A Comprehensive Survey on Farmland Segmentation: Methodologies, Applications, and the Road to Agriculture 5.0

This survey addresses the absence of a unified review that traces the evolution of farmland segmentation from its OBIA roots to the emerging era of Vision-Language and Foundation Models. By providing a dedicated analysis of this trajectory, we bridge the gap between broad remote sensing reviews and specialized algorithmic studies. We systematically categorize the methodological progression into three distinct stages: traditional heuristic methods, deep learning paradigms (CNNs and Transformers), and the nascent frontier of large-scale foundation models. Furthermore, we critically analyze operational bottlenecks-specifically the "smallholder dilemma," label scarcity, and multimodal heterogeneity-that currently hinder deployment in complex agricultural landscapes. To overcome these hurdles, we outline a strategic research agenda centered on Agricultural Foundation Models, Physics-Informed Explainable AI , and Federated Learning. The proposed roadmap consolidates fragmented research avenues into a synergistic vision for Agriculture 5.0, positioning this survey as a foundational reference for shaping the next generation of scalable, trustworthy, and actionable farmland intelligence systems.