Research Projects | Portfolio

3D root reconstruction to improve carbon sequestration in maize roots

Carbon rich soil ensures the fertile soils and the agricultural productivity of plants that sequester atmospheric carbon available as CO2 into the soil.

Root phenotyping is crucial since it provides avenues to quantify deeper root traits of important crops like maize. However, due to the opaque nature of soil, dense and highly occluded maize root system, quantifying these traits such as whorl number, distance and number of crown roots is very challenging.

We developed a 3D reconstruction method that completely digitizes the maize root architecture. Our extensions allow automated reconstruction and measurement of the inner occluded root system.

Our 3D root model reconstruction method is a first promising step towards automated quantification of highly occluded maize root system. It enable the discovery of genes associated with deeper rooting by molecular biologists and pave a promising way to increase soil carbon sequestration in crops.

Accurate high-resolution three-dimensional (3D) models are essential for a non-invasive analysis of phenotypic characteristics of plants.

We present an image-based 3D plant reconstruction system that can be achieved by using a single camera and a rotation stand.

We also deducted the optimal number of images needed for reconstructing a high-quality model. Experiments showed that an accurate 3D model of the plant was successfully could be reconstructed by our approach.

This 3D surface model reconstruction system provides a simple and accurate computational platform for non-destructive, plant phenotyping.

3D plant modeling reconstruction based on structure from motion method

Plant Image analysis module development based on PlantCV platform, in cooperation with Noah Fahlgren and Malia Gehan at Donald Danforth Plant Science Center.

High-throughput non-destructive imaging systems are generating plant images to support plant phenotype study. One of the most vital traits for identifying and evaluating phenotypes of different cultivars of the same plant species is plant growth/mass. Although projected leaf area (PLA, the counting of plant pixels from top-view images), is a good non-destructive measurement of plant growth/mass, this parameter cannot reflect the actual growth conditions, since it is unable to distinguish overlapping leaves. Therefore, an estimate of leaf count can provide more beneficial information for high throughput phenotyping experiments.

We developed a leaf counting method for analyzing Arabidopsis plant images. The main improvement of our method consists of the a mask refinement method and a marker based watershed segmentation method. As a proof of concept we achieved good performance in the analysis of images of Arabidopsis plants grown under various water limiting conditions compared to well-watered controls. Performance comparison based on a benchmark dataset also shows its effectiveness. This new method will contribute to study of abiotic stress and its effect on Arabidopsis growth and development.

All parts of the plant may potentially be subjected to attack by pests or diseases. However, plants can respond with various defense mechanisms such as fortifying cell walls, producing secondary metabolites or programmed cell death to kill the pathogen. Thereafter part of the leaf surface changes its color due to the degradation of light-harvesting pigments and subsequent necrosis, which means the plant health condition can be measure by the color changes in its leaves.

In addition, each plant at all times shows a certain color distribution, which may be the result of different physical colors in the object or caused by different object/leaf orientation, relative to light orientation and shadowing through other leaves. These visible changes can be measured by quantitative image analysis. In order to perform quantitative assessment of shifts between color classes, color classification is the best option to represent color changes.

we developed a novel color classification method to perform quantitative image analysis on the plant colors. Our main motivation is to find the dominant color clusters and its distribution in plant images.

Dominant Color Clustering for Plant Image Analysis based on PlantCV Platform

3D editing & mixed reality

Interactive 3D virtual reality application: Inserting virtual objects into 3D scene

The CG object can be placed at arbitrary position with proper orientation, shadows, especially at places where occlusion happens.

The 3D modeling software has 46 functions. The constructed Airbus plane model file can be parsed as input for embedded solver.

3D model editing and visualization of industrial car & Airbus plane model

Automatic segmentation of hairy boundary: Interactive trimap generation

Given different kind of human face as input, we propose a method to simulate the basic kinds of expression.