Revolutionizing crop breeding: multi-source data fusion enhances phenotype accuracy in field phenotyping platforms

In the face of global food security challenges, the urgency for advanced crop breeding techniques intensifies.High-throughput phenotyping (HTP) platforms, using technologies like RGB cameras and LiDAR, promise detailed 3D insights into plant populations. However, accurately aligning and extracting traits from complex 3D point cloud data remains a significant challenge. This difficulty limits the full exploitation of HTP platforms, indicating a pressing need for research into more effective data mining methods for multi-source, high-dimensional time-series phenotype data.

In April 2023, Plant Phenomics published a perspective article entitled by "Multi-Source Data Fusion Improves Time-Series Phenotype Accuracy under Field Phenotyping Platform".

This study collected high-throughput time-series data from field maize populations via a phenotyping platform equipped with LiDAR and RGB camera technologies. The results revealed that a multi-source data fusion approach markedly increased the precision of phenotypic trait extraction. Plant heights measured using this method closely matched manual measurements (R² = 0.98), outperforming single-source data (R² = 0.93). This demonstrates the platform's effectiveness in capturing detailed phenotypic traits. The study also highlighted the practicality of rail-based field phenotyping platforms for observing plant growth dynamics on individual plant and organ scales. Further results involved the use of a powerful computing setup for efficient image and point cloud processing. The fusion of time-series images with point cloud data led to precise point cloud alignment and effective ground point removal. Segmenting the maize point cloud at various growth stages allowed for detailed analysis of plant and organ structure, essential for phenotypic parameter extraction and understanding growth dynamics. The study has also showen that point cloud data alignment is crucial for accurate phenotype extraction. The integration of image and point cloud data improved the segmentation accuracy, reflected in the increased R² values for plant height, leaf azimuth, and leaf inclination angle. This multi-source data fusion approach proved to be superior in enhancing time-series phenotype estimation accuracy.

In summary, this study successfully demonstrated that the integration of various data sources and advanced algorithms in a high-throughput phenotyping platform can significantly improve the accuracy and efficiency of phenotypic trait extraction in field maize populations. This methodological approach harbors extensive potential across plant phenomics, yielding invaluable insights for contemporary crop breeding and research endeavors.

###

References

Authors

Yinglun  Li^1,2†, Weiliang  Wen^1,2†, Jiangchuan  Fan^1,2, Wenbo  Gou^1,2, Shenghao  Gu^1,2, Xianju  Lu^1,2, Zetao  Yu², Xiaodong  Wang²,  and Xinyu  Guo^1,2*

†These authors contributed equally to this work.

Affiliations

¹Information Technology Research Center, Beijing Academy of Agriculture and Forestry Sciences, Beijing 100097, China.

²Beijing Key Lab of Digital Plant, National Engineering Research Center for Information Technology in Agriculture, Beijing 100097, China.

About Xinyu Guo

He is a professor at the Information Technology Research Center of Beijing Academy of Agriculture and Forestry Sciences.