An Improved Swimming Pool Drowning Detection Method Based on YOLOv8

To better cope with the increasing number of drowning accidents every year, an improved drowning detection method based on YOLOv8 is proposed in this paper. In this approach, the main network structure of YOLOv8 is retained, and the coordinate attention(CA) mechanism and FReLU activation function are added to the model to improve the detection effect. By adding the coordinate attention mechanism, the model obtains a better perception of direction and position on each channel compared to the conventional spatial attention or channel attention mechanisms, and FReLU achieves better spatial perception of the activation function with negligible spatial overhead compared to SiLU in the feature-enhanced network of YOLOv8. Tested on the constructed swimmer dataset, the method in this paper performs better and achieves a mean average precision (MAP) of 91.78%, which is a 1.63% improvement compared to the original YOLOv8.