Abstract

Unmanned aerial vehicles (UAVs) have become an increasingly important tool in precision agriculture by providing flexible high-resolution observations of crop canopies from plot to field scales. Early UAV applications mainly relied on 2D and 2.5D information, such as vegetation indices (VIs), image features, and canopy height models (CHMs), to support crop classification and growth monitoring. The development of structure-from-motion (SfM) photogrammetry and UAV light detection and ranging (lidar) further advanced UAV remote sensing from 2D analysis to 3D characterization of canopy structure. More recently, artificial intelligence (AI)-based approaches, including neural radiance fields (NeRF), 3D Gaussian splatting (3DGS), feed-forward 3D reconstruction networks, and geometry foundation models, have introduced new paradigms for high-fidelity 3D reconstruction from multiview UAV observations. When combined with process-based models (PBMs), including functional-structural plant models (FSPMs), and multitemporal UAV data, these approaches provide a pathway toward 4D (i.e., 3D space plus time) crop modeling for representing canopy growth dynamics, stress responses, and structural changes over time. In this review, we synthesize the evolution of UAV remote sensing in agriculture across 2D, 2.5D, 3D, and emerging 4D frameworks while summarizing the associated UAV platforms, sensor modalities, and data-processing workflows. We further discuss the major technical challenges that currently limit robust 4D UAV applications, including data acquisition, multisensor integration, model generalization, and coupling between AI and process-based modeling. Finally, we outline future directions for integrating multimodal UAV sensing, PBMs, and AI-driven 3D/4D reconstruction to support agricultural digital twins for crop monitoring, scenario analysis, and adaptive field management.

Link: https://ieeexplore.ieee.org/document/11653205