“The environment of the target hydropower station is complex and the risk of fire is high. Traditional smoke and fire detection methods face challenges such as missed detection of small targets, low remote recognition accuracy, and high false alarm rates in complex dynamic environments. The research team proposes a YOLOv8 and Transformer fusion model, introduces feature recalibration gate control mechanism to achieve dynamic complementary fusion, and effectively improves the intelligent monitoring accuracy of fireworks in hydropower stations.”