Answers are grounded in this paper and cite their sources.
Accurate phenological stage classification using unmanned aerial vehicle (UAV) imagery is essential for smart crop monitoring and harvesting management, yet reliable interannual transfer remains challenging due to variations in environmental conditions and background interference, particularly for creeping plants with low canopy height. In this study, we propose a novel framework, termed Vegetation Prior-Guided Domain Adaptation (VPGDA), to improve interannual phenological stage classification of the creeping aromatic plant Thymus mongolicus based on UAV RGB imagery. The framework integrates a vegetation-aware attention mechanism derived from the Excess Green (ExG) index with adversarial domain adaptation to enhance phenology-related feature representation while reducing background influence and domain discrepancy. Experiments conducted on UAV datasets collected over the 2024 and 2025 growing seasons demonstrate that the proposed framework significantly improves year-to-year transfer performance, increasing overall accuracy from 59% to 79% and Macro-F1 score from 0.50 to 0.79 compared with direct model transfer. In addition, independent cross-site validation and controlled illumination perturbation tests further confirm the spatial generalization and robust under moderate illumination variations of the proposed approach without requiring retraining. These results suggest that VPGDA offers an effective solution for UAV-based phenological monitoring under interannual domain shifts and can support intelligent agricultural applications such as harvest-window determination and crop management.
No takes yet. Share an insight, caveat, or question.
Zheng et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: