In order to enhance the accuracy of fire detection, particularly the ability to detect small fire sources, and to increase the speed of fire detection, this paper proposes an improved fire detection algorithm based on YOLOv8. By leveraging the DualConv to improve the C2f and construct a lightweight structure, and introducing the Slou loss function for small fire sources, the system's accuracy and speed are improved. Experiments were conducted on a custom-built fire dataset, and the results show that compared to the original YOLOv8, the improved model increases the mean Average Precision (mAP@50) by 1.5% and reduces model parameters by 10.3%. This effectively lowers the false alarm rate and enhances the response speed to fires.