<![CDATA[Wildfires represent a critical threat to ecosystems, human safety, and economic stability, emphasizing the necessity for rapid and reliable detection mechanisms. Traditional approaches such as satellite monitoring and manual surveillance are often hindered by latency, limited spatial resolution, and environmental constraints, thereby underscoring the value of automated and intelligent solutions. This study presents DeepFire, a real-time wildfire detection framework developed using the YOLOv8 architecture. Data preprocessing involved normalization, removal of irrelevant objects, and extensive data augmentation to enhance generalization and mitigate potential overfitting. The dataset encompassed diverse environmental conditions, including varying smoke intensities, vegetation densities, and viewing perspectives. Experimental evaluation demonstrated outstanding performance, achieving a mean Average Precision (mAP@0.5) of 0.995, precision and recall values of 0.995, and an F1-score of 1.00 at the optimal confidence threshold for detection. The mAP@0.5 metric was selected for its suitability in assessing localization accuracy under real-time constraints, whereas mAP@0.5:0.95 is discussed in the main text for comprehensive benchmarking. Qualitative assessments further verified the model’s robustness in accurately classifying a wide range of fire and non-fire scenarios. Future research will focus on enhancing dataset diversity, improving deployment efficiency, and validating system performance under real-world conditions.]]>
Copyrights © 2026
