Answers are grounded in this paper and cite their sources.
Cross-domain shifts, such as variations across regions and wheat varieties, significantly limit the generalization of UAV-based Fusarium head blight (FHB) detection models. To address this, we propose a novel cross-domain detection approach that integrates a newly developed Earth Mover’s Distance (EMD)-based histogram matching method with adversarial domain adaptation. The proposed EMD histogram matching generates high-quality images that preserve source-domain labels while aligning visual features with the target domain, effectively reducing reliance on annotated target data. In addition, an adversarial domain-adaptive detection framework based on YOLO11, equipped with a gradient reversal-driven domain classifier, enforces domain-invariant feature extraction for robust cross-domain detection. Experiments demonstrate that the proposed method outperforms conventional and existing domain-adaptive approaches, achieving superior cross-domain detection performance for wheat FHB. The mean average precision (mAP) on the three target domains reached 63.3 %, 77.2 %, and 81.7 %, respectively, corresponding to improvements of 7.4 %, 5.6 %, and 6.4 % over the YOLO11 baseline.
No takes yet. Share an insight, caveat, or question.
Yin et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: