FireGuard-YOLOv11: Efficient Deep Vision Architecture for Early Indoor Fire Hazard Detection
Year: 2026
Type: conference-paper
Abstract: Fire and smoke inside buildings pose a serious threat to human life and infrastructure, due to confined spaces, cooking accidents, electrical faults, improper storage of flammable materials, smoking, and human negligence. Conventional sensor-based fire detection systems often produce false alarms and delayed responses due to small environmental changes, reduced sensor sensitivity, and the influence of airflow, among other factors. To overcome these limitations, this study proposes a deep learning-based real-time indoor fire and smoke detection framework based on the advanced YOLOv11 object detection model, which supports computationally efficient real-time detection suitable for edge-oriented applications. The Indoor Fire Smoke Dataset was utilized, which comprises 5000 images, including 2500 fire images and 2500 smoke images, collected from indoor environments. Proper data partitioning and preprocessing techniques have been applied to reduce model bias and ensure reliable results. The proposed model shows a notable improvement of 2.4% in precision, 2.9% in recall, 2.5% in mAP50, 4.5% in mAP50-95, and 2.7% in F1-score compared to the existing model. Furthermore, a comparative performance analysis of improved YOLOv11 against other versions was conducted, including YOLOv9, YOLOv10, and YOLOv12. The K-fold cross-validation was applied to reduce bias, prevent overfitting, and improve generalization. In addition, the proposed study was compared with existing studies, which demonstrated improved accuracy and faster detection performance.
Language: English
Subfield: Safety, Risk, Reliability and Quality
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7204451082