Abstract:In order to realize plant morphological measurement and physiological diagnosis, a multi-modal three-dimensional (3D) reconstruction method was proposed. This reconstruction method further laid the foundation for plant phenotypic measurement. Due to the complexity of 3D geometric morphologies, only two or two-and-a-half-dimensional images of greenhouse plants can be captured at a single angle of view (AOV) by the imaging sensors. However, 3D point cloud reconstruction of plants requires images captured at multiple AOVs. In addition, the 3D geometric morphologies of plants undergo significant changes during the full-growth-cycle and to acquire suitable 3D plant images, it is necessary to frequently adjust the sensor position. Therefore, sensor position and AOV directly affect the plant phenotyping efficiency. Developing an efficient and accurate multi-view 3D point cloud reconstruction method that meets the need for full-growth-cycle, high-throughput 3D reconstruction and phenotyping of greenhouse plants is therefore pivotal to the development of high-throughput plant phenotyping techniques. So a multi-modal three-dimensional reconstruction method of greenhouse tomato plants under different measurement positions and angles was proposed, and to solve the problem of multi-spectral reflectance mapping and multi-view point cloud 3D reconstruction, multi-spectral reflectance images were registered to RGB-D image coordinate system by phase-correlation method, and a multi-view RGB-D image 3D reconstruction method based on self-calibration of the Kinect sensor was established which realized the reconstruction of RGB 3D point cloud model and multi-spectral reflectance 3D point cloud model of the plants. The two-dimensional multi-spectral image registration quality was evaluated objectively by the normalizing gray-scale similarity coefficient D, the spectral overlap rate in the region of interest (ROI) C, and the mutual information value and the Hausdorff distance HD was applied to objectively evaluate the reconstruction accuracy of the three-dimensional point cloud reconstruction of the plant. In total, 30 greenhouse tomato plants were used in this study with each plant reconstructed from four angles of view at angle intervals of 90 degrees. The obtained results showed that the average values of C and D were 0.920 6 and 0.908 5, respectively. After registration, the mutual information value increased by 9.81 % and the canopy multi-spectral images could be registered accurately to the depth coordinate system. The ratio of the HD distance set of reconstruction point cloud less than 0.6 cm was 78.39 %, the ratio of less than 1.0 cm was 91.13%, and the mean value of the tomato distance ensemble HDavg was 0.37 cm, depicting that the tomato plant 3D point cloud model had high reconstruction accuracy and could be applied to multi-modal 3D reconstruction of greenhouse tomato plants. This research integrated the traditional 3D geometric morphology measurement system and plant physiological information diagnosis system, and as such the external morphology and internal physiological information of the plants could be measured in the same imaging room. It provides a precise and efficient measurement method for high-throughput plant phenotypic measurement, and is of great significance to the development of modern intelligent facility horticultural management and plant phenomics.

NeRF-based 3D reconstruction pipeline for acquisition and analysis of tomato crop morphology

Tomato-Nerf: Advancing Tomato Model Reconstruction with Improved Neural Radiance Fields

Three-Dimensional Phenotyping Pipeline of Potted Plants Based on Neural Radiation Fields and Path Segmentation

High-fidelity 3D Reconstruction of Plants using Neural Radiance Field

3Dphenomvs: A Low-Cost 3D Tomato Phenotyping Pipeline Using 3D Reconstruction Point Cloud Based on Multiview Images

Exploring Accurate 3D Phenotyping in Greenhouse through Neural Radiance Fields

Look how they have grown: Non-destructive Leaf Detection and Size Estimation of Tomato Plants for 3D Growth Monitoring

High-Resolution 3D Crop Reconstruction and Automatic Analysis of Phenotyping Index Using Machine Learning

A 3D Surface Reconstruction Pipeline for Plant Phenotyping

PanicleNeRF: low-cost, high-precision in-field phenotypingof rice panicles with smartphone

A Calculation Method of Phenotypic Traits Based on Three-Dimensional Reconstruction of Tomato Canopy

In-field High Throughput Phenotyping and Cotton Plant Growth Analysis Using LiDAR

A comparative study of plant phenotyping workflows based on three-dimensional reconstruction from multi-view images

AgriNeRF: Neural Radiance Fields for Agriculture in Challenging Lighting Conditions

Pheno4D: A spatio-temporal dataset of maize and tomato plant point clouds for phenotyping and advanced plant analysis

Multi-modal three-dimensional reconstruction of greenhouse tomato plants based on phase-correlation method

Evaluating Neural Radiance Fields (NeRFs) for 3D Plant Geometry Reconstruction in Field Conditions

A Novel LiDAR-Based Instrument for High-Throughput, 3D Measurement of Morphological Traits in Maize and Sorghum

Generating high-quality 3DMPCs by adaptive data acquisition and NeREF-based radiometric calibration with UGV plant phenotyping system

Three-dimensional reconstruction and phenotype measurement of maize seedlings based on multi-view image sequences

Evaluating Neural Radiance Fields for 3D Plant Geometry Reconstruction in Field Conditions