针对布匹瑕疵自动化检测,基于传统的机器视觉方法依赖于人工设计特征,对具有复杂背景图案的花色布瑕疵特征提取难度非常大,因此提出一种基于改进Faster R-CNN(faster region with convolutional neural network)的花色布瑕疵检测算法。...针对布匹瑕疵自动化检测,基于传统的机器视觉方法依赖于人工设计特征,对具有复杂背景图案的花色布瑕疵特征提取难度非常大,因此提出一种基于改进Faster R-CNN(faster region with convolutional neural network)的花色布瑕疵检测算法。在Faster R-CNN的基础上使用Resnet-50作为主干网络,嵌入可变形卷积来提高瑕疵特征的学习能力。通过设计多尺度模型来提高小瑕疵的检测,引入级联网络来提高瑕疵检测精度和定位准确度,构造优化的损失函数来降低样本不平衡影响。通过试验验证了该算法的有效性。结果表明,瑕疵检测效果准确率达94.97%,并能精准定位瑕疵位置,可满足工厂的实际需求。展开更多
Digital image design is one of advanced technique in textile design. The investigation into digital Jacquard textile design in the colorful mode is one form of research in digital Jacquard fabric design, which aimed a...Digital image design is one of advanced technique in textile design. The investigation into digital Jacquard textile design in the colorful mode is one form of research in digital Jacquard fabric design, which aimed at expanding past and present jacquard design and production methods towards innovative ends. In this paper, the design principles and design methods for unconventional digital Jacquard fabric design in colorful mode have been analyzed based on the new technologies and computer applied color theory. The results of this study will enhance further research in the area of digital textile.展开更多
文摘针对布匹瑕疵自动化检测,基于传统的机器视觉方法依赖于人工设计特征,对具有复杂背景图案的花色布瑕疵特征提取难度非常大,因此提出一种基于改进Faster R-CNN(faster region with convolutional neural network)的花色布瑕疵检测算法。在Faster R-CNN的基础上使用Resnet-50作为主干网络,嵌入可变形卷积来提高瑕疵特征的学习能力。通过设计多尺度模型来提高小瑕疵的检测,引入级联网络来提高瑕疵检测精度和定位准确度,构造优化的损失函数来降低样本不平衡影响。通过试验验证了该算法的有效性。结果表明,瑕疵检测效果准确率达94.97%,并能精准定位瑕疵位置,可满足工厂的实际需求。
文摘Digital image design is one of advanced technique in textile design. The investigation into digital Jacquard textile design in the colorful mode is one form of research in digital Jacquard fabric design, which aimed at expanding past and present jacquard design and production methods towards innovative ends. In this paper, the design principles and design methods for unconventional digital Jacquard fabric design in colorful mode have been analyzed based on the new technologies and computer applied color theory. The results of this study will enhance further research in the area of digital textile.