Researchers Propose Meta-Learning Framework for Data-Efficient Plant Growth Estimation

A scientific preprint published on arXiv introduces a data-efficient few-shot regression framework designed for greenhouse plant growth estimation. The method tackles the high cost and time required to collect labeled agricultural images.

The framework integrates Vision Transformer (ViT) feature embeddings, fuzzy c-means clustering on unlabeled image pools for task construction, and gradient-based meta-learning. Researchers demonstrated that constructing structured tasks directly within the embedding space serves as a primary driver of overall performance.

Evaluations on two plant datasets showed that second-order meta-learning algorithms, such as MAML++ variants, significantly outperformed traditional baselines under severe label scarcity, enabling reliable greenhouse monitoring.