摘要
在纺织印染行业智能化制造的推进过程中,面料质量检测环节的自动化升级已成为提高生产效率、改善产品品质的重要环节。本文提出一个依靠机器视觉、深度学习技术,完成纺织面料瑕疵自动检测及分类的系统。设计了一种适合高速生产线的、高分辨率的线阵成像平台,利用多角度组合光源的设计来克服染整织物复杂的纹理下成像的问题。在算法层面,利用改进的卷积神经网络模型对采集图像进行深度特征提取,采用纹理抑制算法实现瑕疵区域的精准分割,并通过样本增强策略攻克了工业场景下瑕疵样本数据不平衡导致的分类精度受限问题。
In the advancement of intelligent manufacturing within the textile printing and dyeing industry,automation upgrade of fabric quality inspection has become a critical step for enhancing production efficiency and improving product quality.This paper presents a system leveraging machine vision and deep learning technologies to achieve automated defect detection and classification in textile fabrics.A high-resolution line-scan imaging platform suitable for high-speed production lines was designed,employing a multi-angle combined light source configuration to overcome imaging challenges posed by the complex textures of dyed and finished fabrics.At the algorithmic level,an enhanced convolutional neural network model performs deep feature extraction on captured images.Precise segmentation of defect areas is achieved by taking texture suppression algorithms..Additionally,a sample augmentation strategy overcomes classification accuracy limitations caused by imbalanced defect sample data in industrial settings.
作者
程翠玉
郑明言
焦峰亮
CHENG Cuiyu;ZHENG Mingyan;JIAO Fengliang(Weifang Business Vocational College,Weifang 262234,China;Weifang Economic School,Weifang 262234,China;Zhucheng Education and Sports Bureau,Weifang 262234,China)
出处
《染整技术》
2026年第2期39-41,共3页
Textile Dyeing and Finishing Journal
基金
山东省教育厅职业教育教学改革研究项目“基于知识图谱的职业教育精准化教学实施策略研究与实践”(2023341)
山东省教育厅职业教育教学改革研究项目“基于VR技术的工科专业数字化转型升级研究与实践”(2024551)
山东省教育科学研究院教学研究课题“数智时代知识图谱赋能计算机专业精准化教学研究与实践”(2024JXY544
)
山东省高等教育学会教学研究重点课题“基于知识图谱技术的平面设计一流核心课程建设实践研究”(SDGJ2025E10)研究成果。
关键词
机器视觉
面料瑕疵检测
深度学习
线阵相机
自动分类
machine vision
fabric defect detection
deep learning
line-scan camera
automatic classification
