotypic information of tomato plant height,stem diameter,crown area,leaf area and crop volume under different potassium concentrations;Terahertz time domain spectral imaging system was used to obtain the terahertz spectral tomography image information of tomato leaves in the range of 0-4.0 THz;At the same time,traditional classical methods are used to collect data.(2)The tomato information of different scales obtained by 3D light field imaging system and terahertz imaging system were processed.For the characteristic parameters of stem diameter,leaf area and crown area calculated by the point cloud data,the bilateral filtering algorithm is used to denoise the point cloud data,and then the space algorithm is used to sparse the point cloud.For the characteristic parameters of stem diameter,the maximum and minimum values of the horizontal point cloud are found,and then the difference is calculated;For the characteristic parameters of leaf area and crown area,the delaurot triangulation algorithm is used to transform the point cloud from independent points into a surface graph which can be used to calculate leaf area,and then the programming calculation is carried out by using Python and VTK library;According to the characteristic parameters of plant volume,firstly,the point cloud data of the plant is spliced,and then the convex hull algorithm is used to calculate the volume;According to the characteristic parameters of plant height,firstly,the depth map colors at different heights were obtained by height calibration using standard plate,then the depth map of tomato samples at different heights was obtained,and the sample height was calculated by comparing the highest point of the sample with the depth map of standard plate;Finally,a multiple regression model was established between the growth characteristics of tomato and the nutrient content of crop.Based on the obtained terahertz spectral image data,the moving average algorithm,iterative adaptive weighted penalty least square algorithm and multiple scattering correction algorithm are used to preprocess the data,and then the principal component analysis is used to reduce the dimension of the data.The first three principal components representing most of the information of the data are found for image analysis and the multiple regression model is established.(3)Considering the limitation of single detection method of crop nutrition nondestructive testing,the data measured by two detection devices are substituted into multiple regression algorithm and BP neural network algorithm,and then the correlation coefficients under different model algorithms are obtained respectively.The correlation coefficients under different concentrations in different periods obtained by multiple regression algorithm are between 0.903 and 0.931,The correlation coefficients obtained by BP neural network are between 0.917 and 0.940,while when using the same data for calculation,the correlation coefficients obtained by BP neural network are better than the multivariate regression algorithm.(4)Due to the high price of the above experimental equipment,a portable detection platform suitable for greenhouse with a variety of nutrition and growth detection equipment was built,and the main functions and components of the platform were introduced.更多还原