Abstract:
The transition to sustainable agriculture necessitates a greater understanding of plants in the field with automated plant phenotyping. Leaf segmentation is critical in t...Show MoreMetadata
Abstract:
The transition to sustainable agriculture necessitates a greater understanding of plants in the field with automated plant phenotyping. Leaf segmentation is critical in this domain, enabling the accurate assessment of plant traits essential for managing the growth of economically viable crops and plants. This paper reviews recent developments in leaf segmentation methodologies, focusing on the emerging paradigm of promptable segmentation. Promptable segmentation, exemplified by Meta AI’s Segment Anything Model (SAM), offers flexibility and versatility by allowing users to provide various prompts for segmentation tasks. However, challenges such as complex backgrounds, overlapping leaves, and computational complexity persist. The paper discusses strategies such as prompt engineering, post-processing of segmentation outputs, and fine-tuning segmentation models to address these challenges. Future research includes exploring occlusion handling techniques, adopting parameter-efficient fine-tuning methods, collecting and leveraging publicly available datasets, synthetic data generation, and embracing video-based data collection. By overcoming these challenges and harnessing the potential of promptable segmentation, researchers can create automated plant phenotyping for any plant species, leading to more accurate, efficient, and scalable solutions for agricultural sustainability.
Published in: 2024 30th International Conference on Mechatronics and Machine Vision in Practice (M2VIP)
Date of Conference: 03-05 October 2024
Date Added to IEEE Xplore: 12 November 2024
ISBN Information: