摘要
Abstract
Dust particles adhere to the workpiece surface in industrial environments,which leads to a decline in dimensional measurement accuracy.In view of this,the paper proposes a backlight industrial image deburring method based on segmentation masks.To enhance the ability to capture image detail information and restore overall structure while avoiding the excessive smoothing caused by traditional morphological methods,a global-local feature extraction module(GLFEM)is designed as the core of the feature fusion module(FFM).Selective attention partial convolution(SAPC)and integrated statistical attention(ISA)mechanisms are used to capture key feature information to reduce model complexity and enhance feature representation.A mask adaptive enhancement module and an improved loss function are introduced to further enhance the removal of burrs at contour edges.Experimental results show that in the five measurement indicators for threads,the average errors for major diameter,medium diameter,minor diameter,pitch,and thread angle are 0.000 26 mm,0.004 92 mm,0.005 96 mm,0.000 11 mm,and 0.073°,respectively.In comparison with the existing deep learning methods,the proposed method demonstrates significant advantages in dimensional measurement accuracy.Moreover,the proposed method achieves not only the precise dimensional measurement,but also a balance between real-time performance and accuracy,with parameter count and computational complexity comparable to existing models,making it suitable for deployment in resource-constrained industrial environments.关键词
工业图像/边缘去毛刺/图像复原/注意力机制/部分卷积/图像掩码/轻量化模型Key words
industrial image/edge deburring/image restoration/attention mechanism/partial convolution/image mask/lightweight model分类
信息技术与安全科学
