基于改进YOLOv8的零件图纸裁剪

Part Drawing Cropping Based on an Improved YOLOv8

摘要: 零件图纸不仅代表着零件的制造要求，同时也是原创性设计的体现，图纸比对可以很好地保护其原创性。图纸裁剪是图纸比对技术的先行军，通过裁剪可以高效率地进行信息提取和比对。为了解决传统分割方法在零件图纸分割领域中的不足，本文提出了一种改进的YOLOv8的零件图纸裁剪方法，引入了CBAM注意力机制，能够高效地将图纸图像裁剪为标题栏区域、技术要求区域、其余粗糙度区域和图元区域，并利用轮廓检测方法将图元区域进一步裁剪为若干视图。实验结果表明本文的改进模型比传统的区域生长方法裁剪速度更快，比原始的模型的mAP和召回率分别上升了5.5%和7.6%。

Abstract: Part drawings not only represent the manufacturing requirements of parts, but also the embodiment of original design, and drawing comparison can well protect the originality. Drawing cutting is the pioneer of drawing comparison technology, and information extraction and comparison can be carried out efficiently through cutting. In order to solve the shortcomings of the traditional segmentation method in the field of part drawing segmentation, this paper proposes an improved YOLOv8 part drawing cropping method, which introduces the CBAM attention mechanism, which can efficiently crop the drawing image into the title block area, the technical requirement area, the rest roughness area and the element area, and further crop the element area into several views by using the contour detection method. Experimental results show that the improved model in this paper has a faster cropping speed than the traditional regiometric growth method, and the mAP and recall rates are increased by 5.5% and 7.6%, respectively, compared with the original model.