Weave pattern recognition by measuring fiber orientation with Fourier transform

An effective method based on measuring the fiber orientation of yarn floats with two-dimensional Fourier transform (2-D FFT) is proposed to recognize the weave pattern of yarn-dyed fabric in the high-resolution image. The recognition process consists of four main steps: 1. High-resolution image reduction, 2.Fabric image skew correction, 3.Yarn floats localization, 4. Yarn floats classification. Firstly, the high-resolution image is reduced by the nearest interpolation algorithm. Secondly, the skew of the fabric image is corrected based on Hough transform. Thirdly, the yarn floats in the fabric image is localized by the yarns segmentation method based on the mathematical statistics of sub-images. Fourthly, the high-resolution image is corrected and its yarns are segmented successively according to the inspection information of the reduced image. The fiber orientations are detected by 2-D FFT, and the yarn floats are classified by k-means clustering algorithm. Experimental results and discussions demonstrate that, by measuring the fiber orientation of yarn floats, the proposed method is effective to recognize the yarn floats and the weave pattern for yarn-dyed, solid color, and gray fabrics.     

织物自动检测系统设计与实现

应用4片DSP+FPGA流水阵列结构,用现场可编程门阵列FPGA对采集的视频数字图像信息进行处理,实现了织物疵点自动检测。设计了以4片TMS320C62x为数字图像信息处理核心,由FPGA实现系统控制互连的织物疵点自动检测系统,设计了基于直方图统计和基于支撑矢量机的织物疵点分类识别算法。实验结果表明,当样布传送速度达到100、120 m/min时,该织物疵点自动检测系统对样布的疵点识别准确率分别达到80%和60%。     

A real-time computer vision-based platform for fabric inspection part 2: platform design and real-time implementation

In practice, off-line fabric defect detection is a real-time inspection process, requiring both inspection system and detection algorithms capable of high real-time performance. This paper presents a computer vision-based platform for the fabric’s visual quality assurance. The proposed platform is composed of four subsystems including: fabric drive, illumination and imaging, image acquisition and processing and human–computer interaction. The design of high-performance embedded system based on FPGA + DSP architecture is presented and discussed. After that, an extensive discussion on system settings, real-time implementation issues are presented. Finally, a real-time inspection experiment on 160-m real-world fabrics with 11 defect types is conducted. The experiment results show the proposed platform exhibits a good real-time response, and can achieve an overall 89% detection rate with a low false alarm at a speed of 30 m/min.     

基于机器学习的布匹瑕疵识别研究

纺织工业是我国制造业出口的重要组成部分。布匹的质量控制在纺织工业中尤为重要,而布匹瑕疵是影响布匹质量控制的重要因素之一。在中小企业中,布匹瑕疵识别主要依靠人工流水线作业,存在着人工成本高、人眼识别准确度低等问题。因此,一个有效的布匹瑕疵检验系统是十分必要的,布匹瑕疵分类算法是保证疵点判决效率的核心。基于布匹生产企业存在的问题,有针对性地研究了机器学习与计算机视觉的布匹瑕疵识别算法的基本原理,介绍了各类布匹瑕疵识别中的检测与分类算法,将最近发展迅速的机器学习的理论研究引入布匹瑕疵识别中,对涉及机器学习的模式识别算法进行了介绍。     

基于相似性定位和超像素分割的织物疵点检测

为解决周期性纹理织物图像的疵点检测及其轮廓精确分割问题,提出一种基于相似性定位和超像素分割的织物疵点检测方法。将待检测图像进行中值滤波和对数增强,并利用FT算法估计增强图像的显著图实现待检测图像的预处理;将基于归一化局部均值差分的灰度相似性检测参量和结构相似性检测参量结合,构建可测量更多类型周期性纹理织物图像的相似性度量函数,通过阈值化增强图像分块的相似性测量值实现疵点在显著图中的粗定位;最后对显著图粗定位图像分块进行超像素细分割及其二值化处理,并借助连通域分析剔除孤立点,获得完整的疵点轮廓。结果表明,本方法与常规3种方法相比,对周期性纹理织物图像的疵点检测准确率更高,且提取出的疵点轮廓更精确。     

