UAV-Based Agricultural Canopy Segmentation from OBIA to Foundation Models: A Systematic Review of Methodological Evolution, Operational Trade-Offs, and Future Directions

UAV-based canopy segmentation has become a fundamental step in extracting biologically meaningful information from high-resolution agricultural imagery for precision agriculture. Its methodological landscape has evolved rapidly from object-based image analysis (OBIA) and conventional machine learning, to convolutional neural networks (CNNs), and subsequently to hybrid architectures and foundation models such as the Segment Anything Model (SAM). Most prior reviews emphasise methodological progress or application-specific outcomes but have not systematically evaluated the operational trade-offs that determine real-world deployability — in particular the balance among annotation demand, preprocessing complexity, computational cost, environmental robustness, and transferability across agricultural environments. This gap is the rationale for the present review.