基于卷积神经网络的织物起球等级客观评定

Objective Estimation of Fabric Pilling Grade Based on Convolutional Neural Network

探讨一种基于卷积神经网络的织物起球等级客观评定的方法。首先采集精梳毛织物起球标准样照扫描图像并提取出圆形起球区域,运用傅里叶变换技术和频域滤波去除织物纹理分量,在此基础上,对圆形起球区域进行分割采样得到子样本并制作训练集和测试集。然后建立一个由3个卷积层和2个全连接层组成的卷积神经网络用于起球等级的客观评定。研究了子样本尺寸对评定准确率的影响。结果表明:子样本尺寸为600 pixel×600 pixel时,评定准确率可达98.5%。认为:使用卷积神经网络模型来评定织物起球等级是可行的,并且可以达到实时性要求。

A method for the objective estimation of fabric pilling grade based on convolutional neural network was discussed.The pilling standard sample image of the combed wool fabric was firstly collected and scanned,and the round pilling area was extracted.Fourier converter technique and frequency domain filtering were used to remove fabric texture weight.Based on this,subsample was obtained by dividing the round pilling area and sampling to make the training set and test set.A convolutional neural network with three convolutional layers and two whole connection layers was established for the objective estimation of pilling grade.The influence of subsample size on the accuracy of grade estimation was studied.The results showed that the accuracy of grading estimation could be reached up to 98.5%when the subsample size was 600 pixel×600 pixel.It is considered that using convolutional neural network for fabric pilling grade estimation is feasible.And it can meet the real-time requirement.