Multi-Layer Feature Extraction with Deformable Convolution for Fabric Defect Detection

In the textile industry,the presence of defects on the surface of fabric is an essential factor in determining fabric quality.Therefore,identifying fabric defects forms a crucial part of the fabric production process.Traditional fabric defect detection algorithms can only detect specific materials and specific fabric defect types;in addition,their detection efficiency is low,and their detection results are relatively poor.Deep learning-based methods have many advantages in the field of fabric defect detection,however,such methods are less effective in identifying multiscale fabric defects and defects with complex shapes.Therefore,we propose an effective algorithm,namely multilayer feature extraction combined with deformable convolution(MFDC),for fabric defect detection.In MFDC,multi-layer feature extraction is used to fuse the underlying location features with high-level classification features through a horizontally connected top-down architecture to improve the detection of multi-scale fabric defects.On this basis,a deformable convolution is added to solve the problem of the algorithm’s weak detection ability of irregularly shaped fabric defects.In this approach,Roi Align and Cascade-RCNN are integrated to enhance the adaptability of the algorithm in materials with complex patterned backgrounds.The experimental results show that the MFDC algorithm can achieve good detection results for both multi-scale fabric defects and defects with complex shapes,at the expense of a small increase in detection time.

Review of Fabric Defect Detection Based on Computer Vision

In textile inspection field,the fabric defect refers to the destruction of the texture structure on the fabric surface.The technology of computer vision makes it possible to detect defects automatically.Firstly,the overall structure of the fabric defect detection system is introduced and some mature detection systems are studied.Then the fabric detection methods are summarized,including structural methods,statistical methods,frequency domain methods,model methods and deep learning methods.In addition,the evaluation criteria of automatic detection algorithms are discussed and the characteristics of various algorithms are analyzed.Finally,the research status of this field is discussed,and the future development trend is predicted.

An Enhanced Nonlocal Self-Similarity Technique for Fabric Defect Detection

Fabric defect detection has been an indispensable and important link in fabric production,many studies on the development of vision based automated inspection techniques have been reported.The main drawback of existing methods is that they can only inspect a particular type of fabric pattern in controlled environment.Recently,nonlocal self-similarity(NSS)based method is used for fabric defect detection.This method achieves good defect detection performance for small defects with uneven illumination,the disadvantage of NNS based method is poor for detecting linear defects.Based on this reason,we improve NSS based defect detection method by introducing a gray density function,namely an enhanced NSS(ENSS)based defect detection method.Meanwhile,mean filter is applied to smooth images and suppress noise.Experimental results prove the validity and feasibility of the proposed NLRA algorithm.

Fabric Defect Detection Using Adaptive Wavelet Transform

A method of woven fabric defect detection using the wavelet transform adaptive to the fabric has been developed. With reference to the orthogonality constrains of Daubechies wavelet, by taking the mmimization of the energy or the gray level of the pixels in the output sub-images as the additional conditions and using the random algorithm method, two sets of wavelet filters adapted to the fabric texture were formed. The original images of normal fabric texture and the fabric texture with defects were decomposed into horizontal and vertical sub- images by using these filters and the feature indices of these sub-images were also extracted. By comparing the feature indices of the normal texture with that of the defective texture, the fabric defects can be successfully detected and located.

Fabric Defect Detection Using GMRF Model

It has been testified that the Gauss Markov random field model is most suitable for the characterization of fabric texture among a variety of available models because of its approximately constant character and the normality of the gray-level distribution found with typical fabric images. However, the general Gauss-Markov random field(GMRF) method for fabric defect detection is not always ideal in practice since in some cases, the estimated model parameters make the Markov error covariance not positively definite, which may render the method to fail thoroughly. In this paper, the use of the GMRF model for defect detection of fabric is discussed and an approach to this problem is proposed. Some detailed texture may be overlooked in this way, but good detection results can still be expected as far as fabric defect detection is concerned.

Fabric Defect Detection Using Independent Component Analysis and Phase Congruency

A novel method based on independent component analysis and phase congruency is proposed for detecting defects in textile fabric images. By independent component, we can obtain textile structural features of fabric-free images. By phase congru- ency, structure information is reduced, which can distinguish the defect region from the defect-free regions. Finally, we have the detecting result from binary image which is obtained by a thresh- old step, Compared with other algorithms, the proposed method not only has robustness with high detection rate, but also detects various types of defects quite well.

