深度学习的轮胎带束层X光图像三维重建  

3D Reconstruction of X-ray Image of Tire Belt Based on Deep Learning

作　　者：闫明明 陈亮 YAN Mingming;CHEN Liang(Shenyang Ligong University,Shenyang 110159,China)

机构地区：[1]沈阳理工大学自动化与电气工程学院,沈阳110159

出　　处：《沈阳理工大学学报》2022年第1期20-25,共6页Journal of Shenyang Ligong University

基　　金：辽宁省教育厅基本科研项目(LG201707);辽宁省自然科学基金项目(20170540788);国家重点研发计划(2017YFC0821001,2017YFC0821004)。

摘　　要：针对轮胎缺陷检测中轮胎带束层区域的特殊结构导致该区域缺陷的识别效果较差问题,对轮胎带束层区域的X光图片进行3D重建来拆解其结构。使用循环一致生成对抗网络(CycleGAN)对带束层区域的图片进行3D重建,利用已知钢丝的固定结构作为先验知识来控制生成图像钢丝的角度,生成不同结构的钢丝图片,达到拆解带束层复杂结构的目的,通过对损失函数的修改及参数的选取提高重建效果。对自制数据集进行实验表明,不同倾斜角度、不同清晰程度的钢丝图片均有良好的重建效果。Aiming at the problem of poor recognition of tire belt area defects due to the special structure of the tire belt area in tire defect detection,3D reconstruction of the X-ray image of the tire belt area is carried out to disassemble its structure.In this paper,CycleGAN is used to perform 3D reconstruction of the picture of the belt region.The fixed structure of the known steel wire is used as a priori knowledge to control the angle of the generated image steel wire.The purpose of dismantling the complex structure of the belt layer is achieved by generating steel wire pictures of different structures,and the reconstruction effect is improved by modifying the loss function and selecting the parameters.Experiments on the self-made data set show that it has a good reconstruction effect for steel wire pictures with different inclination angles and different degrees of clarity.

关 键 词：三维重建 循环一致生成对抗网络 轮胎检测 轮胎带束层 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]