基于卷积神经网络的机织物密度均匀性检测  

Detection of fabric density uniformity based on convolutional neural network

作　　者：孟朔 夏旭文 潘如如[1] 周建[1] 王蕾[1] 高卫东[1] MENG Shuo;XIA Xuwen;PAN Ruru;ZHOU Jian;WANG Lei;GAO Weidong(Key Laboratory of Eco-Textiles (Jiangnan University), Ministry of Education, Wuxi, Jiangsu 214122, China)

机构地区：[1]生态纺织教育部重点实验室(江南大学),江苏无锡214122

出　　处：《纺织学报》2021年第2期101-106,共6页Journal of Textile Research

基　　金：国家自然科学基金项目(61976105);江苏省研究生科研与实践创新计划项目(KYCX20_1942);中国纺织工业联合会应用基础研究项目(J202006)。

摘　　要：针对目前基于机器视觉的机织物密度自动检测时织物检测视野小、精度低、品种适应性差的问题,提出一种基于多尺度卷积神经网络的检测方法。首先设计了一套离线图像采集系统连续采集织物图像,并建立一个包含详细织物参数的织物图像数据集;然后采用一种具有不同大小局部感受野的多尺度卷积神经网络适应不同大小的织物结构特征,定位纱线位置;最后利用霍夫变换及灰度投影方法处理网络模型所预测的纱线位置图,计算织物经纬密度,并对织物密度均匀性做出评价。结果表明:与其他方法相比,本文方法对于不同类型织物的经纬密度计算误差小于2%,检测精度更高,品种适应性更强。Aiming at the narrow field of vision,low precision,and low adaptability in the current automatic methods for measuring fabric density,a multi-scale convolutional neural network was proposed.An offline image acquisition system was designed to acquire fabric images continuously,and a fabric image dataset containing detailed fabric parameters was established.Then,a multi-scale convolutional neural network with different sizes of local receptive fields was used for dealing with different fabric structure features and for locating yarns.The Hough transform and gray projection method were used to process the predicted yarn position map in order to calculate the warp and weft density and evaluate the fabric density uniformity.The results show that,compared with other methods,the detecting error of the warp and weft densities for different types of fabrics is less than 2%,which indicates that the proposed method has a higher accuracy and stronger variety adaptability.

关 键 词：机织物 织物密度 织物参数 图像处理 卷积神经网络 

分 类 号：TS101.92[轻工技术与工程—纺织工程]