| Fires pose a serious threat to life and property, making early and accurate warning of critical importance. In recent years, deep learning-based object detection technologies have advanced rapidly in the field of flame and smoke recognition, gradually replacing traditional detection methods. To systematically outline the current research status and development trends in this area, this paper employs the CiteSpace tool to conduct a visual analysis of relevant literature, summarizing mainstream technical frameworks and application progress from 2019 to October 2025, including representative algorithms such as the YOLO series, DETR, Faster R-CNN, SSD series, and various lightweight network architectures. The research shows that although significant results have been achieved, notable challenges remain in areas such as small target and early fire detection, model lightweighting and edge deployment, robustness in complex environments, and data support with false alarm control. In the future, the integration of multi-modal information, the construction of efficient lightweight networks, the exploration of self-supervised and weakly-supervised learning, combined with temporal modeling and interpretability analysis, are expected to drive fire detection and early warning systems toward greater accuracy, robustness, and practicality. |
