A Systematic Literature Review on Deep Learning Applications in Agricultural Pest Detection and Control.

Abstract

This systematic literature review synthesizes research on deep learning for agricultural pest detection and control. Using a PRISMA-guided approach, we screened studies published between 2015 and 2025 across IEEE Xplore, ScienceDirect, Scopus, and SpringerLink. Forty-five studies met the inclusion criteria. We summarize model classes (e.g., Convectional Neural Network (CNNs), You Only Look Once (YOLO)-family detectors, transformers, and hybrids), dataset characteristics, and evaluation metrics (accuracy, precision, recall, F1, and mAP). Findings show that deep learning consistently outperforms traditional techniques for image-based pest identification; lightweight models enable edge deployment, while transfer learning and augmentation improve robustness. Key challenges are generalization across geographies, class imbalance, environmental noise, and limited interpretability. We provide practical recommendations for model selection, reporting standards, and field deployment to advance scalable, trustworthy pest management.