基于高分辨率网络的全钢子午线轮胎钢丝圈区域边界分割  

Regional Boundary Segmentation of Bead Rings in All-steel Radial Tires Based on High Resolution Network

作　　者：李喃 刘华 郝晋一[1,2,3] 夏英杰 李金屏[1,2,3] LI Nan;LIU Hua;HAO Jinyi;XIA Yingjie;LI Jinping(School of Information Science and Engineering,University of Jinan,Jinan 250022,Shandong,China;Shandong Provincial Key Laboratory of Network Based Intelligent Computing,University of Jinan,Jinan 250022,Shandong,China;Shandong Provincial University Key Laboratory of Information Processing and Cognitive Computing,University of Jinan,Jinan 250022,Shandong,China)

机构地区：[1]济南大学信息科学与工程学院,山东济南250022 [2]济南大学山东省网络环境智能计算技术重点实验室,山东济南250022 [3]济南大学信息处理与认知计算山东省高校重点实验室,山东济南250022

出　　处：《济南大学学报（自然科学版）》2024年第6期778-786,共9页Journal of University of Jinan(Science and Technology)

基　　金：山东省重点研发计划项目(2017CXGC0810);山东省教育科学“十三五”规划教育招生考试专项课题项目(BYZK201917)。

摘　　要：针对全钢子午线轮胎中纹理复杂、结构多变的钢丝圈区域边界分割困难的问题,提出一种基于高分辨率网络的轮胎钢丝圈区域边界分割方法;根据垂直投影曲线信息实现轮胎X射线衍射图像各区域的划分;利用直方图均衡化提高图像的明暗对比度,增强纹理信息;根据高分辨率网络输出的热图,基于自适应阈值方法进行边界分析,通过计算热图不同区域的阈值得到对应的二值图,统计热图中边界区域面积并筛除过小的部分,根据剩余区域重构热图并利用边界上下文信息填补被筛除的位置,从而得到整体边界分布均匀、精细的热图;在自建数据集上测试所提出方法的检测性能,通过消融实验探讨所提出的方法及其优化模块对最终边界分割结果的影响,并将所提出的方法与2种常用方法进行定量和定性对比。结果表明,钢丝圈区域的包布、反包边界分割准确率分别达到98.94%、97.23%,相对于2种常用方法,所提出的方法具有较强的稳健性和适用性。Aiming at the difficulty of regional boundary segmentation of bead rings in all-steel radial tires with complex texture and variable structure,a method of regional boundary segmentation of bead rings based on high resolution network was proposed.According to the information of vertical projection curves,each region of X-ray diffraction images of tires was divided.Histogram equalization was used to improve the contrast of light and dark and enhance the texture information.According to the heat map output by the high-resolution network,the boundary analysis was carried out based on the adaptive threshold method.By calculating the threshold value of different regions of the heat map,the corresponding binary map was obtained,and the boundary region area in the heat map was counted and the smaller part was screened out.The remaining areas were used to reconstruct the heat maps,and the boundary context information was incorpor ated to fill in eliminated positions to achieve the finely detailed heat maps with evenly distributed overall borders.The detection performance of the proposed method was tested on the self-built dataset, and the influence of the proposed method and its optimization module on the final boundary segmentation results was discussed through ablation experiments.The proposed method was quantitatively and qualitatively compared with two commonly used methods.The results show that the accuracy of wrapping and backwrapping boundary segmentation is 98.94 % and 97.23 % , respectively.Compared with the two common methods, the proposed method has strong robustness and applicability.

关 键 词：模式识别 区域边界分割 高分辨率网络 全钢子午线轮胎 直方图均衡化 边界优化 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]