应用最优Gabor滤波器的经编织物疵点检测

针对经编织物疵点自动检测问题,提出了一种新的基于最优Gabor滤波器的经编织物疵点检测方法。具体可分为学习阶段和检测阶段;在学习阶段,对于无疵点的经编织物图像构造可调制的二维Gabor滤波器,采用量子行为粒子群优化(QPSO)算法对Gabor滤波器的参数进行优化,得到与无疵点的织物图像纹理特征最匹配的Gabor滤波器参数;在检测阶段,由学习阶段得到的最佳参数构造Gabor滤波器,用该滤波器对待检测织物图像进行卷积处理,然后再对得到的卷积图像进行二值化处理,最终识别出待检测织物是否有疵点存在。结果表明,该方法的检测率可以达到96.67%,具有很好的稳定性和鲁棒性,适合应用于工业生产。     

A computer vision-based system for automatic detection of misarranged warp yarns in yarn-dyed fabric. Part I: continuous segmentation of warp yarns

In textile and garment industries, misarranged warp yarns of yarn-dyed fabrics disorganize the layout of fabrics and lead to poor product quality. This series of studies aims to develop a computer vision-based system for automatic detection of misarranged color warp yarns in terms of high efficiency and good accuracy. Four main parts are included in this series of studies: warp yarn segmentation, fabric image stitching, warp regional segmentation, and yarn layout proofing. This paper proposes a continuous segmentation method of warp yarns to detect the misarranged color warp yarns for yarn-dyed fabrics automatically, which is the foundation of the developed computer vision-based system. The proposed framework consists of two main components: warp yarn segmentation and fabric image stitching. Firstly, the sequence images of a fabric stripe are captured using a designed offline image acquisition platform. Secondly, the warp yarns in the sequence images are segmented by a sub-image projection-based method successively. Thirdly, the sequence images are stitched by a yarn-template matching method based on their warp segmentation results. Finally, the continuous segmentation result of warp yarns is saved for the further processing of warp regional segmentation and color warp layout proofing. The proposed method has been evaluated on 720 fabric images of five fabric examples with plain and 2/2 twill, and experimental results show that the proposed method can realize the continuous segmentation of warp yarns in yarn-dyed fabrics with the yarn segmentation accuracy of 97.43% and image stitching accuracy of 99.53%.     

基于机器视觉的筒子纱缺陷在线检测系统

为提高筒子纱检测过程的自动化程度,设计了一种基于机器视觉的筒子纱缺陷在线检测系统。该系统由2个工业相机、条形LED光源、对照式光电开关和计算机组成。首先,相机与同步光源分时采集筒子纱顶面和侧面过曝模式及正常模式图像。然后通过对顶面过曝图像自适应分割来定位筒子纱中心。其次,通过极坐标变换展开顶面正常图像。最后,在顶面展开图中,分别利用垂直方向边缘分布的投影特征、纹理及强度一致性、局部方向直方图纹理识别菊花芯、多源纱和网纱缺陷;在筒子纱侧面图中,通过投影法快速确定边界位置,并通过轮廓拟合程度识别多层台缺陷。结果表明,该系统可实时识别多层台、网纱、菊花芯、多源纱等筒子纱缺陷,具有较好的检测效果。     

基于改进图像阈值分割算法的纱线疵点检测

针对纺织行业纱线疵点检测方法可靠性差、灵敏度低、检测速度低的问题,提出一种基于数字图像处理的纱线疵点判定方法。首先,搭建纱线图像采集系统,完成纱线图像采集;其次,针对纱线边缘信息难处理以及传统双边滤波对椒盐噪声处理效果差的问题,对双边滤波进行改进,改进后的双边滤波可有效保存纱线边缘信息;再者,针对传统阈值分割计算量大、最佳阈值难以寻找的问题,对传统阈值分割算法进行改进,改进的阈值分割算法在保证处理效果的同时提高了整体算法的处理速度;最后,采用亚像素对纱线边缘进行计算,提高了纱线疵点检测的精确度。实验结果验证了算法的有效性及可靠性,该算法在提高精确度的同时将检测速度提高了20%以上,对提高纱线质量检测的准确性具有重要意义。     

