李明¹，张万腾¹，泮玮婷¹，朱俊科²，宋绪斌¹，王春颖^1,3，刘平^1,3*

(1. 农业装备智能化山东省工程研究中心，山东省设施园艺智慧生产技术装备重点实验室(筹)，山东农业大学机械与电子工程学院，泰安271018，山东，中国；

2. 山东理工大学农业工程与食品科学学院，淄博255100，山东，中国；

3.小麦育种全国重点实验室，泰安271018，山东，中国)

摘要：准确且快速的小麦形态重建与表型解析对于品种选择、科学栽培及精准管理至关重要。单一视角受环境障碍限制，难以实现小麦植株的高通量表型数据解析。因此，该研究提出一种基于多视角的三维点云快速重建方法，以实现小麦表型的高通量精准鉴定。

首先，以小麦分蘖期植株为试验对象，基于RealSense传感器构建多视角采集系统，并从16个视角获取小麦点云数据。其次，通过优化联合光度与几何目标，基于colored-ICP算法实现空间位置配准。此外，采用MVS算法，将粗配准获得的深度图像、RGB图像与空间位置相结合，实现多视角点云的精配准，与SFM-MVS算法相比，该方法的平均重建时间为33.82 s，显著快于SFM-MVS算法。

此外，基于小麦植株的三维点云数据，本研究计算了小麦植株高度、叶片长度、叶片宽度、叶片面积及叶片角度。试验结果显示，计算值与测量值之间的决定系数分别为0.996、0.958、0.956、0.984和0.849。最后，该研究分析并鉴定了不同小麦品种的表型信息，如紧凑度、凸包体积和平均叶面积，证明了所提出的方法能够捕捉品种和个体之间的表型差异。研究中提出的方法可为量化小麦表型性状提供一种快速途径，有助于育种、科学栽培和环境管理。

关键词：小麦植株；RealSense传感器；MVS算法；三维点云；表型特征

DOI: 10.25165/j.ijabe.20251804.9169

引用信息:  Li M, Zhang W T, Pan W T, Zhu J K, Song X B, Wang C Y, et al. Three-dimensional reconstruction and phenotypic identification of the wheat plant using RealSense D455 sensor. Int J Agric & Biol Eng, 2025; 18(4): 254–265.

点击阅读全文：Three-dimensional reconstruction and phenotypic identification of the wheat plant using RealSense D455 sensor

Three-dimensional reconstruction and phenotypic identification of the wheat plant using RealSense D455 sensor

Ming Li¹, Wanteng Zhang¹, Weiting Pan¹, Junke Zhu², Xubin Song¹, Chunying Wang^1,3, Ping Liu^1,3*

(1. Shandong Engineering Research Center of Agricultural Equipment Intelligentization, Shandong Key Laboratory of Intelligent Production Technology and Equipment for Facility Horticulture, College of Mechanical and Electronic Engineering, Shandong Agricultural University, Tai’an 271018, Shandong, China;

2. School of Agricultural Engineering and Food Science, Shandong University of Technology, Zibo 255100, Shandong, China;

3. State Key Laboratory of Wheat Improvement, Shandong Agricultural University, Tai’an 271018, Shandong, China)

Abstract: Accurate and rapid wheat morphology reconstruction and trait collection are essential for selecting varieties, scientific cultivation, and precise management. A single perspective is limited by environmental  obstructions, hindering the collection of high-throughput phenotype data for wheat plants. Therefore, a rapid reconstruction method of multi-view three-dimensional point cloud is proposed to realize the high-throughput and accurate identification of wheat phenotype. Firstly, taking wheat at the tillering stage as the experimental object, a multi-view acquisition system based on a RealSense sensor was constructed, and the point cloud data of wheat were obtained from 16 views. Secondly, a joint photometric and geometric objective was optimized, and space location was registered by colored Point Cloud Registration (colored) and Iterative Closest Point (ICP) algorithms. Furthermore, the Multiple View Stereo (MVS) algorithm was used to combine the depth image, RGB image, and spatial position obtained by coarse registration to enable the fine registration of multi-viewpoint clouds. Compared with the traditional Structure From Motion (SFM)-MVS algorithm, our proposed method is much faster,  with  an  average reconstruction time of 33.82 s. Moreover, the wheat plant height, leaf length, leaf width, leaf area, and leaf angle of wheat were calculated  based on the three-dimensional point cloud of the wheat plant. The experimental results showed that the determination coefficients of the method are 0.996, 0.958, 0.956, 0.984, and 0.849, respectively. Finally, phenotypic information such as compact degree, convex hull volume, and average leaf area of different wheat varieties was analyzed and identified, proving that the method could capture the phenotypic differences between varieties and individuals. The proposed method provides a rapid approach to quantify wheat phenotypic traits, aiding breeding, scientific cultivation, and environmental management.

Keywords: wheat plant, RealSense sensor, MVS, three-dimensional point cloud, phenotypic traits

DOI: 10.25165/j.ijabe.20251804.9169

Citation:  Li M, Zhang W T, Pan W T, Zhu J K, Song X B, Wang C Y, et al. Three-dimensional reconstruction and phenotypic identification of the wheat plant using RealSense D455 sensor. Int J Agric & Biol Eng, 2025; 18(4): 254–265.

往期推荐：

四川大学赵龙副教授等：基于无人机遥感数据估算苹果园土壤水分含量：LST/LAI两阶段特征空间理论的应用

中国农业大学王新副教授等：基于改进YOLOv5s的复杂环境下轻量级烟叶检测方法

东南大学赵立业教授等：便携式土壤硝态氮测定装置的研制

重磅！南京农大发布：国内首个通用农业开源大模型“司农”

支持单位：