结合视觉显著性和卷积神经网络的提花织物疵点检测技术
2021-11-26李敏杨珊何儒汉姚迅崔树芹
李敏 杨珊 何儒汉 姚迅 崔树芹



摘 要:为了实现提花织物疵点自动检测,提出了一种结合视觉显著性和卷积神经网络的提花织物疵点检测方法。针对提花织物背景干扰的问题,利用视觉显著性模型(Context-aware,CA)抑制背景信息,突出疵点区域的显著性来获得图像的显著图;为了区分织物图像中是否存在疵点,使用在通用数据集上训练过的VGG16神经网络模型对提花织物图像的显著图分类。结果表明:该方法在提花织物疵点检测上平均准确率为97.07%,比直接利用VGG16模型对提花织物疵点检测的准确率提高了19.44%,是一种适合提花织物疵点检测的方法。
关键词:提花织物;疵点检测;视觉显著性;卷积神经网络
中图分类号: TS101.9
文献标志码:A
文章编号:1009-265X(2021)06-0062-05
Jacquard Fabric Defect Detection Technology CombiningContext-awareness and Convolutional Neural Network
LI Min, YANG Shan, He Ruhan, YAO Xun, CUI Shuqin
(School of Computer Science and Artificial Intelligence, Wuhan Textile University, Wuhan 430200, China)
Abstract: In order to achieve the automatic detection of jacquard fabric defects, a method is proposed to detect jacquard fabric defects, which combines context-awareness and convolutional neural network. In order to solve the problem of background interference in the jacquard fabric, a context-aware (CA) model was used to suppress the background information and highlight the salience of the defect area to obtain a context-aware view of the image. To distinguish whether there are defects in the fabric image, the VGG16 neural network model trained on the general data set was used to classify the context-aware views of the image. The results show that this method has an average accuracy of 97.07% in the detection of jacquard fabric defects, which is 19.44% higher than that of the detection of jacquard fabric defects by the direct use of the VGG16 model. It is a suitable method for detecting jacquard fabric defects.
Key words: jacquard fabric; defect detection; context-awareness; convolutional neural network
收稿日期:2020-11-03 網络首发日期:2021-04-15
基金项目:湖北省教育厅科技项目(D20161605)
作者简介:李敏(1978-),女,湖北武汉人,副教授,博士,主要从事图像处理和模式识别方面的研究。
织物疵点检测是提高纺织质量的重要环节,其中提花织物的检测是织物疵点检测中的难点[1]。近年来,随着深度学习技术的快速发展,卷积神经网络作为其主要的算法被广泛运用到图像的缺陷检测中[2],如罗俊丽等[3]提出基于卷积神经网络和迁移学习的色织物疵点检测,该方法是分别利用残差网络模型和在通用数据集上训练过的残差网络模型对预处理过的织物图像进行分类训练,然后比较在大小不同的数据集上训练的效果;曹振军等[4]提出了基于树莓派的深……
