本文主要研究内容
作者王珺璞(2019)在《基于层次化特征及改进RPCA的织物疵点检测算法研究》一文中研究指出:织物疵点检测是纺织品生产制造过程中的关键环节,直接决定着纺织品的质量和价值。对于纹理复杂多样、疵点形态各异的织物图像来说,传统模式识别方法存在着检测率不高、适应性不强的问题。鲁棒主成分分析(Robust principal component analysis,RPCA),又称低秩分解模型能将图像分解为目标和背景,可用于织物疵点检测问题。然而,基于低秩分解模型的检测方法的性能依赖于图像的有效表征及模型的构建和求解。因此,本文结合织物图像特性,对图像表征和RPCA模型展开研究,提出了基于层次化特征和改进RPCA的织物疵点检测算法。研究成果如下:1)提出了基于特征融合和TV-RPCA的织物疵点检测算法。首先,根据一阶梯度特征和二阶梯度特征互补的特性,通过典型相关分析(Canonical Correlation Analysis,CCA)对二者进行融合,以提高图像表征能力;然后,构建基于全变差正则项的低秩分解模型(total variation-RPCA,TV-RPCA),在有效地分离出织物疵点的同时,可以在一定程度上消除织物图像中的部分噪声;最后,将优化求解出的稀疏矩阵根据空间对应关系得到疵点显著图,并进行阈值分割处理。实验结果表明,所提算法优于利用单一特征及传统RPCA的检测效果。2)提出了基于深度特征和NTV-RPCA的织物疵点检测算法。首先,通过卷积神经网络提取出多层次深度特征,以实现对织物图像有效表征;然后,构建基于非凸全变差正则项的低秩分解模型(non-convex total variation-RPCA,NTV-RPCA),不仅可以有效地检测出含有较少噪声的疵点显著图,而且非凸优化可以提高求解精度;最后,将由稀疏矩阵生成的多个疵点显著图选择性进行融合,并通过阈值分割得到最终的疵点分割图。实验结果表明,所提算法进一步提高了疵点的检测效果。3)提出了基于深度-低阶特征和NTV-NRPCA的织物疵点检测算法。首先,通过融合新型卷积神经网络提取出的高阶语义特征和一些低阶对比度信息,提高了图像的表征能力;然后,构建基于非凸全变差正则项的非凸低秩分解模型(non-convex total variation-non-convex RPCA,NTV-NRPCA),在有效地检测出含有较少噪声的疵点显著图的同时,可以进一步地提高求解精度;最后,将稀疏矩阵对应的疵点显著图通过一种阈值分割算法,得到疵点分割图以定位疵点位置。实验表明,所提算法在降低计算复杂度的前提下,仍具有很高的检测精度。本文研究成果可用于简单纹理或复杂纹理的织物图像疵点检测,提高了现有检测方法的自适应性及检测精度,相关算法可推广到纸张、铝箔或钢材等工业产品表面的缺陷检测,具有广泛的应用前景。
Abstract
zhi wu ci dian jian ce shi fang zhi pin sheng chan zhi zao guo cheng zhong de guan jian huan jie ,zhi jie jue ding zhao fang zhi pin de zhi liang he jia zhi 。dui yu wen li fu za duo yang 、ci dian xing tai ge yi de zhi wu tu xiang lai shui ,chuan tong mo shi shi bie fang fa cun zai zhao jian ce lv bu gao 、kuo ying xing bu jiang de wen ti 。lu bang zhu cheng fen fen xi (Robust principal component analysis,RPCA),you chen di zhi fen jie mo xing neng jiang tu xiang fen jie wei mu biao he bei jing ,ke yong yu zhi wu ci dian jian ce wen ti 。ran er ,ji yu di zhi fen jie mo xing de jian ce fang fa de xing neng yi lai yu tu xiang de you xiao biao zheng ji mo xing de gou jian he qiu jie 。yin ci ,ben wen jie ge zhi wu tu xiang te xing ,dui tu xiang biao zheng he RPCAmo xing zhan kai yan jiu ,di chu le ji yu ceng ci hua te zheng he gai jin RPCAde zhi wu ci dian jian ce suan fa 。yan jiu cheng guo ru xia :1)di chu le ji yu te zheng rong ge he TV-RPCAde zhi wu ci dian jian ce suan fa 。shou xian ,gen ju yi jie ti du te zheng he er jie ti du te zheng hu bu de te xing ,tong guo dian xing xiang guan fen xi (Canonical Correlation Analysis,CCA)dui er zhe jin hang rong ge ,yi di gao tu xiang biao zheng neng li ;ran hou ,gou jian ji yu quan bian cha zheng ze xiang de di zhi fen jie mo xing (total variation-RPCA,TV-RPCA),zai you xiao de fen li chu zhi wu ci dian de tong shi ,ke yi zai yi ding cheng du shang xiao chu zhi wu tu xiang zhong de bu fen zao sheng ;zui hou ,jiang you hua qiu jie chu de xi shu ju zhen gen ju kong jian dui ying guan ji de dao ci dian xian zhe tu ,bing jin hang yu zhi fen ge chu li 。