The article discusses the development of a new DT/MARS-CycleGAN model to improve object detection in robotic crop phenotyping. This is crucial in assessing and quantifying phenotypic traits related to crop growth, yield, and adaptation to environmental stresses. The model uses image augmentation to better detect crops from complex and variable backgrounds. The generated synthesized crop images closely mimic real images in terms of realism and are used to fine-tune object detectors such as YOLOv8. The new framework significantly boosts the performance of the MARS crop object/row detector, contributing to the field of robotic crop phenotyping.
Publication date: 20 Oct 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2310.12787
