Realtime Animal Intrusion Detection System Using YOLOv5

Abstract

Farmers working land near forests or open grazing tracts deal with a problem that rarely makes it into agricultural planning discussions: animals wandering into their fields at night and destroying weeks of work in minutes. Fences get pushed over, scarecrows get ignored, and a farmer cannot realistically stay awake every night watching the boundary. This paper walks through a system built to take over that watching job. It pairs a small camera and motion sensor sitting at the field edge with a compact deep learning model that can tell, in under two seconds, whether something moving near the crop is actually an animal worth worrying about. Once it decides yes, it fires off a text message or app alert with a photo attached, so the farmer knows exactly what showed up and where. The hardware is built around a Raspberry Pi, a PIR sensor that keeps the unit asleep until something moves, a YOLO-based detector trained on wild boar, nilgai, deer, and stray cattle, and a GSM or Wi-Fi link for sending the alert itself. Testing across a labelled image set and a two-week trial run at a field boundary produced detection accuracy above ninety percent for the animals it was trained on, with the whole detect-and-alert cycle finishing in under two seconds when Wi-Fi was available. The paper lays out how the pieces fit together, why certain design trade-offs were made to keep cost and power draw low, and what would need to change to scale this up more cameras, solar charging, and eventually some kind of automated deterrent that fires before the farmer even reaches the field.