Semantic Segmentation Using DeepLabv3+ Model for Fabric Defect Detection

Currently, numerous automatic fabric defect detection algorithms have been proposed. Traditional machine vision algorithms that set separate parameters for different textures and defects rely on the manual design of corresponding features to complete the detection. To overcome the limitations of traditional algorithms, deep learning-based correlative algorithms can extract more complex image features and perform better in image classification and object detection. A pixel-level defect segmentation methodology using DeepLabv3+, a classical semantic segmentation network, is proposed in this paper. Based on ResNet-18,ResNet-50 and Mobilenetv2, three DeepLabv3+ networks are constructed, which are trained and tested from data sets produced by capturing or publicizing images. The experimental results show that the performance of three DeepLabv3+ networks is close to one another on the four indicators proposed(Precision, Recall, F1-score and Accuracy), proving them to achieve defect detection and semantic segmentation, which provide new ideas and technical support for fabric defect detection.

Detection of Fabric Defects with Fuzzy Label Co-occurrence Matrix Set

Co-occurrence matrices have been successfully applied in texture classification and segmentation.However,they have poor computation performance in real-time application.In this paper,the efficient co-occurrence matrix solution for defect detection is focused on,and a method of Fuzzy Label Co-occurrence Matrix (FLCM) set is proposed.In this method,all gray levels are supposed to subject to some fuzzy sets called fuzzy tonal sets and three defective features are defined.Features of FLCM set with various parameters are combined for the final judgment.Unlike many methods,image acquired for learning hasn't to be entirely free of defects.It is shown that the method produces high accuracy and can be a competent candidate for plain colour fabric defect detection.

Global Fabric Defect Detection Based on Unsupervised Characterization

Fabric texture intelligent analysis comprises the following characteristics:objective detection results,high detection efficiency,and accuracy.It is significantly vital to replace manual inspection for smart green manufacturing in the textile industry,such as quality control and rating,and online testing.For detecting the global image,an unsupervised method is proposed to characterize the woven fabric texture image,which is the combination of principal component analysis(PCA)and dictionary learning.First of all,the PCA approach is used to reduce the dimension of fabric samples,the obtained eigenvector is used as the initial dictionary,and then the dictionary learning method is operated on the defect-free region to get the standard templates.Secondly,the standard templates are optimized by choosing the appropriate dictionary size to construct a fabric texture representat ion model that can effectively characterize the defec-free texture region,while ineffectively representing the defective sector.That is to say,through the mechanism of identifying normal texture from imperfect texture,a learned dictionary with robustness and discrimination is obtained to adapt the fabric texture.Thirdly,after matching the detected image with the standard templates,the average filter is used to remove the noise and suppress the background texture,while retaining and enhancing the defect region.In the final part,the image segmentation is operated to identify the defect.The experimental results show that the proposed algorithm can adequately inspect fabrics with defects such as holes,oil stains,skipping,other defective types,and non-defective materials,while the detection results are good and the algorithrm can be operated flexibly.

Automatic Image Inspection of Fabric Defects Based on Optimal Gabor Filter

An effective method for automatic image inspection of fabric defects is presented. The proposed method relies on a tuned 2D-Gabor filter and quantum-behaved particle swarm optimization( QPSO) algorithm. The proposed method consists of two main steps:( 1) training and( 2) image inspection. In the image training process,the parameters of the 2D-Gabor filters can be tuned by QPSO algorithm to match with the texture features of a defect-free template. In the inspection process, each sample image under inspection is convoluted with the selected optimized Gabor filter.Then a simple thresholding scheme is applied to generating a binary segmented result. The performance of the proposed scheme is evaluated by using a standard fabric defects database from Cotton Incorporated. Good experimental results demonstrate the efficiency of proposed method. To further evaluate the performance of the proposed method,a real time test is performed based on an on-line defect detection system. The real time test results further demonstrate the effectiveness, stability and robustness of the proposed method,which is suitable for industrial production.

