Research on Forest Fire Detection Model Based on Transformer and YOLOv5

Addressing challenges such as the destructive power of wildfires and difficulties in timely detection, governments have begun adopting camera-based monitoring for wildfire detection with the advancement of deep learning technology. Existing YOLOv5-based wildfire detection models suffer from insufficient accuracy and robustness in complex scenarios. This study enhances the YOLOv5 model algorithm by incorporating a lightweight Transformer attention module to replace certain convolutional layers. This approach aggregates multi-modal wildfire features across scales and employs data augmentation to simulate complex interferences. While controlling computational costs, the model gains enhanced adaptability to variable lighting conditions and terrain occlusions in mountainous areas, thereby improving detection accuracy and robustness in complex scenarios.