基于小波能量基特征和BP神经网络的洗涤去污等级识别  

The Classification of Grade of Stain Release based on Wavelet Energy-based Signature and BP Neural Network

作　　者：巫志远 朱晓伟 王嘉[1] 刘建立[2] Wu Zhiyuan;Zhu Xiaowei;Wang Jia;LIU Jianli(Wuxi Little Swan Company Limited,Wuxi 214028,China;College of Textile and Clothing,Jiangnan University,Wuxi 214122,China)

机构地区：[1]无锡小天鹅股份有限公司,江苏无锡214028 [2]江南大学纺织服装学院,江苏无锡214122

出　　处：《纺织报告》2019年第6期8-12,25,共6页

摘　　要：为了研究小波能量基特征在织物洗涤去污等级识别中的应用,采用db6小波对多次洗涤后的沾污织物试样的图像进行小波分解,并提取高频子带小波系数的1-范数和2-范数作为能量基小波纹理特征。分别以从试样图像提取的小波纹理特征向量和其对应的洗涤去污等级为BP神经网络的输入和输出,对BP神经网络进行训练,并采用测试样本检验BP神经网络的正确识别率以衡量其泛化能力。实验结果表明:训练后的BP神经网络的正确识别率高达97%,其中,纹理特征类型和小波分解层数对网络的正确识别率的影响较大,而测试样本个数对正确识别率的影响不显著。In order to research the application of wavelet texture analysis on the recognition of grade of stain release,the db6 wavelet is used to decompose stain images of stained fabric samples after several times washing,and then the 1-norm and 2-norm values of wavelet coefficients are extracted as energy-based textural features.The generalization capability of the trained BP neural network is evaluated with the correct recognition ratio as criterion after which is trained by using the wavelet feature vector extracted from each sample and the corresponding of grade of stain release as input and output,respectively.Experimental result indicates that the highest correct recognition rate of the trained BP neural network is 97%.Additionally,the correct recognition rate is more influenced by the type of texture feature and the decomposition level of wavelet transform,while the effect of the number of test samples is insignificant.

关 键 词：织物 小波变换 能量基特征 BP神经网络 去污等级 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]