改进YOLOv5的织物缺陷检测方法

为了在不增加网络参数量的条件下提升深度学习方法对织物缺陷检测的精度,提出了一种基于改进YOLOv5的织物缺陷检测方法。通过深度卷积改造通道注意力,剪裁最大池化优化空间注意力,并通过二者构建的双级联注意力机制来搭建特征提取子网络,从而提高网络对缺陷区域纹理和语义特征的提取能力;采用鬼影混洗卷积改进特征融合子网络,强化对提取特征的筛选,在降低模型参数量的同时,改善缺陷信息丢失和无效信息冗余问题;在检测端引入具有角度损失的新型损失函数SIOU,来促进真实框和预测框的拟合并提升对缺陷预测的准确性。实验结果表明:改进的YOLOv5方法在降低YOLOv5基准模型复杂度和计算量的同时,与YOLOv7等六种先进方法相比,可获得更高的检测精度,相较原模型mAP@0.5值提高了2.6个百分点,mAP@0.5:0.9值提高了1.3个百分点。

基于改进YOLOv5算法的织物缺陷检测

基于现有织物缺陷检测算法受疵点尺寸与织物纹理背景的影响导致检测精度较低,同时检测模型过于复杂,难以部署到工控设备上,无法满足织物缺陷实时检测等现状,提出一种改进YOLOv5算法的织物缺陷检测算法。以YOLOv5算法为基准模型,采用跨阶段部分连接残差网络替代原模型的主干网络,增强模型上下文特征信息学习能力;将SimAM注意力机制融入到模型中,提升对有用特征的提取能力,抑制无用纹理背景特征的干扰;引入WIoU与Varifocal Loss损失函数,提高回归框准确性的同时降低负样本权重;最后,针对织物的小目标疵点难以检测的问题,提出增加小目标检测层的方法,提高模型的检测能力。试验结果表明:该研究算法能够快速准确地检测织物疵点,精确率与mAP分别达到86.46%与84.4%,与基准模型相比,分别提高6.16个百分点和5.8个百分点。

基于多度量多模型图像投票的织物表面瑕疵检测方法

为解决自动化生产线上织物表面瑕疵检测准确率低和计算速度慢的问题,利用织物表面具有周期纹理的特性提出了一种改进的RANSac检测方法,即多度量多模型图像投票。首先将输入图像裁剪为尺寸一致的子图,计算出子图多维度量的输出值矩阵;然后与改进RANSac计算出的无瑕疵背景的多维度量标准值分别对应作差,采用投票得出每张子图的基础分;再将其在4个记数模型下得到的综合评分排序,根据顺序和偏移量在输出端得到外点所代表的瑕疵子图。实验结果表明:在自采样的织物瑕疵数据集上,选择单度量和单模型的预测精度平均可达到90.9%,平均预测时间达到0.139 s,综合多度量多模型投票的平均预测精度可达到92.7%。该算法不需要大量前期数据进行训练,适用于纯色和条纹状织物的实时表面缺陷检测。

基于直方图均衡化的毛织物服装印花缺陷检测

为了保证毛织物印花服装的生产质量,提出基于直方图均衡化的毛织物服装印花缺陷检测方法。设置毛织物服装印花不同缺陷特征作为检测标准,利用光学成像原理采集毛织物服装印花图像,通过颜色转换、图像滤波等步骤,实现初始印花图像的预处理;利用直方图均衡化技术提取服装印花图像特征,通过特征匹配确定缺陷状态与类型,实现毛织物服装印花缺陷的检测。测试结果表明,优化设计方法得出缺陷面积检测误差的平均值为0.09 mm^(2),缺陷类型检测错误率较低。

基于改进甲壳虫全域搜索算法的机织物疵点检测

为解决深度学习模型在面对跨场景的织物疵点检测时存在泛化性能差的问题,在甲壳虫全域搜索算法(BAS)的基础上添加了本地搜索能力构建了一种基于甲壳虫算法的混合算法,该算法可具体分为训练阶段和检测阶段。在训练阶段,通过对无疵点织物进行训练构建二维Gabor滤波器,然后使用改进BAS的混合模型对Gabor滤波器的参数进行了优化,使改进后的算法具备全局搜索和局部搜索的能力;在检测阶段,根据在训练阶段获得最佳参数构造Gabor滤波器,对待检测的织物图像进行卷积运算,并对卷积后图像进行二值化处理,最终识别待测试织物是否含有疵点。实验结果表明:该方法的特征提取具有良好的类别区分性,更加集中在疵点范围内,检测准确率可达99.26%,具有良好的稳定性和泛化性能。

