结合实例分割和双目结构光的钢丝绳分割与尺寸测量OA

To effectively measure the dimensions of steel wire ropes with complex geometries and textures,a segmentation algorithm combining binocular structured light and an improved YOLOv8-seg is proposed.First,for the complex surface characteristics of steel wire ropes,dynamic snake convolution(DSConv)is introduced to construct a specialized feature extraction module,namely C2f-DSC.Then,a BoTNet(bottleneck transformer)module,which combines the advantages of convolutional neural networks(CNNs)and Transformers,is introduced into the Neck.Finally,SIoU is adopted as the bounding box regression loss function.Experimental results show that,compared with the traditional YOLOv8-seg algorithm,the improved algorithm achieves a 7.4%increase in mAP@0.5(seg)and a 15.1%increase in mAP@0.5∶0.95(seg),effectively improving the segmentation accuracy.In the measurement module,binocular structured light is used for dimensional measurement within a range of 2 m,and the diameter error of the steel wire rope is controlled within 0-5%of its nominal diameter,meeting industry standards.