Improved fire detection algorithm based on YOLOV8s

To enhance the accuracy and speed of fire detection in complex scenarios while reducing false positives and missed detections, this paper constructs a lightweight model by replacing the backbone network of YOLOv8 with a RepViT network.Structural reparameterization technology is introduced to optimize training and inference efficiency. The loss function is improved with EIoU to mitigate instability in aspect ratio parameterization. Additionally,the Focal Loss function is incorporated into the loss function to enhance the model's robustness in complex environments.Experimental results on the dataset demonstrate that the improved algorithm achieves a 6.84%increase in precision,a 1.25%improvement in mAP@0.5,a 1.14%higher recall rate, a 50-frame increase in FPS, and a 40%reduction in parameter count compared to the original algorithm. Consequently,the enhanced model is better suited for real-time monitoring applications.