基于相对总变差模型与自适应形态学的织物瑕疵检测

为解决目前基于图像处理的织物瑕疵检测算法中,因织物纹理的多样性与瑕疵形状尺寸的不确定性所造成的检测效果差的问题,提出一种基于结构-纹理模型与自适应数学形态学的织物瑕疵检测算法。首先采用相对总变差模型对织物图像进行滤波以去除织物纹理,然后在得到的灰度图像上直接进行基于自适应邻域的灰度形态学运算,形态学算子采用开运算算子,最终得到织物瑕疵的增强图像。采用基于相对总变差模型与自适应形态学相结合的方法与2种已知的Gabor算法进行比对,对4类典型织物瑕疵进行检测实验和分析。结果表明,本文方法能更好地提取出织物瑕疵。     

基于深度学习的布匹疵点检测方法

为快速、准确检测布匹疵点,提出以深度学习目标检测框架YOLOv4为基础的布匹疵点检测方式,首先将5种常见疵点图像(吊经、百脚、结点、破洞、污渍)进行预处理,然后将图像输入到YOLOv4算法中进行分类。YOLOv4采用CSPDarknet53作为主干网络提取疵点特征,SPP模块、FPN+PAN的方式作为Neck层进行深层疵点特征提取,预测层采用3种尺度预测方式,对不同大小的疵点进行检测。研究结果表明:经600个测试集样本的验证,该方法对疵点图像的检测准确率达95%,检测单张疵点图像的速率为33 ms。与SSD、Faster R-CNN、YOLOv3方法进行比较,采用YOLOv4方法准确率更高,速度更快。     

基于深度信念网络的织物疵点检测

为提高织物疵点检测精度和效率,提出了一种基于深度信念网络的织物疵点检测方法。用改进的受限玻尔兹曼机模型对深度信念网络进行训练,完成模型识别参数的构建。利用同态滤波方法对图像进行预处理,使疵点图像更加清晰,同时抑制了背景图像。以Python语言,基于TensorFlow框架构建深度信念网络模型,对织物疵点图像进行处理得到学习样本,确定模型激活函数后,分析了各模型参数对织物疵点检测准确率的影响规律,得到激活函数为Relu, Dropout值为0.3,预训练学习率为0.1,微调学习率为0.000 1,批训练个数为64时,模型参数值达到最优。最后,利用在无缝内衣机上采集到的各类疵点图像,对深度信念网络织物疵点检测模型进行验证。结果表明：所提出的织物疵点检测方法能够快速、有效地对织物疵点进行检测和分类识别,准确率达到98%。     

Fabric defect detection based on the saliency map construction of target-driven feature

In inspection of fabric surface quality in production line, small defects have to be detected in a large background. In this paper, a new method is put forward to detect fabric surface defect by target-driven features. First of all, surface defect feature of fabric is analyzed; and then, area feature of and number feature of defects are used as tasks, which drive to enhance saliency of defective regions and to form feature saliency maps; finally, by using threshold segmentation, fusion, and filtering, fabric defect is gained from the feature saliency maps. Experiments show that the detection algorithm, compared with classic defect algorithm, can achieve accurate segmentation of the surface defects, better anti-noise ability, higher detection accuracy, which has a strong applicability on the fabric defect detection, and provides the possibility for realizing automatic detection of textile industrial product surface defect.     

应用积分图的织物瑕疵检测快速算法

为解决现有基于图像处理的织物瑕疵检测算法实时性较差、正确率偏低等问题,提出一种包含学习和检测2个阶段的瑕疵检测算法。通过对无瑕疵模板图像的梯度能量特征及其分布特性的学习,自适应获得检测阶段所需的参数。一方面利用积分图原理将任意大小的图像块内的求和运算化简为三次加法运算,快速提取织物图像的梯度能量特征,实现织物瑕疵的实时检测,另一方面利用核函数拟合特征参数分布,结合均值漂移法求解分布峰值获得自适应的瑕疵判定阈值参数,实现织物瑕疵的准确分割。通过实验将本文算法与现有基于局部二值模式特征、小波特征、规则带特征等算法进行对比,针对包含3种纹理6类瑕疵的织物图像数据集的测试结果显示,本文算法平均处理时间为56ms,正确率为97%。     

A computer vision-based system for automatic detection of misarranged warp yarns in yarn-dyed fabric. Part II: warp region segmentation

This series of studies aim to develop a computer vision-based system for automatic detection of misarranged color warp yarns to replace manpower and improve efficiency. Based on the warp yarn segmentation and fabric image stitching methods presented in Part I, this paper proposes a stepwise segmentation method of warp regions, as a core of the developed computer vision-based system, to detect the layout of color yarns for yarn-dyed fabrics automatically. The proposed framework consists of two main components: rough warp region segmentation and precise warp region merging which are realized by analyzing correlation coefficient of color histograms among segmented warp yarns and warp regions successively. The proposed method has been evaluated on 543 fabric images of four fabric samples consisting of 5533 warp regions, and experimental results show that the proposed method can realize the warp region segmentation in yarn dyed fabrics with the average accuracy of 99.47%.     

