ARTIFICIAL INTELLIGENCE AND COMPUTER VISION FOR AUTOMATED IDENTIFICATION, POPULATION MONITORING, AND EARLY WARNING OF AGRICULTURAL INSECT PESTS

Abstract

Global agricultural production loses over 40% of its potential yield annually to biotic stressors, primarily insect pests, threatening food security and supply chain stability. Traditional pest management, reliant on manual field scouting and calendar-based chemical spraying, is labor-intensive, costly, and prone to observer bias, sampling errors, and reporting delays. This review synthesizes recent advances in Artificial Intelligence (AI) and Computer Vision (CV) for automated pest identification, population monitoring, and early warning systems. We analyze four benchmark datasets—IP102, AgriPest, Pest24, and GrainPest—highlighting domain-specific challenges including fine-grained morphological distinctions, multi-instar variation, long-tailed class distributions, and micro-target detection below 15×15 pixels. We evaluate deep learning architectures spanning handcrafted feature extractors, Convolutional Neural Networks (CNNs), Vision Transformers (ViTs), and one-stage real-time detectors. Refined ResNet-50 pipelines achieve 76.0% Top-1 accuracy on IP102, while specialized YOLO variants such as EP-YOLO, AEP-PEST, and AgriPest-YOLO reach mAP50 scores of 70.5–75.2% on dense micro-target benchmarks. We examine cyber-physical smart traps integrating edge computing hardware, anti-occlusion mechanical clearing systems, and low-power LoRaWAN telemetry for autonomous field operation. Furthermore, we assess microclimate-driven predictive models combining Growing Degree-Day phenological calculations with XGBoost and LSTM architectures, where five-day rolling input windows elevate emergence prediction accuracy to 86.3% while reducing false-positive rates to 11%. Key challenges include visual distribution shifts under field conditions, thermal constraints on edge processors, and extreme class imbalances affecting rare species detection. We conclude by outlining strategic pathways toward scalable, proactive Integrated Pest Management deployment.