AI in Agricultural Pest Management: A Critical Review
Artificial intelligence is widely promoted as a solution for automating pest monitoring and reducing crop losses, but many models fail when moved from laboratory conditions to real farms. This review critically examines AI application across three sensing platforms i.e., retrofitted sticky traps with computer vision, drones and rovers for distinguishing mite damage from drought stress and acoustic sensors for detecting wood borers hidden inside trees. This review highlights that while deep learning reliably counts whiteflies and codling moths in controlled settings, performance collapses under variable lighting, dust, dew and occlusion. The core problems are rarely discussed in the literature; for instance, training datasets are often labeled by non-specialists, leading to systematic misclassification, models suffer from domain shift over a single growing season and the black box nature of neural networks erodes trust among farmers and regulators. False positives drive unnecessary pesticide use while false negatives cause crop loss. To the best of current knowledge, AI systems generally struggle to recognize a novel invasive pest in its first year of arrival. Through case studies from China, California and Brazil, it is demonstrated that trust depends on explainability and field validation, not on benchmark accuracy. This study proposes a tiered integration into existing IPM workflows. AI-only alerts, remote human checks and field visits. Unlike previous reviews that focus on model accuracy, this review critically examines why most AI fails in the field and proposes a tiered, human-centered integration pathway.
Authors
- Surender Kumar Sharma (ORCID: https://orcid.org/0000-0002-1283-450X)
Publication Details
- Journal
- Black Sea Journal of Agriculture
- Published
- 2026-11-15
- DOI
- https://doi.org/10.47115/bsagriculture.1972876
- Primary Topic
- Smart Agriculture and AI
- Type
- article
- Field-Weighted Citation Impact
- 0.00