基于LBP和神经网络的织物疵点分类

Classification of Fabric Defects Based on LBP and Neural Network

织物疵点在销售中严重影响着产品的价格与品质,传统的织物疵点检测主要依靠人工检测,这种检测方式如今无法满足机器化时代下的高速度、高精度、高质量的要求。针对织物疵点检测难度大,效率低的问题,文章采用局部二值模式(LBP)和神经网络对织物疵点分类。首先,采用局部二值模式(LBP)对织物疵点纹理特征进行提取;其次,将特征值进行归一化处理并且将获得的特征值如能量、方差、熵等送入到已经训练好的BP神经网络中;最后,通过BP神经网络将前面送入的织物疵点特征值进行织物疵点先识别再分类;研究认为:基于局部二值模式和神经网络的织物疵点检测方法是一种可行的方法。该方法的平均准确率达到80%以上,平均召回率达到80%以上,分类的平均正确率达到85%以上。

The price and quality of products of fabric defects seriously affect sales.The traditional fabric defect detection mainly relies on manual detection,which can not meet the requirements of high speed,high precision and high quality in the era of mechanization.In view of the difficulty and low efficiency of fabric defect detection,in this paper,the rotation-invariant local binary model(ULBP)and neural network are used to classify fabric defects.Firstly,the rotation-invariant local binary model(ULBP)is used to extract the texture features of fabric defects;Secondly,the eigenvalues are normalized and the obtained eigenvalues such as energy,variance,entropy,etc.They are sent into the trained BP neural network;Finally,the fabric defect feature values sent in front are classified by BP neural network;It is considered that the fabric defect detection method based on local binary mode and neural network is a feasible method.The average accuracy rate of this method is more than 80%,the average recall rate is more than 80%,and the average accuracy rate of classification is more than 85%.