Abstract: Rapid, reliable detection of forest fires is critical to minimize ecological, economic, and human losses. This paper presents a practical AI-driven system for early forest-fire detection trained on the D-Fire dataset and deployed end-to-end with a Flask web service and HTML/CSS/JS front-end. The proposed pipeline detects both fire and smoke in single images, videos, and live streams, triggers configurable alerts, and exposes REST endpoints for easy integration. We report model and system design choices, training regimen, evaluation results, and field-deployment considerations. The solution supports live IP-camera feeds, user image/video upload, and dashboards for incident review. A modular microservice design enables scaling, audit logging, and integration with emergency notification channels.

Keywords: Forest fire detection, smoke detection, deep learning, YOLO, Flask, real-time inference, edge deployment, early warning.

Downloads: | DOI: 10.17148/IARJSET.2026.13119