Article recently published in Remote Sensing It answers the crucial question about the correctness and accuracy of 3D plant measurements. By combining experimental measurements with theoretical models of artificial plant models, Professor Evgeny Gladelin and colleagues at the Leibniz Institute for Plant Genetics and Crop Plant Research (IPK) have proposed solutions for reliable quantitative interpretation and analysis of 3D plant measurements.
In the past 30 years, methods for measuring 3D surface data from plants have been continuously developed.
Access to the plant structure makes it possible to determine the parameters of plant canopies, individual plants and organs, as well as to trace the geometric development of the plant.
Since 3D measurement is non-destructive, it is possible to observe the plant for a long time.
To distinguish between plant movement and actual growth at the level of plants and organs, 3D imaging is vital.
Plant phenotyping aims to bridge the gap between agronomic traits, plant function and genomics. As a result, 3D measurement tools are a good choice as they allow accurate measurements of geometry and growth.
Three-dimensional plant measurement methods
Several 3D scanning and engineering reconstruction techniques have been used to objectively assess the 3D plant organization. Some of these techniques are listed below:
- laser scanning
- Stereo vision (SV) or structure of motion (SfM)
- Regulated light (SL)
- Sculpt space/volume
- Flight time (ToF)
- Optical field scanning (LF) and ground laser scanning (TLS)
- range (lidar)
While the above 3D scanning methods have had varying degrees of success, the accuracy of 3D-scanned plant models and how they affect plant phenotypic properties has rarely been explored.
Professor Gladelin’s group used four plant architectures to develop an extensive virtual laser scanning architecture of 3D artificial plants. They intended to answer the following basic questions:
- To what extent do 3D plant traits change when 3D plant reconstructions become inaccurate over time, as simulated by different sources of geometric noise?
- By integrating the imprecision of 3D plant features with the results of computer simulations of composite plants, is it possible to produce a consistent quantitative description of plant morphology and physiology?
artificial plants simulation
the study Using the open source modeling tool, GroIMP v1.6, for modeling and simulation.
Artificial 3D models of the imaging architectures of tomato, maize, cucumber and Arabidopsis plants were created and used in virtual laser scanning studies.
The four composite factories presented different challenges to reach the accuracy of the 3D image.
Complex plant morphology, interlacing effects and self-shading of leaves were simulated due to larger physical size, distinctive 2D vertical geometry, and circular, flat, nearly 2D shape with large overhanging leaves.
A virtual laser scanner was used to simulate a 3D factory scan. The physical ray tracing model that GroIMP uses to simulate light is the basis of the laser scanner.
Several standard scanning techniques, including spherical and cylindrical scanning, are available with the virtual laser scanner.
The method of operation has been changed so that point cloud models of scanned objects can be generated without regard to the optical properties of objects or the color of the laser beam, unlike the usual calculation methods.
Calculation of the light absorption of the 3D structure was used as an additional indicator of the virtual laser scanner simulation to measure how “complete” the synthetic structure was with respect to the original plant structure.
Complexity, size, number of objects, light rays used, and their maximum bounce depth directly affect the computing time of the light simulation.
In this work, three perturbation cases were studied. Each scenario begins with three ways in which plant parts are lost. These events can happen randomly, from outside in, and from inside out.
Laser scanning analysis of the initial and slim 3D plant models was performed respectively to determine the quantitative properties of the 3D plant architecture.
Results and expectations
Experimental results showed that different phenotypic characteristics of the whole plant architecture generally show varying responses to progressive model perturbation.
As a result, some metrics, for example, plant height, appear independently of the percentage of lost surface area and are reasonably robust to 3D scanning errors.
According to the results, the phenotypic characteristics of the plant tend to be more or less related to the level of geometric errors in determining the three-dimensional structure of plants.
Integrative traits, such as plant area, volume, and physiologically important light absorption, have a stronger correlation with effectively visible plant area than linear growth traits, such as total plant height and width.
By integrating the measurement results with computer simulations of artificial plant models, Professor Gladion’s team addressed a fundamental problem related to the validity and accuracy of 3D plant measurements.
Their findings provide recommendations for consistent quantitative analysis and interpretation of inaccurate data.
More research is needed to expand on the results of the feasibility study and to make the analysis of synthetic model results useful for estimating actual 3D scans of plant structures.
Henk, Michael and Evgeny Gladlin. (2022) A virtual laser scanning approach to assessing the effect of geometric inaccuracies on 3D plant traits. Remote Sensing 14, no. 19: 4727. https://www.mdpi.com/2072-4292/14/19/4727