基于语义分割的纱筒余纱量检测方法研究OACSTPCD

In order to solve the problems of complex background of circular knitting robot automatic production line and the large variation of the yarn bobbin sizes,which led to lower accuracy and lower precision of the detection algorithm,a method to detect yarn surplus of bobbin based on semantic segmentation was proposed.Firstly,on the basis of YOLOv8,the C2F module was replaced by DSSConv module to prevent feature redundancy and feature information loss.Aimed at the influence of multiple yarn bobbin sizes and background of yarn bobbin on detection effect,EMA attention mechanism was introduced to improve the ability of obtaining foreground of yarn bobbins.Finally,SQConv module was used to replace C3 module in the Neck layer.The improved group convolution is used to improve the reasoning speed of the model in the Neck layer.SENet attention mechanism was added to reduce the missing rate of the bobbin detail feature.The experiment showed that the improved model mAP@0.5:0.95 was reached 94.1%and the reasoning speed was 65.71 frames/s,which was better than the original YOLOv8 model.The average error of the algorithm was less than 2 mm.It could be used to measure the remaining yarn in different imaging distance and it could meet the actual production demand.