Research on Fire Smoke Detection Algorithm Based on YOLOv8

The SDE-YOLOv8 fire detection model is proposed to solve the problems of low accuracy and slow speed in the detection of fire smoke.Based on YOLOv8, SimAM attention mechanism is introduced in this model, and the C2f layer in the network structure is improved to obtain D-C2f module, which enhances the feature expression ability of the model in complex scenes and inhibits the interference of irrelevant features.Secondly, in order to improve the speed of model detection, a lightweight backbone network EfficientViT was used for feature extraction, which reduced the computing overhead during detection. The experimental results show that compared with the original YOLOv8 on DFS data set, SDE-YOLOv8's average accuracy mAP50 is increased by 2.2%, and the detection speed is increased by 14.3%, achieving a good improvement in accuracy and speed.