基于改进YOLOv8-seg的浆纱过程经轴纱辊图像分割研究OA

In order to achieve accurate and rapid segmentation of winding rolls,which vary in size during the unwinding process of warp yarns in the sizing process,an improved instance segmentation model based on YOLOv8-seg was proposed.By introducing a mask boundary loss function,the precision of edge segmentation for the winding rolls was enhanced.The feature fusion module was replaced by BIC(Bidirectional Information Conveying)module to strengthen the multi-scale feature capturing ability,and EffectiveSE attention mechanism was added to reinforce the feature map representation capability,thereby the model's segmentation accuracy of winding roll boundaries and its feature extraction capability were improved.The experimental results showed that mAP@0.5 and mAP@0.5∶0.95 of the improved YOLOv8s-seg segmentation model were reached 98.4%and 97.3%,respectively,which were 2.6 percentage points and 3.1 percentage points higher than the original YOLOv8s-seg model and the effectiveness of the model was verified.It is believed that the improved YOLOv8s-seg model can effectively adapt to images of winding rolls with different diameters,positions and quantity,providing a solid technical foundation for subsequent applications such as breakage detection of winding rolls.