Solar-Powered Motion-Triggered Edge AI Monitoring System for Small-Animal Surveillance in Farm Environments

Distributed farm environments require monitoring systems that can operate with limited power, intermittent connectivity, and minimal human supervision. Small-animal intrusion is particularly difficult to detect because targets are small, fast-moving, and frequently occluded by vegetation, soil structures, or low-light conditions. This paper presents a demonstration-scale solar-powered monitoring system that integrates photovoltaic energy harvesting, lithium-ion storage, passive infrared (PIR) motion triggering, camera activation, YOLOv8n edge detection, event logging, and deterrence using floodlight and speaker modules. The system is designed as an event-driven pipeline in which image capture and visual inference are activated only after a motion event, reducing unnecessary computation relative to continuous video analysis. A custom five-class dataset covering rat, mouse, squirrel, chipmunk, and mole was used to fine-tune the prototype detector. Under constrained proof-of-concept conditions, the five-class model achieved an [email protected] of 0.81, an [email protected]:0.95 of 0.54, a mean retained-detection confidence of approximately 0.81, and about 22 FPS on a Raspberry Pi 4B. To provide a more stable assessment of the visual detection component, a supplementary single-class rat-detection experiment was conducted using 4,127 annotated images, achieving 0.8555 precision, 0.9042 recall, 0.9331 [email protected], and 0.7475 [email protected]:0.95 on the test set. A first-order power-budget analysis estimated a daily demand of 47.07 Wh, a 2.12x solar safety margin using a 20 W panel under five peak-sun-hours, and approximately 4.08 days of autonomy with a 12 V, 20 Ah battery at 80% usable depth of discharge. The results support the feasibility of combining solar-assisted operation, motion-gated sensing, lightweight edge inference, and autonomous deterrence for low-cost farm monitoring. However, larger field-collected multi-class datasets, controlled Raspberry Pi benchmarking, and long-duration outdoor energy measurements are required before operational deployment claims can be made.