基于改进B-CNN模型的羊绒与羊毛纤维识别  

Recognition of cashmere and wool fiber based on improved B-CNN model

作　　者：朱耀麟[1,2] 穆婉婉 王进美 李文雅 ZHU Yaolin;MU Wanwan;WANG Jinmei;LI Wenya(School of Electronics and Information, Xi’an Polytechnic University, Xi’an 710048, China;School of Electronics and Information, Northwestern Polytechnical University,Xi’an 710072, China;School of Textile Science and Engineering, Xi’an Polytechnic University, Xi’an 710048, China)

机构地区：[1]西安工程大学电子信息学院,陕西西安710048 [2]西北工业大学电子信息学院,陕西西安710072 [3]西安工程大学纺织科学与工程学院,陕西西安710048

出　　处：《西安工程大学学报》2021年第6期46-53,共8页Journal of Xi’an Polytechnic University

基　　金：陕西省教育厅2020重点研究计划产业用纺织品协同创新中心项目(20JY026);榆林市科技局科创新城项目(2018-2-24);榆林科技局科技计划(CXY-2020-052)。

摘　　要：由于物体本身较小的类间差异和因拍摄环境、背景等导致的较大的类内差异,羊绒和羊毛的图像识别一直是纺织领域的难题。为解决羊绒与羊毛纤维难以鉴别的问题,提出一种改进的双线性卷积神经网络(bilinear convolutional neural network,B-CNN)模型用于羊绒和羊毛纤维识别。该方法通过对两路网络进行改进,提取纤维原始样本图像和骨架图像不同层次特征向量,采用向量拼接方式融合2幅图像特征,实现信息互补,从而增强特征表达能力,最后使用迁移训练,解决纤维扫描电子显微镜(scanning electron microscope,SEM)图像小样本问题,提高分类精度和效率。实验结果表明:该模型与经典B-CNN模型相比,测试集准确率最高可达98.06%,说明该模型能够有效解决羊绒与羊毛纤维识别问题。Due to the small inter-class differences of the object itself and the larger intra-class differences caused by the shooting environment and background,the image recognition of cashmere and wool has always been a problem in the textile field.In order to solve the problem,an improved bilinear convolutional neural network model for cashmere and wool fiber recognition was proposed.The two-way network of the B-CNN model was improved to extract feature vectors of different levels of fiber original sample images and skeleton images,and the features of the two images were fused using vector stitching in this method,so as to complement information and enhance the ability of feature expression.Finally,transfer training was used to solve the problem of small samples of fiber images and improve classification accuracy and efficiency.The experimental results show that the test set accuracy of this model can be up to 98.06%as opposed to that of the classic B-CNN model.It shows that the model can effectively solve the problem of cashmere and wool fiber recognition.

关 键 词：羊绒 羊毛 双线性卷积神经网络模型 特征融合 迁移训练 

分 类 号：TS102.3[轻工技术与工程—纺织工程]