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

Meta-analysis of 121 UAV-based crop pest and disease detection studies (2018–2024) found 89% lacked truly independent test datasets, inflating performance estimates and limiting generalisability.

• Why it matters: Reported AI accuracy may not reflect real field performance because validation often fails to test on independent data. This gap undermines confidence in deploying models for practical pest and disease monitoring.
• What they did: The study reviewed dataset construction and model validation practices across 121 UAV-based papers from 2018–2024, focusing on independence of test sets and evaluation on independent fields.
• The result: Only 11% evaluated models on independent fields, and transferability was uncommon; the authors recommend changes to improve robustness, reproducibility, and field-level relevance.

AI summary of the abstract · may contain errors · not medical advice · read the paper.