Submitted:
18 November 2026
Posted:
19 November 2026
You are already at the latest version
Abstract
High-throughput plant phenotyping using unmanned/uncrewed aerial vehicles (UAVs) has emerged as a powerful approach for monitoring plant growth, stress responses, and productivity across diverse agroecosystems. Recent advances in artificial intelligence (AI), particularly deep learning and foundation models, are expanding UAV-based phenotyping through automated feature extraction, multimodal integration, and transferable representations. Unlike conventional task-specific models, foundation models are pretrained on broad datasets and can potentially provide reusable representations that are adapted to multiple downstream applications with reduced annotation requirements. This review synthesizes recent developments in AI-driven UAV plant phenotyping, with particular emphasis on foundation-model approaches, including self-supervised pretraining, transfer and parameter-efficient adaptation, vision transformers, large vision-language models, promptable segmentation, generative AI, and multimodal representation learning. We examine applications in crop segmentation, stress and disease detection, canopy trait estimation, and longitudinal phenotyping, while assessing the transferability of models pretrained on satellite or natural-image datasets to centimeter-scale UAV imagery. Beyond conventional prediction, we explore how longitudinal UAV phenomics, genomic, environmental, management, and mechanistic information could be integrated within foundation-model frameworks to support climate-resilient breeding and environment-specific ideotype discovery. We discuss key challenges involving limited annotations, domain shift, multimodal integration, computational demands, interpretability, uncertainty, and equitable access. Finally, we highlight future opportunities in multimodal agricultural foundation models, self-supervised and continual learning, AI copilots, satellite-UAV-ground data fusion, edge AI, autonomous phenotyping, digital twins, and open agricultural AI ecosystems. Together, these developments point toward a transition from task-specific image analysis toward transferable crop-intelligence systems capable of supporting both precision agriculture and climate-resilient crop improvement.
Keywords:
foundation models
; UAV phenotyping
; artificial intelligence
; precision agriculture
; climate resilience
; vision transformers
; plant phenomics
; remote sensing
; deep learning
; multimodal learning
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.