Real-Time Wild Animal Detection in Farmland Using Camera-Based Deep Learning and an Android Alert System

Abstract

Wild animals entering farms is a problem. This happens because their homes are getting smaller. Animals like leopards, elephants and wild boars damage crops. Dig up fields. This causes losses for farmers. This paper presents a real-time detection and alert system that uses a camera to identify animals and immediately notifies the farmer through a mobile app. The dataset contains two classes domestic (human, cat, dog) and wild (leopard, elephant, wild boar) with 3,000 original images (500 per class). After augmentation using Roboflow [21], the final dataset totals 7,446 images split into training (6,669 / 90%), validation $(\mathbf{4 4 3} / \mathbf{6} \boldsymbol{\%})$, and test $(\mathbf{3 3 4} / \mathbf{4} \boldsymbol{\%})$ sets. Three sizes of YOLO26 were trained and tested on the same test set. The smallest model, YOLO26n, performed best with mAP50 = 0.8469 and 66.1 FPS on a Tesla T4 GPU, outperforming YOLO26s (0.8399) and YOLO26m (0.772). The confusion matrix shows the model correctly identifies 77% of wild animals, which is the most critical class for farm safety. The mobile app is built with React Native [18] and Expo [12], connected to Firebase. Farmers log in once and receive live DANGER alerts when a wild animal is detected, along with detection history and daily, weekly, and monthly charts. The total time from camera capture to phone notification is under 3 seconds.