作物根系构型三维探测与重建方法研究进展_温维亮

中国农业科学 2015,48(3):436-448 Scientia Agricultura Sinica
doi: 10.3864/j.issn.0578-1752.2015.03.04
作Байду номын сангаас根系构型三维探测与重建方法研究进展
温维亮
1,2
，郭新宇 ，赵春江 ，王传宇 ，肖伯祥
2
2
2
2
（1 北京工业大学计算机学院，北京 100124；2 北京农业信息技术研究中心，北京 100097）
摘要：根系是作物获取水分和养分的重要器官，由于土壤的观测阻碍，根系三维形态的认知与表达成为作物 根系深入研究的瓶颈。三维数字化、可视化是研究和认知作物形态结构的重要方法，研究具有表征根系长相长势 及土壤中水分、养分等物质对作物根系构型的影响具有重要意义。本文从三维角度，综述了近年来作物根系构型 探测手段、三维重构与可视化方面的研究进展。首先从破坏性探测与原位探测两方面介绍了近年来根系构型三维 探测的方法。破坏性探测主要包括直接挖掘法、土块保护挖掘清洗法和平板扫描图像分析法，破坏性探测方法在 获取全局或局部根系拓扑结构与平面几何构型参数方面具有较大优势；原位探测方法主要包括土壤中安置观察装 置法、地面穿透雷达法、特殊培养环境法、CCD 相机法、三维数字化方法及穿透射线成像法等，作物根系的原位 探测保持了根系构型的空间分布信息，但大部分方法仅能针对作物生长初期或可控生长环境下的作物根系开展数 据获取。由于作物根系探测数据大多以局部二维图像形式存在，文章综述了基于二维图像的作物根系平面几何构 型解析的相关内容，包括基于二维图像的根系识别与参数提取的算法与相关软件。分析表明，目前作物根系数据 获取仍存在： （1）数据获取费时费力； （2）方法局限性大； （3）数据完整性低； （4）各种方法所获取数据融合应 用度低等问题。在根系探测的基础上，从三维建模与生长建模两方面介绍了作物根系三维建模与可视化方面的相 关工作。其中，根系三维建模包括了基于模拟算法的几何建模和基于原位探测的三维重建两部分，基于模拟算法 的几何建模是在人们对作物根系认识的基础上，结合计算机模拟算法，构建与实际根系具有形态相似性的根系三 维几何模型；与之相比，基于原位探测的三维重建更能真实地反映作物根系的实际形态，其主要包括 XCT、三维 数字化等方法。最后，文章展望了数据缺失条件下的作物根系三维重建研究，认为在目前技术手段前提下，已可 实现根系拓扑结构三维解析，但根系的空间分布重建难度较大，尤其是在大田环境下作物根系原位测量的前提下； 此外，目前作物根系三维数据主要存在着数据缺失、各种数据各为己用等问题，认为有必要将小子样理论与数据 融合相关方法引入到作物根系三维重建研究，实现缺失数据条件下有效利用多种数据获取手段的作物根系三维重 建。 关键词：作物；根系构型；探测方法；三维重建；可视化
Crop Roots Configuration and Visualization: A Review
WEN Wei-liang1,2, GUO Xin-yu2, ZHAO Chun-jiang2, WANG Chuan-yu2, XIAO Bo-xiang2
(1 College of Computer Science, Beijing University of Technology, Beijing 100124; 2 Beijing Research Center for Information Technology in Agriculture, Beijing 100097)
收稿日期：2014-01-27；接受日期：2014-10-28 基金项目：国家“863”计划（2013AA102404） 、国家科技支撑计划（2012BAD35B01） 联系方式：温维亮，Tel：010-51503362；E-mail：wenwl@。通信作者赵春江，Tel：010-51503411；E-mail：zhaocj@
Abstract: As an important organ of absorbing water and nutrients for crops, root system architecture (RSA) has become a bottleneck in-depth study of crop roots due to the observing obstacles of soil. Three dimensional and visualization are important methods for studying and recognizing the morphological and structural traits of plants. In the aspect of crop root phenotype, these methods have a significance for characterizing the growth appearance of roots and for the influence of water, nutrients and other substances in the soil on RSA. In this paper, from the three dimensional perspective, the research progresses in detecting methods,
3期
温维亮等：作物根系构型三维探测与重建方法研究进展
437
three dimensional reconstruction and visualization of RSA in recent years were reviewed. Destructive and in situ detection methods of RSA were introduced. Destructive detecting methods mainly include (i) direct mining, (ii) protective soil block mining and cleaning, (iii) root scanning and image analysis, etc. These destructive methods have great advantages in obtaining global or local topology and extracting the plane geometric parameters. Meanwhile, in situ method mainly include six classes: (i) installation of root observation windows in field, (ii) ground penetrating radar, (iii) cultivation in controllable environment, (iv) CCD camera imaging, (v) 3D digitizing and (vi) X-ray computed tomography (CT), etc. These in situ methods retained the distribution information of RSA in space. However, most of the in situ detecting methods are only available for young or grown in controllable environment roots. Because a great deal of the detected data of RSA exist in the form of two-dimensional image, studies on extraction and analysis of plane geometry RSA parameters from two dimensional images were introduced as well, including root identification and parameter extraction algorithms and manual extraction software. Analysis showed that, the crop root data acquisition still exist some problems, such as time-consuming and laborious, redundant limitations, low integrity, difficult to integrate of each other, etc. Therefore, detecting and analyzing morphological traits of roots will still be the focus of RSA research. On the basis of RSA detecting technology, 3D modeling and visualization methods of RSA were reviewed on the aspects of 3D static modeling and dynamic growth simulation. 3D static modeling of RSA includes two kinds of methods: 3D modeling based on simulating algorithms and 3D reconstruction based on in situ detecting. Combining the computer simulation algorithms and understanding of RSA, the 3D modeling methods using simulating algorithms construct virtual roots with high morphological similarity of the real one. In contrast, 3D reconstruction based on in situ detection, which mainly includes XCT and 3D digital methods, could reflect the actual morphological structure of the RSA. Finally, the paper prospected the research on 3D reconstruction of crop roots under the condition of data missing. It was considered that 3D measurement and analysis of root topology could be realized using current technical means, while the 3D reconstruction of spatial distribution of RSA was difficult, especially in the premise of in situ measurement of crop roots in field. In addition, current 3D root data mainly have the data missing and low integration problems, so it is necessary to introduce the statistical methods like small sample theory and data fusion method into the research of 3D reconstruction of crop root to achieve a better reconstruction strategy by using multiple RSA data effectively in the premise of data missing. Key words: crop; root system architecture; detecting method; 3D reconstruction; visualization