Abstract:Conventional fire detection approaches typically relied on distinct models to address the varying characteristics of fires and smoke, particularly under different day and night conditions. Additionally, the distinction between wildfires and urban fires, which is influenced by camera shooting distances, often requires the use of separate detection algorithms. To address this, we introduce a novel domain-free (day, night, urban, and forest) fire detection algorithm within YOLOv5, incorporating linear attention for spatial attention extraction and gated temporary pooling (GTP) for temporal attention extraction. This study utilizes YOLOv5 as a base framework, and deviates from the conventional approach of modifying only pre-processing and downstream tasks. Within the dynamic attention block of GTP, our method extracts fire-related features by effectively discerning fires from background, while considering the spatiotemporal characteristics of fire flames and smoke. Despite the compact model size, our proposed approach significantly outperforms state-of-the-art methods for still image and continuous video datasets, demonstrating the effectiveness of our approach. This performance holds true for various fire types, locations, and under day and night conditions. This study marks a significant advancement in domain-free fire detection, offering a unified solution capable of addressing the diverse challenges presented by different fire scenarios and lighting conditions.

Domain-free fire detection using the spatial–temporal attention transform of the YOLO backbone