基于U-Net的工件轮廓提取方法的研究  

Research on Workpiece Contour Extraction Method Based on U-Net

作　　者：郭凯旋 王吉芳[1] 刘相权[1] 王凯 GUO Kaixuan;WANG Jifang;LIU Xiangquan;WANG Kai(College of Mechanical and Electrical Engineering,Beijing Information Science&Technology University,Beijing 100192,China)

机构地区：[1]北京信息科技大学机电工程学院,北京100192

出　　处：《组合机床与自动化加工技术》2025年第4期8-12,共5页Modular Machine Tool & Automatic Manufacturing Technique

摘　　要：工件轮廓提取是实现工件识别和定位的重要前提,为解决传统机器视觉算法对复杂环境中的工件轮廓提取不完整、易受干扰的问题,提出了一种改进U-Net的工件轮廓分割模型,将VGG16网络应用于解码器部分,增加网络特征提取能力;将ECA注意力机制引入每个跳跃连接层中,提高了工件轮廓特征在模型中的权重;在编码器末端引入ASPP空洞空间卷积池化金字塔模块,以获取高层特征图中不同尺度的特征信息,进而提高目标的分割精度。试验结果表明,EVA-UNet模型在交并比、召回率、精准率和综合性能F1分数4个方面表现良好,对工件轮廓提取能力优于其他经典模型,能够为实现复杂环境下工件轮廓提取提供良好的解决方案。The contour extraction is an important prerequisite for the identification and positioning of a workpiece.To solve the problem that traditional machine vision algorithms extract incomplete and vulnerable artifact profiles in complex environments.The EVA-UNet model for object contour segmentation is proposed.The VGG16 network is incorporated into the decoder section to enhance the capabilities of the features extraction.The ECA(efficient channel attention)attention mechanism is introduced into each skip connection layer to increase the weight of workpiece contour features in the model.At the end of the encoder,the ASPP(atrous spatial pyramid pooling)dilated spatial convolutional pooling pyramid module is introduced to obtain feature information at different scales in high-level feature maps,thereby improving the segmentation precision of the target.Experimental results demonstrate that the EVA-UNet model performs well in terms of intersection over Iou,Recall,Precision,and F1_score,and is better than other classical models,and can provide a good solution for workpiece contours extracted in complex environments.

关 键 词：注意力机制模块 轮廓提取 语义分割 工件轮廓 

分 类 号：TH164[机械工程—机械制造及自动化] TG659[金属学及工艺—金属切削加工及机床]