Edge AI and IoT-Integrated Wildfire Real-Time Monitoring System

Increasing wildfire frequency and severity, driven by climate change and urban expansion, demands rapid and reliable detection in remote and bandwidth-constrained environments. Existing AI-based wildfire detection systems face multiple challenges: high false positive rates from vision-only models, latency from cloud-based inference, poor performance in network-limited scenarios, power inefficiency, and vulnerability to transient/noisy sensor readings. To address these, we present an edge-deployed wildfire detection framework that integrates a lightweight YOLOv8n model on GPU-enabled devices with multi-modal IoT sensor fusion and temporal smoothing. This architecture eliminates cloud dependency for detection, significantly reduces false positives, and enables real-time operation on battery-powered UAV platforms. In field tests, the proposed system achieved an F1-score of $93.3 \%$, mAP@0.5 of $93.8 \%$, less than 1 false positive per hour, and inference latency under 50 ms while operating at $5-10 \mathrm{~W}$. These results demonstrate that our proposed system is a low-power, false-positive–resilient, real-time wildfire detection solution capable of maintaining reliable performance in challenging operational environments.