Abstract :

The Growing Severity And Frequency Of Wildfires, Largely Driven By Climate Change, Highlight The Urgent Need For Intelligent And Proactive Monitoring Systems. Traditional Approaches, Including Manual Surveillance And Satellite-based Observation, Are Constrained By Delayed Response Times, High Operational Costs, And Limited Temporal Resolution. A Critical Limitation In Current Automated Detection Systems Is The High Rate Of False Alarms. While Modern Convolutional Neural Networks (CNNs) Such As YOLO (You Only Look Once) Enable Fast, Real-time Detection, They Often Lack Contextual Awareness, Leading To Misclassification Of Visually Similar Elements Like Sunsets, Artificial Lighting, Or Coloured Objects As Fire. Such Inaccuracies Result In Inefficient Emergency Responses And Increased Alert Fatigue. To Overcome These Challenges, This Work Presents VLM-FireNet, A Hybrid Cascade Framework That Combines The Speed Of Edge-based Detection With The Contextual Reasoning Capabilities Of Advanced Multimodal Models. In The Proposed System, YOLOv8 Is Deployed At The Edge To Perform Rapid Initial Detection With Inference Times Below 50 Milliseconds. These Detections Are Subsequently Verified Using A Transformer-based Vision-Language Model (VLM), Which Leverages A Global Self-attention Mechanism To Analyse The Broader Scene Context And Filter Out False Positives Effectively. The System Is Implemented Within A Multithreaded Python Environment, Integrating A Tkinter-based Graphical Interface With A Telegram Bot API For Real-time Remote Alerting. The Core Contribution Of This Research Is Its Dual-validation Strategy, Which Enhances Detection Accuracy Without Compromising Speed. Experimental Evaluation Shows That The Proposed Approach Reduces False Positives By Approximately 20% While Maintaining Real-time Performance. This Hybrid Methodology Offers A Scalable And Efficient Solution For AI-enabled IoT (AIoT) Applications In Wildfire Detection And Disaster Management.

Published:

10-4-2026

Issue:

Vol. 26 No. 4 (2026)

Page Nos:

2180-2190

Section:

Articles

License:

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

How to Cite