Hybrid Multi-Model Deep Learning for Forest Fire Prediction and Real-Time Detection: FireGuard AI
TL;DR
The proposed Hybrid Multi-Fire Model unifies a fine-tuned ResNet50 for real-time fire and smoke detection from webcam images and a BiLSTM network for district-level fire-risk prediction using weather and land-condition data from 2015–2025.
Abstract
Forest fires are a major environmental hazard, and early prediction and detection are essential for reducing damage. To address the limitations of traditional forest surveillance in Andhra Pradesh, this project presents FireGuard AI, a hybrid deep learning system that combines historical weather data with live visual detection. The proposed Hybrid Multi-Fire Model unifies a fine-tuned ResNet50 for real-time fire and smoke detection from webcam images and a BiLSTM network for district-level fire-risk prediction using weather and land-condition data from 2015–2025. The system is implemented using Flask and provides risk visualization, interactive district maps, model performance metrics, and automated Twilio SMS alerts for high-risk conditions. By combining predictive weather analysis with real-time image detection, FireGuard AI provides an integrated approach to early forest-fire warning and monitoring. Keywords Forest Fire Prediction, Real-Time Fire Detection, Deep Learning, ResNet50, BiLSTM, Hybrid Model, Weather Data, Computer Vision, Andhra Pradesh, FireGuard AI, Early Warning System, Flask