Advanced Animal Detection System with Cascaded YOLOv8 and Dynamic Feature Extraction

Human-animal conflict is a major issue in agricultural fields and forested regions, risking both human and animal lives and causing huge resource losses. Crop destruction, livestock loss, and, in severe circumstances, human losses demonstrate the gravity of the situation. With advances in IoT sensors and cloud computing, remote image processing has become more efficient and cost-effective. This study describes an advanced animal detection system that uses real-time video monitoring to identify animal invasions. The system uses a laptop camera module with Wi-Fi to collect live video, which is then transferred to a remote server for processing. For object detection, a pre-trained deep learning system based on YOLOv8 is used, with continuous training on a dataset of annotated animal photos yielding an accuracy of roughly 98%. Upon detection, the system sends out alerts in the form of SMS or alarms, notifying farmers or forest authorities based on the detected animal. In addition, a speaker-based deterrence device is used to successfully prevent animal infiltration. The proposed approach improves farm security by delivering an automated, high-accuracy solution for real-time animal detection and management.