Proceedings of the 3rd International Conference on Smart and Innovative Agriculture (ICoSIA 2022)
School of Science
Greater knowledge of wheat crop phenology and growth and improvements in measurement are beneficial to wheat agronomy and productivity. This is constrained by a lack of public plant datasets. Collecting plant data is expensive and time consuming and methods to augment this with synthetic data could address this issue. This paper describes a cost-effective and accurate Synthetic Wheat dataset which has been created by a novel L-system, based on technological advances in cameras and deep learning. The dataset images have been automatically created, categorised, masked and labelled, and used to successfully train a synthetic neural network. This network has been shown to accurately recognise wheat in pasture images taken from the Global Wheat dataset, which provides for the ongoing interest in the phenotyping of wheat characteristics around the world. The proven Mask R-CNN and Detectron2 frameworks have been used, and the created network is based on the public COCO format. The research question is “How can L-system knowledge be used to create an accurate synthetic wheat dataset and to make cost-effective wheat crop measurements?”.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License