A Multithreaded Hybrid System for Autonomous Wildfire Detection and Mobile Emergency Response

Abstract

The rapid escalation of global wildfires, exacerbated by climate change, necessitates a shift from reactive suppression to proactive, intelligence-driven intervention. Historically, wildfire monitoring relied on human lookouts or satellite-based remote sensing; however, these traditional systems suffer from high latency, significant labor costs, and low temporal resolution. The core problem definition lies in the "False Alarm" crisis: while modern Convolutional Neural Networks (CNNs) like You Only Look Once (YOLO) provide real-time speed, they lack semantic reasoning, frequently misidentifying sunsets, industrial glare, or red vehicles as fire. These limitations lead to wasted emergency resources and "alert fatigue" among responders. There is a critical need for a context-aware framework that balances the low latency of edge computing with the high-level reasoning of multimodal models. This research introduces VLM-FireNet, a unique proposed system utilizing a Hybrid Cascade Architecture. The model employs YOLOv8 at the edge for rapid initial feature extraction, achieving sub-50ms inference. High-confidence detections are then cross-verified by a Transformer-based VisionLanguage Model (VLM). By utilizing a global Self-Attention mechanism, the VLM analyzes the entire scene context to validate the hazard, effectively filtering out environmental noise that fools standard CNNs. The system is integrated through a multithreaded Python environment, bridging a local Tkinter dashboard with a remote Telegram Bot API for instantaneous mobile alerting. The significance of this study lies in its "Dual-Check" logic, which provides a scalable, cost-effective solution for smart city and forest management. By reducing false positives by an estimated 20% while maintaining real-time performance, this hybrid approach sets a new benchmark for AI-Enabled IoT (AIoT) in disaster management, ensuring that emergency responses are both swift and accurate.