RepC-MVSNet: A Reparameterized Self-Supervised 3D Reconstruction Algorithm for Wheat 3D Reconstruction

The application of 3D digital models to high-throughput plant phenotypic analysis is a research hotspot nowadays. Traditional methods, such as manual measurement and laser scanning, have high costs, and multi-view, unsupervised reconstruction methods are still blank in the field of crop research. It...

Full description

Saved in:
Bibliographic Details
Published in:Agronomy (Basel) 2023-08, Vol.13 (8), p.1975
Main Authors: Liu, Hui, Xin, Cheng, Lai, Mengzhen, He, Hangfei, Wang, Yongzhao, Wang, Mantao, Li, Jun
Format: Article
Language:English
Subjects:
Agricultural production
Algorithms
Analysis
Cameras
Cereal crops
Crops
Data collection
Datasets
Deep learning
Digitization
Feature maps
Genotype & phenotype
Image reconstruction
Morphology
Motion systems
multi-view stereo
Neural networks
neural rendering
Optimization
point cloud
Research methodology
structure from motion
Three dimensional models
Visualization
Wheat
Wheat industry
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The application of 3D digital models to high-throughput plant phenotypic analysis is a research hotspot nowadays. Traditional methods, such as manual measurement and laser scanning, have high costs, and multi-view, unsupervised reconstruction methods are still blank in the field of crop research. It is challenging to obtain a high-quality 3D crop surface feature composition for 3D reconstruction. In this paper, we propose a wheat point cloud generation and 3D reconstruction method based on SfM and MVS using sequential wheat crop images. Firstly, the camera intrinsics and camera extrinsics of wheat were estimated using a structure-from-motion system with feature maps, which effectively solved the problem of camera point location design. Secondly, we proposed the ReC-MVSNet, which integrates the heavy parametric structure into the point cloud 3D reconstruction network, overcoming the difficulty of capturing complex features via the traditional MVS model. Through experiments, it was shown that this research method achieves non-invasive reconstruction of the 3D phenotypic structure of realistic objects, the accuracy of the proposed model was improved by nearly 43.3%, and the overall value was improved by nearly 14.3%, which provided a new idea for the development of virtual 3D digitization.
ISSN:2073-4395
2073-4395
DOI:10.3390/agronomy13081975