shi yan jie guo biao ming ,suo di suan fa you yu li yong chan yi te zheng ji chuan tong RPCAde jian ce xiao guo 。2)di chu le ji yu shen du te zheng he NTV-RPCAde zhi wu ci dian jian ce suan fa 。shou xian ,tong guo juan ji shen jing wang lao di qu chu duo ceng ci shen du te zheng ,yi shi xian dui zhi wu tu xiang you xiao biao zheng ;ran hou ,gou jian ji yu fei tu quan bian cha zheng ze xiang de di zhi fen jie mo xing (non-convex total variation-RPCA,NTV-RPCA),bu jin ke yi you xiao de jian ce chu han you jiao shao zao sheng de ci dian xian zhe tu ,er ju fei tu you hua ke yi di gao qiu jie jing du ;zui hou ,jiang you xi shu ju zhen sheng cheng de duo ge ci dian xian zhe tu shua ze xing jin hang rong ge ,bing tong guo yu zhi fen ge de dao zui zhong de ci dian fen ge tu 。shi yan jie guo biao ming ,suo di suan fa jin yi bu di gao le ci dian de jian ce xiao guo 。3)di chu le ji yu shen du -di jie te zheng he NTV-NRPCAde zhi wu ci dian jian ce suan fa 。shou xian ,tong guo rong ge xin xing juan ji shen jing wang lao di qu chu de gao jie yu yi te zheng he yi xie di jie dui bi du xin xi ,di gao le tu xiang de biao zheng neng li ;ran hou ,gou jian ji yu fei tu quan bian cha zheng ze xiang de fei tu di zhi fen jie mo xing (non-convex total variation-non-convex RPCA,NTV-NRPCA),zai you xiao de jian ce chu han you jiao shao zao sheng de ci dian xian zhe tu de tong shi ,ke yi jin yi bu de di gao qiu jie jing du ;zui hou ,jiang xi shu ju zhen dui ying de ci dian xian zhe tu tong guo yi chong yu zhi fen ge suan fa ,de dao ci dian fen ge tu yi ding wei ci dian wei zhi 。shi yan biao ming ,suo di suan fa zai jiang di ji suan fu za du de qian di xia ,reng ju you hen gao de jian ce jing du 。ben wen yan jiu cheng guo ke yong yu jian chan wen li huo fu za wen li de zhi wu tu xiang ci dian jian ce ,di gao le xian you jian ce fang fa de zi kuo ying xing ji jian ce jing du ,xiang guan suan fa ke tui an dao zhi zhang 、lv bo huo gang cai deng gong ye chan pin biao mian de que xian jian ce ,ju you an fan de ying yong qian jing 。
论文参考文献
论文详细介绍
论文作者分别是来自中原工学院的王珺璞,发表于刊物中原工学院2019-07-03论文,是一篇关于织物疵点检测论文,特征融合论文,卷积神经网络论文,深度特征论文,全变差正则项论文,非凸优化论文,中原工学院2019-07-03论文的文章。本文可供学术参考使用,各位学者可以免费参考阅读下载,文章观点不代表本站观点,资料来自中原工学院2019-07-03论文网站,若本站收录的文献无意侵犯了您的著作版权,请联系我们删除。
标签:织物疵点检测论文; 特征融合论文; 卷积神经网络论文; 深度特征论文; 全变差正则项论文; 非凸优化论文; 中原工学院2019-07-03论文;