Dynamic monitoring of canopy structure by time serial photography of early stage growth of wheat

Abstract

Early vigour is an important physiological trait to improve establishment, water-use efficiency and grain yield for wheat. Phenotyping large numbers of genotypes is challenging due to the fast growth and development of wheat seedlings. Several technologies have been well developed to obtain 3D spatial structures of crop canopies, e.g. LiDAR, time-of-flight laser and ultrasonic sensing. The technologies, however, are expensive with limited operation environments. Multi-view images can be used to reconstruct 3D canopies given recent advances in photogrammetry information computing. Combined with pixel analysis of RGB images or high spatial resolution broad-band imagery, crop phenotypes can be extracted from images.

Imaged based phenotyping has already been used to analyse the complete structure of plants with wider leaves, e.g. estimation of leaf length, width and stem height. The algorithm is mature for broad leaves to reconstruct 3D point cloud and leaf surface to estimate the phenotypic parameters related to growth. But wheat has slender and narrow leaves with few identifiable features, and it is challenging to reconstruct wheat canopies from images. Here we developed a new photo-based method to dynamically monitor growth and development of wheat canopy of two wheat genotypes with contrasting expression of early vigour.

In this paper, the multi-view images were taken using a ‘vegetation stress’ camera at two day intervals from emergence to 6th leaf stage for single and stand plants in a glasshouse experiment. Point clouds were extracted using VisualSFM which is based on the Multi-View Stereo and Structure From Motion (MVS-SFM) algorithm, and segmented into individual organs using a segmentation submodel based on Octree method in CloudCompare. The leaf midribs were fitted using a local polynomial function . Finally, phenotypic parameters were calculated from the reconstructed point cloud, and included tiller and leaf number, plant height, Haun index, phyllochron, leaf length, angle, and growth rate . Significant contrasts of phenotype parameters were observed for the two genotypes and were consistent and highly correlated with manual observations. The final growth stage at which the method works depends on the complexity of tillering when plant organs become occluded. Here the method works until about 6 main stem leaves are emerged.

The workflow presented in this study provides an efficient method to reconstruct the dense point cloud without damaging and phenotype individual plants using a low-cost camera. It could be applied in high throughput phenotyping for applications in both physiology and breeding. The rapidity and accuracy of this novel method can characterise the results of specific selection criteria that have been or can now be utilised to breeding program.