Phenotyping and Skeletonization for Agricultural Robotics

Abstract

Scientific phenotyping of plants is a crucial aspect of experimental plant breeding. By accurately measuring plant characteristics, phenotyping plays a vital role in the development of new plant varieties that are better adapted to specific environments and have improved yield, quality, and resistance to stress and disease.

In addition to observing plants, robotic plant manipulation is becoming an increasingly important area of research in agriculture, as it has the potential to revolutionize farming practices. By using robots to interact with plants, farmers could eventually achieve greater precision and efficiency in tasks such as pruning, pollinating, and harvesting, leading to improved yields and reduced labor costs.

However, obtaining labeled data for the assessment of phenotype estimates or plant models can be an extremely challenging and time-consuming process in agriculture. We tackle this common problem in agricultural robotics along several avenues. First, we propose an unsupervised assessment method for reconstructed 3D sorghum clouds, which are used to count sorghum seeds for non-destructive phenotyping. Second, we use highly consistent outdoor imagery to simplify vine segmentation with low amounts of training data. Finally, we build 3D skeletal vine models intended for vine pruning, and assess these vine models using an unsupervised approach from previous work.

These skeletal vine models are then used in a case study in which we predict pruning weight in grapevines, one of the factors in optimizing grape quality and yield. Our results show that our approach outperforms previous methods in predicting pruning weight, demonstrating the potential for our method to improve agricultural practice.

Overall, our work highlights the benefits of 3D plant models for phenotyping and manipulation in agriculture, and presents a new approach to assessing reconstructed point clouds. Our findings have implications for the development of more efficient and effective agricultural practices, with the potential to play a role in simplifying sorghum breeding and grapevine pruning efforts through automation.