基于互相关的印花织物疵点检测

为实现印花织物中疵点的自动检测,以互相关理论为基础,结合图像处理技术,以Matlab7.0构建了一套印花织物疵点自动检测系统。在疵点检测过程中,提出以加和表理论为基础实现互相关系数的快速计算。通过对软件模拟的印花花纹疵点的识别,说明这个系统能够实现印花过程中常见的花纹偏移、颜色色差等疵点的自动检测。实际印花织物疵点的检测实验表明,所提出的算法具有有效性、鲁棒性等优点。通过比较不同子窗口大小的检测结果,选定25像素×25像素作为最终检测系统中子窗口的大小。     

基于机器视觉和图像处理的织物疵点检测研究新进展

按照被检测的织物类型并根据当前研究中所使用的方法,简要综述了近年来基于机器视觉和图像处理的织物疵点检测系统新的应用和发展情况。首先分析了织物疵点自动检测研究的理论和现实意义。给出了织物疵点检测系统中视觉图像获取和疵点图像检测两个关键部分的架构。说明了迫切需要进行检测的两类织物白坯布和色织布,着重讨论了对这两类织物进行疵点检测的各种新方法,并详细说明了其检测效果和存在的不足。最后给出了疵点检测研究的几点建议。     

Textural fabric defect detection using statistical texture transformations and gradient search

The inspection of the fabric defects is an important problem, which highly affects both the quality and the cost in the textile industry. Because of consistency and accuracy problems, the inspection of the fabric defect by human experts is neither feasible nor efficient. This requires development and use of automated inspection techniques. Thus, in this study, a texture analysis method, which uses sum and difference histograms (SDH) conjointly with co-occurrence matrices, is proposed to introduce an objective criterion for defect detection. To accomplish the detection task with high accuracy, several features were extracted from SDH and then, a defect search technique, which was developed in the context of this study, was applied. Moreover, several experiments and parameter analysis were performed to carry out detection at feasible computation time and memory storage. The developed method was applied to 28 kinds of raw woven fabric defects and 27 of them (i.e. 93.1%) were successfully recognized by the proposed detection system. The quantitative results and qualitative discussions show the effectiveness of the developed strategy.     

基于视觉校准的环锭纺细纱条干特征在线提取方法

针对纱线高速回转、毛羽条干交织导致的条干轮廓特征难以准确提取的问题,提出了深度学习与形态学运算融合的在线提取方法,设计了图像在线采集系统与校准定焦方法,为轮廓特征提取提供高质量输入,构建了基于整体嵌套边缘检测神经网络和形态学运算的细纱条干轮廓特征提取重构模型,实现毛羽干扰下的条干轮廓在线准确提取。实验结果表明,所提方法的轮廓提取准度指标OIS-F(optimal image scale)、ODS-F(optimal dataset scale)达到了0.91,平均准确率AP达到了0.89,相对于当前方法提高了7%以上。基于提取的轮廓特征计算的条干不匀CV值,与CT3000均匀度检测仪的平均误差小于4%。     

应用机器视觉的织物表面绒毛率测试系统

目前织物表面绒毛含量大都采用人工方式检测,存在效率低、准确度不高等问题。为此,应用机器视觉和图像处理技术,研制了一套织物表面绒毛率测试系统。介绍了织物表面绒毛率测试原理,包括织物表面绒毛率检测数学模型、检测算法和阈值的确定方法,并介绍了织物表面绒毛率测试系统的软硬件组成。采用该测试系统检测了5种织物的表面绒毛率,并与人工检测结果进行了对比分析。结果表明：该测试系统能够高效地测定织物表面绒毛率,且与人工检测结果呈现高度正相关;系统重复检测偏差范围为1.18%～7.25%,可满足织物表面绒毛率的检测需求。