Research on people recognition in ship fire smoke based on YOLOv5 algorithm with added attention mechanism

This electronic document proposes a video-based method for identifying people in ship fire smoke for the scene of people in ship fire smoke. By collecting a large number of images of ship fire and crew escape, a fire smoke personnel dataset was produced for training. Considering the limitations of the YOLOv5 algorithm that is prone to false detection and missed detection of occluded targets. This paper proposes an improved YOLOv5 algorithm, which improves the accuracy and classification of the model by adding the shuffle attention mechanism to the group feature channel structure of the model precision. Then the ship fire smoke personnel recognition model was obtained through dataset training and comparative experiments were designed. The results show that the ship fire smoke personnel recognition model based on the YOLOv5-ShuffleAttention algorithm proposed in this paper has higher recognition performance, and can quickly and accurately identify ship fire smoke and Trapped people.