基于改进YOLOv5s的轻量化布匹瑕疵检测算法

针对纺织生产中布匹瑕疵检测高精度、实时性的需求,提出了一种基于改进YOLOv5s的轻量化布匹瑕疵检测算法(GhostNet-CBAM-Partial convolution-YOLOv5s,GCP-YOLOv5s)。该算法首先引入GhostNet中的GhostConv模块,对原主干网络进行优化重构,大幅减少网络参数;其次,在主干特征提取网络中加入CBAM(Convolutional block attention module)注意力机制,增加网络的特征提取能力;最后,设计了基于Partial convolution的改进C3模块(C3-Partial convolution,C3-P),在降低模型参数量的同时提高特征融合能力。在自建布匹瑕疵数据集上进行了对比测试,结果表明:与基准模型YOLOv5s相比,GCP-YOLOv5s的参数量降低了41.6%,计算量降低了43.1%,检测速度提高了12 FPS,检测精度提升了1.7%。GCP-YOLOv5s算法在保证模型轻量化的同时具有较高的检测精度,可以满足布匹瑕疵检测的高精度和实时性要求。

基于改进YOLOv7的织物疵点小目标检测算法

针对织物疵点种类多、尺度变化大、小目标易漏检等问题,提出了一种基于YOLOv7的改进算法(YOLOv7⁃ESL)。首先,将注意力机制ECA⁃Net融入到Neck层中取代CBS模块,在少量增加计算成本的情况下显著提高了检测精度。其次,设计了专用探测头检测小目标,充分利用网络的浅层特征信息,使模型能够有效地检测多尺度的目标。最后,在特征加强部分增加Swin Transformer V2 Block,能够捕捉全局和局部之间的丰富关系,提高模型检测小目标缺陷的能力。试验结果表明:YOLOv7⁃ESL算法精确率为97.7%,召回率为90.3%,平均精度均值为94.9%,FPS为29.9帧/s;与原始YOLOv7模型相比,分别提高了3.8个百分点、3.4个百分点、3.3个百分点、3.1帧/s,可满足纺织工业领域的应用要求。

基于DCA-CenterNet的棉布瑕疵检测研究

针对棉布生产过程中常存在各类瑕疵,且瑕疵的尺度差异较大、形态各异、对比度低、部分为小目标瑕疵等特点,提出了一种基于CenterNet网络的改进目标检测模型DCA-CenterNet,并首次应用于棉布瑕疵检测.在骨干网络Hourglass的残差模块引入CA注意力机制,可以捕捉到不同位置之间的空间关系和上下文信息,提高了网络对于棉布瑕疵的特征表达能力;设计了基于定位质量的关键点筛选模块,可以有效地捕获关键位置信息,提高算法模型检测精度;采用多组改进的带有基于定位质量的关键点筛选模块的检测器,以更好地适应棉布瑕疵的种类多样性和尺度差异,有效解决棉布瑕疵极端长宽比问题.实验结果表明:提出的模型相较于改进前的模型,mAP提高了4.14%,比YOLOv5、FasterRCNN算法分别高出了4.20%和9.11%,验证了所提模型的有效性.

改进YOLOv5的布匹缺陷检测方法

现阶段布匹缺陷种类繁杂,且包含大量人眼难以辨别的小目标缺陷和长宽比极端不平衡缺陷,使得在复杂背景下的布匹缺陷检测成为一项艰巨任务。为此,提出一种改进YOLOv5的布匹缺陷检测方法。首先,在YOLOv5的C3模块中增加注意力机制NAM,设计为C3NAM模块,其可以抑制特征值中不显著的权重,在保持性能的同时进行高效计算;其次,采用一个新的CNN模块SPD-Conv,以解决大部分的布匹缺陷检测在分辨率较低或者瑕疵较小时性能迅速下降的问题;最后,在检测端引入新的损失函数Alpha-IoU,促进真实框和预测框的拟合,并提升对缺陷预测的准确性。实验结果表明:改进的YOLOv5网络模型较原YOLOv5网络模型mAP@0.5值提高了5.4%,mAP@0.5:0.95值提高了2.2%,且检测效果优于原网络模型和其他主流目标检测模型。

基于改进YOLOv5的织物瑕疵检测方法

针对织物瑕疵中部分瑕疵目标小、长宽比极端等问题,提出一种基于改进YOLOv5的织物瑕疵检测方法。该方法在YOLOv5模型基础上引入自注意力机制CoTNet网络,并将颈部网络中的PAFPN网络优化为BiFPN网络,同时将目标损失函数改进为CIoU损失函数,加强模型对邻近键以及上下文之间特征信息的收集,在增强模型对小目标和尺寸变化大类型瑕疵检测能力的同时可获得更准确的边界框回归,加快收敛速度。实验证明,本文改进的模型在织物瑕疵检测数据集上的检测效果和YOLOv5模型相比平均精度均值提升了6.8%,准确率提升了6.7%,模型验证有效。