Enhancing Plant Phenotyping Accuracy in Sorghum through Analysing Aerial Data

The school of EECS is hosting the following Progress Review 1 seminar:

Enhancing Plant Phenotyping Accuracy in Sorghum through Analysing Aerial Data

Speaker: Rukshan Karannagoda
Host: Dr. Alina Bialkowski 

Abstract:

The task of phenotyping genetically diverse plant lines is a crucial component in plant breeding programs. With the ever-increasing food demand of the modern world, it has become critical for plant breeders and scientists alike to develop more efficient and accurate plant phenotyping methods, in order to selectively breed plants with high-yield and higher stress tolerance and manage systems of agriculture. The efficient and accurate phenotyping of sorghum plants via the analysis of crop images holds significant potential for advancing plant science and breeding.

Yet, as traditional phenotyping methods require a large expenditure of time and labour, computer vision and machine learning (ML) as well as recent deep learning (DL) algorithms and the automatic analysis of aerial imagery have been instrumental in enhancing the accuracy and efficiency of plant phenotyping, diminishing the need for human input. As these methods are only as accurate as the training data itself, the effects of phenotyping a phenotypically and genotypically diverse a crop as sorghum has been limited largely due to the diversity of the sorghum germplasm not being captured by existing datasets and methods.

This thesis advances a framework for capturing the diversity in sorghum genotypes as well as the diversity in environment, to increase the performance of sorghum phenotyping methods from a computer vision perspective.  It first situates the research within the history of computer vision-based plant phenotyping and motivates the shift toward domain invariant phenotyping methods. It then studies and addresses the need for enhancement of phenotyping accuracy by introducing new datasets and algorithmic enhancements for sorghum organs via detection, segmentation, classification, and domain generalization tasks. These contributions enable more precise and comprehensive plant trait identification, which leads to new insights into plant growth and development and ultimately contribute to the development of more resilient and productive crop varieties.

Biography:

Mr Rukshan Karannagoda is a PhD candidate from the Data Science group at the School of Electrical Engineering and Computer Science, the University of Queensland (UQ), Australia.  He received his Bachelor’s Honours degree in Information Technology from the University of Moratuwa, Sri Lanka. His research focuses on Computer Vision and Domain Generalization for Agriculture, under the supervision of Assistant Professor Mahsa Baktashmotlagh, Dr. Yadan Luo, and Professor Scott Chapman.