Classifying fabric defects with evolving Inception v3 by improved L2,1-norm regularized extreme learning machine

Author:

Zhou Zhiyu1ORCID,Yang Xingfan1,Ji Jiangfei1,Wang Yaming2,Zhu Zefei3

Affiliation:

1. School of Information Science and Technology, Zhejiang Sci-Tech University, China

2. Lishui University, China

3. School of Mechanical Engineering, Hangzhou Dianzi University, China

Abstract

To improve efficiency and classification accuracy, and overcome the issue of poor generalization performance of traditional fabric defect classification methods, we present a new fabric defect classification algorithm with evolving Inception v3 by improved L2,1-norm regularized extreme learning machine. Herein, first, the features of fabric images are extracted using Inception v3, which reduces the amount of computation and improves the accuracy of feature extraction. Second, an L2,1-norm regularization extreme learning machine algorithm based on multi-verse optimizer–Henry gas solubility optimization, called MHL21-RELM, is proposed. The proposed algorithm solves the problems of the low classification accuracy of the traditional ELM and prevents the algorithm from easily falling into the local optimal solution. Next, instead of SoftMax, MHL21-RELM is used as the classification network layer of Inception V3, and a new fabric defect classification algorithm named Inv3-MHL21-RELM is proposed to enhance the fabric defect classification performance. Finally, the performance of our algorithm is verified by experiments on different fabric defect datasets. The proposed method achieves 97.29% classification accuracy with the Textile Texture Database; it achieves a 6–30% improvement in classification accuracy compared with other traditional fabric defect classification methods, thus effectively improving the classification accuracy of fabric defects. In addition, this method achieves good classification results with different types of fabric defect datasets, thereby showing strong generalization.

Publisher

SAGE Publications

Subject

Polymers and Plastics,Chemical Engineering (miscellaneous)

同舟云学术

1.学者识别学者识别

2.学术分析学术分析

3.人才评估人才评估

"同舟云学术"是以全球学者为主线,采集、加工和组织学术论文而形成的新型学术文献查询和分析系统,可以对全球学者进行文献检索和人才价值评估。用户可以通过关注某些学科领域的顶尖人物而持续追踪该领域的学科进展和研究前沿。经过近期的数据扩容,当前同舟云学术共收录了国内外主流学术期刊6万余种,收集的期刊论文及会议论文总量共计约1.5亿篇,并以每天添加12000余篇中外论文的速度递增。我们也可以为用户提供个性化、定制化的学者数据。欢迎来电咨询!咨询电话:010-8811{复制后删除}0370

www.globalauthorid.com

TOP

Copyright © 2019-2024 北京同舟云网络信息技术有限公司
京公网安备11010802033243号  京ICP备18003416号-3