Technology-Based Mitigation of Human-Elephant Conflict in Sri Lanka: An Integrated IoT and AI Early Warning System

Human-elephant conflict (HEC) in Sri Lanka kills over 260 elephants and more than 100 people annually, driven by habitat loss and agricultural expansion into elephant corridors. Traditional deterrents such as electric fences and manual night watches have proved inadequate because elephants adapt to predictable barriers and human vigilance cannot be sustained around the clock. This paper proposes and documents a two-version integrated early warning system combining passive infrared motion detection, thermal imaging, AI-assisted species confirmation using a YOLOv8 convolutional neural network, GSM communication, and a community mobile application. Version 1.0 provides an off-grid, SMS-based sensor network deployable at USD 400 to 500 per unit. Version 2.0 adds conditional thermal imaging and on-device AI classification achieving greater than 94% species-confirmation accuracy, reducing false alerts and extending hardware lifespan, at USD 460 to 800 per unit. The paper includes a field deployment layout, hardware component breakdown, cost comparison against electric fencing and GPS collar programmes, and a community training protocol designed for rural Sri Lankan conditions. The system is intended for pilot deployment across high-risk villages in the North Central, Eastern, and Southern provinces. All diagrams and technical specifications are included in the uploaded document.