AI-Enabled IoT Security Architecture for Real-Time Intrusion Detection in Maize Farms Using Custom You Only Look Once (YOLO) Models and GSM Alert Systems.

Maize production in resource-constrained agricultural environments is increasingly threatened by human intrusion, livestock encroachment, bird infestation, and tampering with field-deployed monitoring infrastructure, resulting in crop losses, reduced productivity, and operational inefficiencies. Conventional security approaches, including manual guarding, perimeter fencing, and motion-triggered alarm systems, lack contextual intelligence, suffer from high false-alarm rates, and provide limited real-time responsiveness. This study presents an AI-enabled Internet of Things (IoT) security architecture for real-time intrusion detection in maize farms, integrating a custom-trained You Only Look Once version 8 (YOLOv8) object detection model, lightweight face-recognition algorithms, GSM-based communication, and edge computing implemented on a Raspberry Pi 5 platform. The system autonomously performs motion sensing, image acquisition, object classification, identity verification, alert generation, and dashboard-based event monitoring without cloud dependency. Experimental evaluation through laboratory validation and controlled field simulations demonstrated strong detection capability and operational robustness. The model achieved an overall detection accuracy of 84%, human detection precision of 0.856, recall of 1.000, and a mean Average Precision (mAP@50) of 0.848, confirming reliable multi-class detection performance. F1-scores of 0.741 and 0.801 for animal and bird classes, respectively, further indicate robust generalization under dynamic agricultural conditions. The system maintained an average end-to-end response latency of approximately 2 s, satisfying real-time operational requirements. To formalize integrated system effectiveness, performance is analytically modelled as a function of detection confidence, model generalization, communication reliability, environmental robustness, intrusion risk, and cumulative latency. Designed for low-connectivity rural environments, the architecture offers a scalable, low-cost, and context-aware solution for intelligent agricultural security in developing regions.