AI-based UAV pest and disease detection: Time for a reset?

Remote sensing using uncrewed aerial vehicles (UAVs) and AI, particularly machine learning and deep learning, is increasingly applied to crop pest and disease detection. However, the real-world robustness of these models remains uncertain. We conducted a meta-analysis of 121 UAV-based studies published between 2018 and 2024, examining dataset construction and model validation practices. We found that 89% of studies lacked truly independent test datasets, resulting in inflated performance estimates and limited generalisability. Only 11% evaluated models on independent fields, and successful transferability was uncommon. Our analysis identifies key methodological limitations underlying this issue and provides recommendations to improve robustness, reproducibility, and practical relevance. Overall, current validation practices require substantial improvement to ensure reported model performance reflects field-level applicability.