Real-Time Object Detection in Disaster Zones and UAV Thermal-RGB Imagery

This research presents an innovative real-time disaster detection framework that leverages YOLOv11, a deep learning model, to enhance situational awareness and decision-making in emergency response operations. Unlike traditional UAV-based systems that often suffer reduced accuracy in low-visibility or complex environments, the proposed approach fuses RGB and thermal imagery from quadcopter drones with the advanced feature extraction and high-speed inference capabilities of YOLOv11. Integrated into an edge computing platform, the system supports low-latency, real-time object detection, making it highly effective for time-critical disaster scenarios. To further support operational decision-making, a multi-criteria decision-making (MCDM) module based on the Analytic Hierarchy Process (AHP) is embedded within the pipeline, enabling automated prioritization of detected threats.

The model was trained and validated on a 10,000-image multimodal dataset comprising annotated UAV data from wildfire, flood, and earthquake zones. YOLOv11 consistently outperformed baseline models such as YOLOv5, achieving 88% detection accuracy, with precision, recall, and F1-scores all exceeding 0.85, and reduced response time by 40% compared to manual inspection workflows. The integration of YOLOv11 with thermal-RGB fusion significantly improved detection robustness under smoke, haze, and debris-obscured conditions.

This study validates YOLOv11 on multimodal UAV disaster imagery with an integrated decision-support layer to improve emergency response effectiveness. The proposed framework sets a new benchmark in intelligent aerial surveillance, combining high detection accuracy with real-time processing capabilities. Designed for cost-efficiency and modular deployment, the framework supports scalability across local governments, first responders, and humanitarian organizations.

Article file ID not found.

The authors declare no conflict of interest.

Not applicable

The datasets used in this study are openly available at [repository link] and the source code is available on GitHub at [GitHub link].

This work did not receive any external funding.